{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Getting started with astir"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Table of Contents:\n",
    "* [0 Loading necessary libraries](#zero)\n",
    "* [1 Load data](#one)\n",
    "* [2 Fitting cell type](#two)\n",
    "* [3 Fitting cell state](#three)\n",
    "* [4 Saving results](#four)\n",
    "* [5 Accessing internal functions and data](#five)\n",
    "* [6 Saving models](#six)\n",
    "* [7 Plot clustermap of expression data](#seven)\n",
    "* [8 Hierarchical model specification](#eight)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 0. Load necessary libraries <a class=\"anchor\" id=\"zero\"></a>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "outputs": [],
   "source": [
    "# !pip install -e ../../..\n",
    "import os\n",
    "import sys\n",
    "module_path = os.path.abspath(os.path.join('../../..'))\n",
    "if module_path not in sys.path:\n",
    "    sys.path.append(module_path)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "The autoreload extension is already loaded. To reload it, use:\n",
      "  %reload_ext autoreload\n"
     ]
    }
   ],
   "source": [
    "from astir.data import from_csv_yaml\n",
    "\n",
    "import pandas as pd\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "import seaborn as sns\n",
    "\n",
    "%load_ext autoreload\n",
    "%autoreload 2\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 1. Load data <a class=\"anchor\" id=\"one\"></a>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We start by reading expression data in the form of a csv file and marker gene information in the form of a yaml file:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "outputs": [],
   "source": [
    "expression_mat_path = \"../../../astir/tests/test-data/sce.csv\"\n",
    "# expression_mat_path = \"data/sample_data.csv\"\n",
    "yaml_marker_path = \"../../../astir/tests/test-data/jackson-2020-markers.yml\""
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    ".. note:: \n",
    "    Expression data should already be cleaned and normalized. In our workflow, we perform this by a log-transformation of the data with a pseudocount of 1 (i.e. `log(x+1)`), followed by winsorization at the (1%,99%) percentiles."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We can view both the expression data and marker data:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "cell_states:\r\n",
      "  RTK_signalling:\r\n",
      "    - Her2\r\n",
      "    - EGFR\r\n",
      "  cell_growth:\r\n",
      "    - Ki-67\r\n",
      "    - phospho mTOR\r\n",
      "  mTOR_signalling:\r\n",
      "    - phospho mTOR\r\n",
      "    - phospho S6\r\n",
      "  apoptosis:\r\n",
      "    - cleaved PARP\r\n",
      "    - Cleaved Caspase3\r\n",
      "\r\n",
      "cell_types:\r\n",
      "  stromal:\r\n",
      "    - Vimentin\r\n",
      "    - Fibronectin\r\n",
      "  B cells:\r\n",
      "    - CD45\r\n"
     ]
    }
   ],
   "source": [
    "!head -n 20 ../../../astir/tests/test-data/jackson-2020-markers.yml"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>EGFR</th>\n",
       "      <th>E-Cadherin</th>\n",
       "      <th>CD45</th>\n",
       "      <th>Cytokeratin 5</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>BaselTMA_SP41_186_X5Y4_3679</th>\n",
       "      <td>0.346787</td>\n",
       "      <td>0.938354</td>\n",
       "      <td>0.227730</td>\n",
       "      <td>0.095283</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BaselTMA_SP41_153_X7Y5_246</th>\n",
       "      <td>0.833752</td>\n",
       "      <td>1.364884</td>\n",
       "      <td>0.068526</td>\n",
       "      <td>0.124031</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BaselTMA_SP41_20_X12Y5_197</th>\n",
       "      <td>0.110006</td>\n",
       "      <td>0.177361</td>\n",
       "      <td>0.301222</td>\n",
       "      <td>0.052750</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BaselTMA_SP41_14_X1Y8_84</th>\n",
       "      <td>0.282666</td>\n",
       "      <td>1.122174</td>\n",
       "      <td>0.606941</td>\n",
       "      <td>0.093352</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BaselTMA_SP41_166_X15Y4_266</th>\n",
       "      <td>0.209066</td>\n",
       "      <td>0.402554</td>\n",
       "      <td>0.588273</td>\n",
       "      <td>0.064545</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                 EGFR  E-Cadherin      CD45  Cytokeratin 5\n",
       "BaselTMA_SP41_186_X5Y4_3679  0.346787    0.938354  0.227730       0.095283\n",
       "BaselTMA_SP41_153_X7Y5_246   0.833752    1.364884  0.068526       0.124031\n",
       "BaselTMA_SP41_20_X12Y5_197   0.110006    0.177361  0.301222       0.052750\n",
       "BaselTMA_SP41_14_X1Y8_84     0.282666    1.122174  0.606941       0.093352\n",
       "BaselTMA_SP41_166_X15Y4_266  0.209066    0.402554  0.588273       0.064545"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "pd.read_csv(expression_mat_path, index_col=0)[['EGFR','E-Cadherin', 'CD45', 'Cytokeratin 5']].head()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Then we can create an astir object using the `from_csv_yaml` function. For more data loading options, see the data loading tutorial."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Astir object with 6 cell types, 4 cell states, and 4931 cells.\n"
     ]
    }
   ],
   "source": [
    "ast = from_csv_yaml(expression_mat_path, marker_yaml=yaml_marker_path)\n",
    "print(ast)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 2. Fitting cell types <a class=\"anchor\" id=\"two\"></a>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "To fit cell types, simply call"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "training restart 1/5: 100%|██████████| 5/5 [ 3.76s/epochs, current loss: -1693.0]\n",
      "training restart 2/5: 100%|██████████| 5/5 [ 4.03s/epochs, current loss: 1273.4] \n",
      "training restart 3/5: 100%|██████████| 5/5 [ 3.54s/epochs, current loss: -640.5] \n",
      "training restart 4/5: 100%|██████████| 5/5 [ 3.41s/epochs, current loss: 198.5]  \n",
      "training restart 5/5: 100%|██████████| 5/5 [ 3.44s/epochs, current loss: -415.8] \n",
      "training restart (final):  94%|█████████▍| 47/50 [ 3.51s/epochs, current loss: -17252.6]\n"
     ]
    }
   ],
   "source": [
    "ast.fit_type(max_epochs=10, n_init=3, n_init_epochs=2)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    ".. note:: \n",
    "    **Controlling inference**\n",
    "    There are many different options for controlling inference in the `fit_type` function, including\n",
    "    `max_epochs` (maximum number of epochs to train),\n",
    "    `learning_rate` (ADAM optimizer learning rate),\n",
    "    `batch_size` (minibatch size),\n",
    "    `delta_loss` (stops iteration once the change in loss falls below this value),\n",
    "    `n_inits` (number of restarts using random initializations).\n",
    "    For full details, see the function documentation."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We should always plot the losses to assess convergence:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Text(0.5, 0, 'Epoch')"
      ]
     },
     "execution_count": 25,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 360x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(5,4))\n",
    "plt.plot(np.arange(len(ast.get_type_losses())), ast.get_type_losses())\n",
    "plt.ylabel(\"Loss\")\n",
    "plt.xlabel(\"Epoch\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We can then get cell type assignment probabilities by calling"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>stromal</th>\n",
       "      <th>B cells</th>\n",
       "      <th>T cells</th>\n",
       "      <th>macrophage</th>\n",
       "      <th>epithelial(basal)</th>\n",
       "      <th>epithelial(luminal)</th>\n",
       "      <th>Other</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>BaselTMA_SP43_115_X4Y8_1</th>\n",
       "      <td>5.981264e-05</td>\n",
       "      <td>4.351427e-01</td>\n",
       "      <td>4.413378e-06</td>\n",
       "      <td>4.778704e-05</td>\n",
       "      <td>0.005910</td>\n",
       "      <td>0.000002</td>\n",
       "      <td>5.588340e-01</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BaselTMA_SP43_115_X4Y8_2</th>\n",
       "      <td>7.080264e-05</td>\n",
       "      <td>3.849646e-01</td>\n",
       "      <td>5.889273e-06</td>\n",
       "      <td>4.314298e-05</td>\n",
       "      <td>0.044710</td>\n",
       "      <td>0.000007</td>\n",
       "      <td>5.701980e-01</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BaselTMA_SP43_115_X4Y8_3</th>\n",
       "      <td>2.086748e-04</td>\n",
       "      <td>1.871444e-01</td>\n",
       "      <td>9.592506e-04</td>\n",
       "      <td>4.061270e-04</td>\n",
       "      <td>0.290781</td>\n",
       "      <td>0.039960</td>\n",
       "      <td>4.805403e-01</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BaselTMA_SP43_115_X4Y8_4</th>\n",
       "      <td>2.980559e-04</td>\n",
       "      <td>4.438626e-01</td>\n",
       "      <td>6.048962e-06</td>\n",
       "      <td>1.707915e-04</td>\n",
       "      <td>0.011950</td>\n",
       "      <td>0.000006</td>\n",
       "      <td>5.437063e-01</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BaselTMA_SP43_115_X4Y8_5</th>\n",
       "      <td>1.121071e-04</td>\n",
       "      <td>4.681895e-01</td>\n",
       "      <td>2.136392e-05</td>\n",
       "      <td>3.064268e-05</td>\n",
       "      <td>0.017083</td>\n",
       "      <td>0.000002</td>\n",
       "      <td>5.145617e-01</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BaselTMA_SP43_115_X4Y8_4927</th>\n",
       "      <td>4.338320e-07</td>\n",
       "      <td>1.912927e-03</td>\n",
       "      <td>2.400906e-07</td>\n",
       "      <td>1.376804e-06</td>\n",
       "      <td>0.993716</td>\n",
       "      <td>0.000003</td>\n",
       "      <td>4.365914e-03</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BaselTMA_SP43_115_X4Y8_4928</th>\n",
       "      <td>2.193680e-04</td>\n",
       "      <td>3.908094e-01</td>\n",
       "      <td>3.504072e-04</td>\n",
       "      <td>2.164537e-04</td>\n",
       "      <td>0.009062</td>\n",
       "      <td>0.000018</td>\n",
       "      <td>5.993251e-01</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BaselTMA_SP43_115_X4Y8_4929</th>\n",
       "      <td>1.189970e-08</td>\n",
       "      <td>7.369497e-07</td>\n",
       "      <td>5.310283e-09</td>\n",
       "      <td>8.920690e-09</td>\n",
       "      <td>0.999534</td>\n",
       "      <td>0.000464</td>\n",
       "      <td>8.322494e-07</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BaselTMA_SP43_115_X4Y8_4930</th>\n",
       "      <td>5.812946e-08</td>\n",
       "      <td>6.018526e-09</td>\n",
       "      <td>2.364501e-08</td>\n",
       "      <td>5.991192e-09</td>\n",
       "      <td>0.000590</td>\n",
       "      <td>0.999410</td>\n",
       "      <td>6.340690e-08</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BaselTMA_SP43_115_X4Y8_4931</th>\n",
       "      <td>1.341373e-05</td>\n",
       "      <td>1.754594e-06</td>\n",
       "      <td>2.443332e-06</td>\n",
       "      <td>8.300306e-07</td>\n",
       "      <td>0.060859</td>\n",
       "      <td>0.939114</td>\n",
       "      <td>8.438225e-06</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>4931 rows × 7 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "                                  stromal       B cells       T cells  \\\n",
       "BaselTMA_SP43_115_X4Y8_1     5.981264e-05  4.351427e-01  4.413378e-06   \n",
       "BaselTMA_SP43_115_X4Y8_2     7.080264e-05  3.849646e-01  5.889273e-06   \n",
       "BaselTMA_SP43_115_X4Y8_3     2.086748e-04  1.871444e-01  9.592506e-04   \n",
       "BaselTMA_SP43_115_X4Y8_4     2.980559e-04  4.438626e-01  6.048962e-06   \n",
       "BaselTMA_SP43_115_X4Y8_5     1.121071e-04  4.681895e-01  2.136392e-05   \n",
       "...                                   ...           ...           ...   \n",
       "BaselTMA_SP43_115_X4Y8_4927  4.338320e-07  1.912927e-03  2.400906e-07   \n",
       "BaselTMA_SP43_115_X4Y8_4928  2.193680e-04  3.908094e-01  3.504072e-04   \n",
       "BaselTMA_SP43_115_X4Y8_4929  1.189970e-08  7.369497e-07  5.310283e-09   \n",
       "BaselTMA_SP43_115_X4Y8_4930  5.812946e-08  6.018526e-09  2.364501e-08   \n",
       "BaselTMA_SP43_115_X4Y8_4931  1.341373e-05  1.754594e-06  2.443332e-06   \n",
       "\n",
       "                               macrophage  epithelial(basal)  \\\n",
       "BaselTMA_SP43_115_X4Y8_1     4.778704e-05           0.005910   \n",
       "BaselTMA_SP43_115_X4Y8_2     4.314298e-05           0.044710   \n",
       "BaselTMA_SP43_115_X4Y8_3     4.061270e-04           0.290781   \n",
       "BaselTMA_SP43_115_X4Y8_4     1.707915e-04           0.011950   \n",
       "BaselTMA_SP43_115_X4Y8_5     3.064268e-05           0.017083   \n",
       "...                                   ...                ...   \n",
       "BaselTMA_SP43_115_X4Y8_4927  1.376804e-06           0.993716   \n",
       "BaselTMA_SP43_115_X4Y8_4928  2.164537e-04           0.009062   \n",
       "BaselTMA_SP43_115_X4Y8_4929  8.920690e-09           0.999534   \n",
       "BaselTMA_SP43_115_X4Y8_4930  5.991192e-09           0.000590   \n",
       "BaselTMA_SP43_115_X4Y8_4931  8.300306e-07           0.060859   \n",
       "\n",
       "                             epithelial(luminal)         Other  \n",
       "BaselTMA_SP43_115_X4Y8_1                0.000002  5.588340e-01  \n",
       "BaselTMA_SP43_115_X4Y8_2                0.000007  5.701980e-01  \n",
       "BaselTMA_SP43_115_X4Y8_3                0.039960  4.805403e-01  \n",
       "BaselTMA_SP43_115_X4Y8_4                0.000006  5.437063e-01  \n",
       "BaselTMA_SP43_115_X4Y8_5                0.000002  5.145617e-01  \n",
       "...                                          ...           ...  \n",
       "BaselTMA_SP43_115_X4Y8_4927             0.000003  4.365914e-03  \n",
       "BaselTMA_SP43_115_X4Y8_4928             0.000018  5.993251e-01  \n",
       "BaselTMA_SP43_115_X4Y8_4929             0.000464  8.322494e-07  \n",
       "BaselTMA_SP43_115_X4Y8_4930             0.999410  6.340690e-08  \n",
       "BaselTMA_SP43_115_X4Y8_4931             0.939114  8.438225e-06  \n",
       "\n",
       "[4931 rows x 7 columns]"
      ]
     },
     "execution_count": 26,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "assignments = ast.get_celltype_probabilities()\n",
    "assignments"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We can also visualize the assignment probabilities using a heatmap:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.axes._subplots.AxesSubplot at 0x7f8d611c72d0>"
      ]
     },
     "execution_count": 27,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "sns.heatmap(assignments)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "where each row corresponds to a cell, and each column to a cell type, with the entry being the probability of that cell belonging to a particular cell type.\n",
    "\n",
    "To fetch an array corresponding to the most likely cell type assignments, call"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>cell_type</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>BaselTMA_SP43_115_X4Y8_1</th>\n",
       "      <td>Unknown</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BaselTMA_SP43_115_X4Y8_2</th>\n",
       "      <td>Unknown</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BaselTMA_SP43_115_X4Y8_3</th>\n",
       "      <td>Unknown</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BaselTMA_SP43_115_X4Y8_4</th>\n",
       "      <td>Unknown</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BaselTMA_SP43_115_X4Y8_5</th>\n",
       "      <td>Unknown</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BaselTMA_SP43_115_X4Y8_4927</th>\n",
       "      <td>epithelial(basal)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BaselTMA_SP43_115_X4Y8_4928</th>\n",
       "      <td>Unknown</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BaselTMA_SP43_115_X4Y8_4929</th>\n",
       "      <td>epithelial(basal)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BaselTMA_SP43_115_X4Y8_4930</th>\n",
       "      <td>epithelial(luminal)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BaselTMA_SP43_115_X4Y8_4931</th>\n",
       "      <td>epithelial(luminal)</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>4931 rows × 1 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "                                       cell_type\n",
       "BaselTMA_SP43_115_X4Y8_1                 Unknown\n",
       "BaselTMA_SP43_115_X4Y8_2                 Unknown\n",
       "BaselTMA_SP43_115_X4Y8_3                 Unknown\n",
       "BaselTMA_SP43_115_X4Y8_4                 Unknown\n",
       "BaselTMA_SP43_115_X4Y8_5                 Unknown\n",
       "...                                          ...\n",
       "BaselTMA_SP43_115_X4Y8_4927    epithelial(basal)\n",
       "BaselTMA_SP43_115_X4Y8_4928              Unknown\n",
       "BaselTMA_SP43_115_X4Y8_4929    epithelial(basal)\n",
       "BaselTMA_SP43_115_X4Y8_4930  epithelial(luminal)\n",
       "BaselTMA_SP43_115_X4Y8_4931  epithelial(luminal)\n",
       "\n",
       "[4931 rows x 1 columns]"
      ]
     },
     "execution_count": 28,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "ast.get_celltypes()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Cell type diagnostics"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "It is important to run diagnostics to ensure that cell types express their markers at higher levels than other cell types. To do this, run the `diagnostics_celltype()` function, which will alert to any issues if a cell type doesn't express its marker signficantly higher than an alternative cell type (for which that protein isn't a marker):"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>feature</th>\n",
       "      <th>should be expressed higher in</th>\n",
       "      <th>than</th>\n",
       "      <th>mean cell type 1</th>\n",
       "      <th>mean cell type 2</th>\n",
       "      <th>p-value</th>\n",
       "      <th>note</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>Fibronectin</td>\n",
       "      <td>stromal</td>\n",
       "      <td>B cells</td>\n",
       "      <td>2.045195</td>\n",
       "      <td>1.565420</td>\n",
       "      <td>inf</td>\n",
       "      <td>Only 1 cell in a type: comparison not possible</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>Vimentin</td>\n",
       "      <td>stromal</td>\n",
       "      <td>B cells</td>\n",
       "      <td>2.994335</td>\n",
       "      <td>1.095477</td>\n",
       "      <td>inf</td>\n",
       "      <td>Only 1 cell in a type: comparison not possible</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>CD20</td>\n",
       "      <td>B cells</td>\n",
       "      <td>stromal</td>\n",
       "      <td>0.403677</td>\n",
       "      <td>0.081944</td>\n",
       "      <td>inf</td>\n",
       "      <td>Only 1 cell in a type: comparison not possible</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>CD20</td>\n",
       "      <td>B cells</td>\n",
       "      <td>T cells</td>\n",
       "      <td>0.403677</td>\n",
       "      <td>0.123195</td>\n",
       "      <td>inf</td>\n",
       "      <td>Only 1 cell in a type: comparison not possible</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>CD20</td>\n",
       "      <td>B cells</td>\n",
       "      <td>macrophage</td>\n",
       "      <td>0.403677</td>\n",
       "      <td>0.212008</td>\n",
       "      <td>inf</td>\n",
       "      <td>Only 1 cell in a type: comparison not possible</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>CD20</td>\n",
       "      <td>B cells</td>\n",
       "      <td>epithelial(basal)</td>\n",
       "      <td>0.403677</td>\n",
       "      <td>0.118228</td>\n",
       "      <td>inf</td>\n",
       "      <td>Only 1 cell in a type: comparison not possible</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>CD20</td>\n",
       "      <td>B cells</td>\n",
       "      <td>epithelial(luminal)</td>\n",
       "      <td>0.403677</td>\n",
       "      <td>0.131991</td>\n",
       "      <td>inf</td>\n",
       "      <td>Only 1 cell in a type: comparison not possible</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>CD20</td>\n",
       "      <td>B cells</td>\n",
       "      <td>Other</td>\n",
       "      <td>0.403677</td>\n",
       "      <td>0.017356</td>\n",
       "      <td>inf</td>\n",
       "      <td>Only 1 cell in a type: comparison not possible</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>CD45</td>\n",
       "      <td>B cells</td>\n",
       "      <td>stromal</td>\n",
       "      <td>0.077596</td>\n",
       "      <td>0.227257</td>\n",
       "      <td>inf</td>\n",
       "      <td>Only 1 cell in a type: comparison not possible</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>CD45</td>\n",
       "      <td>B cells</td>\n",
       "      <td>epithelial(basal)</td>\n",
       "      <td>0.077596</td>\n",
       "      <td>0.182078</td>\n",
       "      <td>inf</td>\n",
       "      <td>Only 1 cell in a type: comparison not possible</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "       feature should be expressed higher in                 than  \\\n",
       "0  Fibronectin                       stromal              B cells   \n",
       "1     Vimentin                       stromal              B cells   \n",
       "2         CD20                       B cells              stromal   \n",
       "3         CD20                       B cells              T cells   \n",
       "4         CD20                       B cells           macrophage   \n",
       "5         CD20                       B cells    epithelial(basal)   \n",
       "6         CD20                       B cells  epithelial(luminal)   \n",
       "7         CD20                       B cells                Other   \n",
       "8         CD45                       B cells              stromal   \n",
       "9         CD45                       B cells    epithelial(basal)   \n",
       "\n",
       "   mean cell type 1  mean cell type 2  p-value  \\\n",
       "0          2.045195          1.565420      inf   \n",
       "1          2.994335          1.095477      inf   \n",
       "2          0.403677          0.081944      inf   \n",
       "3          0.403677          0.123195      inf   \n",
       "4          0.403677          0.212008      inf   \n",
       "5          0.403677          0.118228      inf   \n",
       "6          0.403677          0.131991      inf   \n",
       "7          0.403677          0.017356      inf   \n",
       "8          0.077596          0.227257      inf   \n",
       "9          0.077596          0.182078      inf   \n",
       "\n",
       "                                             note  \n",
       "0  Only 1 cell in a type: comparison not possible  \n",
       "1  Only 1 cell in a type: comparison not possible  \n",
       "2  Only 1 cell in a type: comparison not possible  \n",
       "3  Only 1 cell in a type: comparison not possible  \n",
       "4  Only 1 cell in a type: comparison not possible  \n",
       "5  Only 1 cell in a type: comparison not possible  \n",
       "6  Only 1 cell in a type: comparison not possible  \n",
       "7  Only 1 cell in a type: comparison not possible  \n",
       "8  Only 1 cell in a type: comparison not possible  \n",
       "9  Only 1 cell in a type: comparison not possible  "
      ]
     },
     "execution_count": 29,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "ast.diagnostics_celltype().head(n=10)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    ".. note:: \n",
    "    In this tutorial, we end up with many \"Only 1 cell in a type: comparison not possible\" notes - this is simply because the small dataset size results in only a single cell assigned to many types, making statistical testing infeasible."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Calling `ast.diagnostics_celltype()` returns a `pd.DataFrame`, where each column corresponds to a particular protein and two cell types, with a warning if the protein is not expressed at higher levels in the cell type for which it is a marker than the cell type for which it is not."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The diagnostics:\n",
    "\n",
    "1. Iterates through every cell type and every marker for that cell type\n",
    "\n",
    "2. Given a cell type *c* and marker *g*, find the set of cell types *D* that don't have *g* as a marker\n",
    "\n",
    "3. For each cell type *d* in *D*, perform a t-test between the expression of marker *g* in *c* vs *d*\n",
    "\n",
    "4. If *g* is not expressed significantly higher (at significance *alpha*), output a diagnostic explaining this for further investigation."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "If multiple issues are found, the markers and cell types may need refined."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 3. Fitting cell state <a class=\"anchor\" id=\"three\"></a>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Similarly as before, to fit cell state, call"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/Users/jinelles.h/Documents/Camlab/astir-top-level/astir/astir/models/cellstate.py:176: UserWarning: Delta loss batch size is greater than the number of epochs\n",
      "  warnings.warn(\"Delta loss batch size is greater than the number of epochs\")\n",
      "training restart 1/5: 100%|██████████| 5/5 [ 3.41s/epochs, current loss: 59.8]\n",
      "training restart 2/5: 100%|██████████| 5/5 [ 3.40s/epochs, current loss: 98.0] \n",
      "training restart 3/5: 100%|██████████| 5/5 [ 3.51s/epochs, current loss: 78.2]\n",
      "training restart 4/5: 100%|██████████| 5/5 [ 3.48s/epochs, current loss: 88.0] \n",
      "training restart 5/5: 100%|██████████| 5/5 [ 3.37s/epochs, current loss: 60.0]\n",
      "training restart (final): 100%|██████████| 50/50 [ 3.60s/epochs, current loss: 14.5]\n",
      "/Users/jinelles.h/Documents/Camlab/astir-top-level/astir/astir/astir.py:268: UserWarning: Maximum epochs reached. More iteration may be needed to complete the training.\n",
      "  warnings.warn(msg)\n"
     ]
    }
   ],
   "source": [
    "ast.fit_state(batch_size = 1024, learning_rate=1e-3, max_epochs=10)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "and similary plot the losses via"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Text(0.5, 0, 'Epoch')"
      ]
     },
     "execution_count": 31,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 360x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(5,4))\n",
    "plt.plot(np.arange(len(ast.get_state_losses())), ast.get_state_losses())\n",
    "plt.ylabel(\"Loss\")\n",
    "plt.xlabel(\"Epoch\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "and cell state assignments can be inferred via"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>RTK_signalling</th>\n",
       "      <th>cell_growth</th>\n",
       "      <th>mTOR_signalling</th>\n",
       "      <th>apoptosis</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>BaselTMA_SP43_115_X4Y8_1</th>\n",
       "      <td>0.219178</td>\n",
       "      <td>0.178849</td>\n",
       "      <td>0.133436</td>\n",
       "      <td>0.144105</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BaselTMA_SP43_115_X4Y8_2</th>\n",
       "      <td>0.296468</td>\n",
       "      <td>0.119678</td>\n",
       "      <td>0.200147</td>\n",
       "      <td>0.157937</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BaselTMA_SP43_115_X4Y8_3</th>\n",
       "      <td>0.181820</td>\n",
       "      <td>0.185739</td>\n",
       "      <td>0.117271</td>\n",
       "      <td>0.117391</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BaselTMA_SP43_115_X4Y8_4</th>\n",
       "      <td>0.271369</td>\n",
       "      <td>0.091272</td>\n",
       "      <td>0.145709</td>\n",
       "      <td>0.170535</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BaselTMA_SP43_115_X4Y8_5</th>\n",
       "      <td>0.238632</td>\n",
       "      <td>0.219416</td>\n",
       "      <td>0.147383</td>\n",
       "      <td>0.165110</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BaselTMA_SP43_115_X4Y8_4927</th>\n",
       "      <td>0.377885</td>\n",
       "      <td>0.229766</td>\n",
       "      <td>0.257093</td>\n",
       "      <td>0.232393</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BaselTMA_SP43_115_X4Y8_4928</th>\n",
       "      <td>0.144620</td>\n",
       "      <td>0.335371</td>\n",
       "      <td>0.105722</td>\n",
       "      <td>0.134628</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BaselTMA_SP43_115_X4Y8_4929</th>\n",
       "      <td>0.494012</td>\n",
       "      <td>0.387052</td>\n",
       "      <td>0.286395</td>\n",
       "      <td>0.357899</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BaselTMA_SP43_115_X4Y8_4930</th>\n",
       "      <td>0.624909</td>\n",
       "      <td>0.492357</td>\n",
       "      <td>0.315533</td>\n",
       "      <td>0.519452</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BaselTMA_SP43_115_X4Y8_4931</th>\n",
       "      <td>0.631358</td>\n",
       "      <td>0.453455</td>\n",
       "      <td>0.292822</td>\n",
       "      <td>0.536574</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>4931 rows × 4 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "                             RTK_signalling  cell_growth  mTOR_signalling  \\\n",
       "BaselTMA_SP43_115_X4Y8_1           0.219178     0.178849         0.133436   \n",
       "BaselTMA_SP43_115_X4Y8_2           0.296468     0.119678         0.200147   \n",
       "BaselTMA_SP43_115_X4Y8_3           0.181820     0.185739         0.117271   \n",
       "BaselTMA_SP43_115_X4Y8_4           0.271369     0.091272         0.145709   \n",
       "BaselTMA_SP43_115_X4Y8_5           0.238632     0.219416         0.147383   \n",
       "...                                     ...          ...              ...   \n",
       "BaselTMA_SP43_115_X4Y8_4927        0.377885     0.229766         0.257093   \n",
       "BaselTMA_SP43_115_X4Y8_4928        0.144620     0.335371         0.105722   \n",
       "BaselTMA_SP43_115_X4Y8_4929        0.494012     0.387052         0.286395   \n",
       "BaselTMA_SP43_115_X4Y8_4930        0.624909     0.492357         0.315533   \n",
       "BaselTMA_SP43_115_X4Y8_4931        0.631358     0.453455         0.292822   \n",
       "\n",
       "                             apoptosis  \n",
       "BaselTMA_SP43_115_X4Y8_1      0.144105  \n",
       "BaselTMA_SP43_115_X4Y8_2      0.157937  \n",
       "BaselTMA_SP43_115_X4Y8_3      0.117391  \n",
       "BaselTMA_SP43_115_X4Y8_4      0.170535  \n",
       "BaselTMA_SP43_115_X4Y8_5      0.165110  \n",
       "...                                ...  \n",
       "BaselTMA_SP43_115_X4Y8_4927   0.232393  \n",
       "BaselTMA_SP43_115_X4Y8_4928   0.134628  \n",
       "BaselTMA_SP43_115_X4Y8_4929   0.357899  \n",
       "BaselTMA_SP43_115_X4Y8_4930   0.519452  \n",
       "BaselTMA_SP43_115_X4Y8_4931   0.536574  \n",
       "\n",
       "[4931 rows x 4 columns]"
      ]
     },
     "execution_count": 32,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "states = ast.get_cellstates()\n",
    "states"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.collections.PathCollection at 0x7f8d60937b90>"
      ]
     },
     "execution_count": 33,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.scatter(\n",
    "    states['RTK_signalling'],\n",
    "    ast.get_state_dataset().get_exprs_df()['Her2']\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Cell state diagnostics"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "It is important to run diagnostics on cell states model for the same reasons\n",
    "stated for the cell type model. `Astir.diagnostics_cellstate()` spots any non\n",
    " marker protein and pathway pairs whose expressions are higher than those of\n",
    " the marker proteins of the pathway."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>pathway</th>\n",
       "      <th>protein A</th>\n",
       "      <th>correlation of protein A</th>\n",
       "      <th>protein B</th>\n",
       "      <th>correlation of protein B</th>\n",
       "      <th>note</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>RTK_signalling</td>\n",
       "      <td>EGFR</td>\n",
       "      <td>0.823298</td>\n",
       "      <td>Cleaved Caspase3</td>\n",
       "      <td>0.831128</td>\n",
       "      <td>EGFR is marker for RTK_signalling but Cleaved ...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>RTK_signalling</td>\n",
       "      <td>EGFR</td>\n",
       "      <td>0.823298</td>\n",
       "      <td>cleaved PARP</td>\n",
       "      <td>0.831128</td>\n",
       "      <td>EGFR is marker for RTK_signalling but cleaved ...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>cell_growth</td>\n",
       "      <td>Ki-67</td>\n",
       "      <td>0.058879</td>\n",
       "      <td>Cleaved Caspase3</td>\n",
       "      <td>0.862459</td>\n",
       "      <td>Ki-67 is marker for cell_growth but Cleaved Ca...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>cell_growth</td>\n",
       "      <td>Ki-67</td>\n",
       "      <td>0.058879</td>\n",
       "      <td>EGFR</td>\n",
       "      <td>0.897650</td>\n",
       "      <td>Ki-67 is marker for cell_growth but EGFR isn't</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>cell_growth</td>\n",
       "      <td>Ki-67</td>\n",
       "      <td>0.058879</td>\n",
       "      <td>Her2</td>\n",
       "      <td>0.750985</td>\n",
       "      <td>Ki-67 is marker for cell_growth but Her2 isn't</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>cell_growth</td>\n",
       "      <td>Ki-67</td>\n",
       "      <td>0.058879</td>\n",
       "      <td>cleaved PARP</td>\n",
       "      <td>0.862459</td>\n",
       "      <td>Ki-67 is marker for cell_growth but cleaved PA...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>cell_growth</td>\n",
       "      <td>Ki-67</td>\n",
       "      <td>0.058879</td>\n",
       "      <td>phospho S6</td>\n",
       "      <td>0.272989</td>\n",
       "      <td>Ki-67 is marker for cell_growth but phospho S6...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>mTOR_signalling</td>\n",
       "      <td>phospho S6</td>\n",
       "      <td>0.699387</td>\n",
       "      <td>EGFR</td>\n",
       "      <td>0.720013</td>\n",
       "      <td>phospho S6 is marker for mTOR_signalling but E...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>mTOR_signalling</td>\n",
       "      <td>phospho S6</td>\n",
       "      <td>0.699387</td>\n",
       "      <td>Her2</td>\n",
       "      <td>0.757867</td>\n",
       "      <td>phospho S6 is marker for mTOR_signalling but H...</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "           pathway   protein A  correlation of protein A         protein B  \\\n",
       "0   RTK_signalling        EGFR                  0.823298  Cleaved Caspase3   \n",
       "1   RTK_signalling        EGFR                  0.823298      cleaved PARP   \n",
       "2      cell_growth       Ki-67                  0.058879  Cleaved Caspase3   \n",
       "3      cell_growth       Ki-67                  0.058879              EGFR   \n",
       "4      cell_growth       Ki-67                  0.058879              Her2   \n",
       "5      cell_growth       Ki-67                  0.058879      cleaved PARP   \n",
       "6      cell_growth       Ki-67                  0.058879        phospho S6   \n",
       "7  mTOR_signalling  phospho S6                  0.699387              EGFR   \n",
       "8  mTOR_signalling  phospho S6                  0.699387              Her2   \n",
       "\n",
       "   correlation of protein B                                               note  \n",
       "0                  0.831128  EGFR is marker for RTK_signalling but Cleaved ...  \n",
       "1                  0.831128  EGFR is marker for RTK_signalling but cleaved ...  \n",
       "2                  0.862459  Ki-67 is marker for cell_growth but Cleaved Ca...  \n",
       "3                  0.897650     Ki-67 is marker for cell_growth but EGFR isn't  \n",
       "4                  0.750985     Ki-67 is marker for cell_growth but Her2 isn't  \n",
       "5                  0.862459  Ki-67 is marker for cell_growth but cleaved PA...  \n",
       "6                  0.272989  Ki-67 is marker for cell_growth but phospho S6...  \n",
       "7                  0.720013  phospho S6 is marker for mTOR_signalling but E...  \n",
       "8                  0.757867  phospho S6 is marker for mTOR_signalling but H...  "
      ]
     },
     "execution_count": 34,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "ast.diagnostics_cellstate().head(n=10)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Calling `ast.diagnostics_cellstate()` returns a `pd.DataFrame`, where each\n",
    "column corresponds to a particular protein and two cell types, with a warning\n",
    " if the protein is not expressed at higher levels in the cell state for which\n",
    " it is a marker than the cell state for which it is not."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The diagnostics:\n",
    "\n",
    "1. Get correlations between all cell states and proteins\n",
    "\n",
    "2. For each cell state *c*, get the smallest correlation with marker *g*\n",
    "\n",
    "3. For each cell state *c* and its non marker *g*, find any correlation that is\n",
    "bigger than those smallest correlation for *c*.\n",
    "\n",
    "4. Any *c* and *g* pairs found in step 3 will be included in the output of\n",
    "`Astir.diagnostics_cellstate()`, including an explanation."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "If multiple issues are found, the markers and cell states may need refined.\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 4. Saving results <a class=\"anchor\" id=\"four\"></a>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Both cell type and cell state information can easily be saved to disk via"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "outputs": [],
   "source": [
    "ast.type_to_csv(\"data/cell-types.csv\")\n",
    "ast.state_to_csv(\"data/cell-states.csv\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      ",stromal,B cells,T cells,macrophage,epithelial(basal),epithelial(luminal),Other\r\n",
      "BaselTMA_SP43_115_X4Y8_1,5.9812641356773485e-05,0.43514272442399615,4.4133780759206555e-06,4.778704083811438e-05,0.00590960852651017,1.6351770806453748e-06,0.5588340188121422\r\n",
      "BaselTMA_SP43_115_X4Y8_2,7.080264289768121e-05,0.38496458175145926,5.889272783675721e-06,4.3142984449199766e-05,0.04471036731616506,7.2442437475747445e-06,0.5701979717884975\r\n"
     ]
    }
   ],
   "source": [
    "!head -n 3 data/cell-types.csv"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      ",RTK_signalling,cell_growth,mTOR_signalling,apoptosis\r\n",
      "BaselTMA_SP43_115_X4Y8_1,0.2191776340692214,0.17884886731101723,0.13343558969907734,0.1441052515306596\r\n",
      "BaselTMA_SP43_115_X4Y8_2,0.29646752007327026,0.11967817733248733,0.20014708539822942,0.1579368118796618\r\n"
     ]
    }
   ],
   "source": [
    "!head -n 3 data/cell-states.csv"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "where the first (unnamed) column always corresponds to the cell name/ID."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 5. Accessing internal functions and data <a class=\"anchor\" id=\"five\"></a>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Data stored in `astir` objects is in the form of an `SCDataSet`. These can be retrieved via"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<astir.data.scdataset.SCDataset at 0x7f8d60bd6ed0>"
      ]
     },
     "execution_count": 39,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "celltype_data = ast.get_type_dataset()\n",
    "celltype_data"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "and similarly for cell state via `ast.get_state_dataset()`.\n",
    "\n",
    "These have several helper functions to retrieve relevant information to the dataset:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['BaselTMA_SP43_115_X4Y8_1',\n",
       " 'BaselTMA_SP43_115_X4Y8_2',\n",
       " 'BaselTMA_SP43_115_X4Y8_3',\n",
       " 'BaselTMA_SP43_115_X4Y8_4']"
      ]
     },
     "execution_count": 40,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "celltype_data.get_cell_names()[0:4] # cell names"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['stromal',\n",
       " 'B cells',\n",
       " 'T cells',\n",
       " 'macrophage',\n",
       " 'epithelial(basal)',\n",
       " 'epithelial(luminal)']"
      ]
     },
     "execution_count": 41,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "celltype_data.get_classes() # cell type names"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "6\n",
      "14\n"
     ]
    }
   ],
   "source": [
    "print(celltype_data.get_n_classes()) # number of cell types\n",
    "print(celltype_data.get_n_features()) # number of features / proteins"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "tensor([[4.2121e-02, 5.2044e-02, 1.7348e-03,  ..., 6.1314e-01, 4.4827e-02,\n",
       "         3.8841e-01],\n",
       "        [0.0000e+00, 3.4770e-02, 0.0000e+00,  ..., 9.4025e-01, 1.9424e-02,\n",
       "         5.6716e-01],\n",
       "        [1.8200e-02, 7.8596e-02, 0.0000e+00,  ..., 6.2852e-01, 4.0905e-02,\n",
       "         8.4946e-01],\n",
       "        ...,\n",
       "        [1.4403e-01, 7.0877e-02, 2.0325e-01,  ..., 1.7303e+00, 2.1434e-01,\n",
       "         9.4889e-01],\n",
       "        [3.6400e-02, 9.9307e-02, 1.1815e-01,  ..., 1.6467e+00, 1.3463e-01,\n",
       "         1.9238e+00],\n",
       "        [5.4069e-02, 1.1861e-01, 5.3894e-02,  ..., 1.4265e+00, 4.9443e-01,\n",
       "         1.6126e+00]], dtype=torch.float64)"
      ]
     },
     "execution_count": 43,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "celltype_data.get_exprs() # Return a torch tensor corresponding to the expression data used"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>CD20</th>\n",
       "      <th>CD3</th>\n",
       "      <th>CD45</th>\n",
       "      <th>CD68</th>\n",
       "      <th>Cytokeratin 14</th>\n",
       "      <th>Cytokeratin 19</th>\n",
       "      <th>Cytokeratin 5</th>\n",
       "      <th>Cytokeratin 7</th>\n",
       "      <th>Cytokeratin 8/18</th>\n",
       "      <th>E-Cadherin</th>\n",
       "      <th>Fibronectin</th>\n",
       "      <th>Her2</th>\n",
       "      <th>Vimentin</th>\n",
       "      <th>pan Cytokeratin</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>BaselTMA_SP43_115_X4Y8_1</th>\n",
       "      <td>0.042121</td>\n",
       "      <td>0.052044</td>\n",
       "      <td>0.001735</td>\n",
       "      <td>0.042628</td>\n",
       "      <td>0.046076</td>\n",
       "      <td>0.076746</td>\n",
       "      <td>0.113790</td>\n",
       "      <td>0.073571</td>\n",
       "      <td>0.110530</td>\n",
       "      <td>0.800019</td>\n",
       "      <td>0.324533</td>\n",
       "      <td>0.613138</td>\n",
       "      <td>0.044827</td>\n",
       "      <td>0.388409</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BaselTMA_SP43_115_X4Y8_2</th>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.034770</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.084581</td>\n",
       "      <td>0.059064</td>\n",
       "      <td>0.056487</td>\n",
       "      <td>0.098873</td>\n",
       "      <td>0.019242</td>\n",
       "      <td>0.152825</td>\n",
       "      <td>0.972369</td>\n",
       "      <td>0.384034</td>\n",
       "      <td>0.940250</td>\n",
       "      <td>0.019424</td>\n",
       "      <td>0.567159</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BaselTMA_SP43_115_X4Y8_3</th>\n",
       "      <td>0.018200</td>\n",
       "      <td>0.078596</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.042628</td>\n",
       "      <td>0.026254</td>\n",
       "      <td>0.126815</td>\n",
       "      <td>0.165966</td>\n",
       "      <td>0.310404</td>\n",
       "      <td>0.253835</td>\n",
       "      <td>1.108996</td>\n",
       "      <td>0.256832</td>\n",
       "      <td>0.628524</td>\n",
       "      <td>0.040905</td>\n",
       "      <td>0.849462</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BaselTMA_SP43_115_X4Y8_4</th>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.018701</td>\n",
       "      <td>0.015682</td>\n",
       "      <td>0.108294</td>\n",
       "      <td>0.071382</td>\n",
       "      <td>0.052757</td>\n",
       "      <td>0.022484</td>\n",
       "      <td>0.093607</td>\n",
       "      <td>0.062275</td>\n",
       "      <td>1.204820</td>\n",
       "      <td>0.661950</td>\n",
       "      <td>0.530076</td>\n",
       "      <td>0.093848</td>\n",
       "      <td>0.418222</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BaselTMA_SP43_115_X4Y8_5</th>\n",
       "      <td>0.099789</td>\n",
       "      <td>0.073795</td>\n",
       "      <td>0.093230</td>\n",
       "      <td>0.122463</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.238924</td>\n",
       "      <td>0.193627</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.125397</td>\n",
       "      <td>1.225161</td>\n",
       "      <td>0.623664</td>\n",
       "      <td>0.493992</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.296923</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BaselTMA_SP43_115_X4Y8_4927</th>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.036476</td>\n",
       "      <td>0.142845</td>\n",
       "      <td>0.384770</td>\n",
       "      <td>0.046530</td>\n",
       "      <td>0.000362</td>\n",
       "      <td>0.139654</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>1.730274</td>\n",
       "      <td>0.233316</td>\n",
       "      <td>1.276169</td>\n",
       "      <td>0.171234</td>\n",
       "      <td>0.396967</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BaselTMA_SP43_115_X4Y8_4928</th>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.100622</td>\n",
       "      <td>0.089643</td>\n",
       "      <td>0.091315</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.098437</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.155443</td>\n",
       "      <td>0.908461</td>\n",
       "      <td>0.411734</td>\n",
       "      <td>0.365833</td>\n",
       "      <td>0.110548</td>\n",
       "      <td>0.789458</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BaselTMA_SP43_115_X4Y8_4929</th>\n",
       "      <td>0.144029</td>\n",
       "      <td>0.070877</td>\n",
       "      <td>0.203246</td>\n",
       "      <td>0.503439</td>\n",
       "      <td>0.051248</td>\n",
       "      <td>0.200929</td>\n",
       "      <td>0.138156</td>\n",
       "      <td>0.045556</td>\n",
       "      <td>0.057138</td>\n",
       "      <td>2.136403</td>\n",
       "      <td>0.841147</td>\n",
       "      <td>1.730263</td>\n",
       "      <td>0.214340</td>\n",
       "      <td>0.948885</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BaselTMA_SP43_115_X4Y8_4930</th>\n",
       "      <td>0.036400</td>\n",
       "      <td>0.099307</td>\n",
       "      <td>0.118155</td>\n",
       "      <td>0.447235</td>\n",
       "      <td>0.285294</td>\n",
       "      <td>0.270529</td>\n",
       "      <td>0.107387</td>\n",
       "      <td>0.423265</td>\n",
       "      <td>0.234452</td>\n",
       "      <td>1.884903</td>\n",
       "      <td>0.767309</td>\n",
       "      <td>1.646707</td>\n",
       "      <td>0.134634</td>\n",
       "      <td>1.923792</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BaselTMA_SP43_115_X4Y8_4931</th>\n",
       "      <td>0.054069</td>\n",
       "      <td>0.118611</td>\n",
       "      <td>0.053894</td>\n",
       "      <td>0.484753</td>\n",
       "      <td>0.340787</td>\n",
       "      <td>0.133814</td>\n",
       "      <td>0.150674</td>\n",
       "      <td>0.365944</td>\n",
       "      <td>0.163528</td>\n",
       "      <td>1.843372</td>\n",
       "      <td>1.161388</td>\n",
       "      <td>1.426493</td>\n",
       "      <td>0.494433</td>\n",
       "      <td>1.612622</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>4931 rows × 14 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "                                 CD20       CD3      CD45      CD68  \\\n",
       "BaselTMA_SP43_115_X4Y8_1     0.042121  0.052044  0.001735  0.042628   \n",
       "BaselTMA_SP43_115_X4Y8_2     0.000000  0.034770  0.000000  0.084581   \n",
       "BaselTMA_SP43_115_X4Y8_3     0.018200  0.078596  0.000000  0.042628   \n",
       "BaselTMA_SP43_115_X4Y8_4     0.000000  0.018701  0.015682  0.108294   \n",
       "BaselTMA_SP43_115_X4Y8_5     0.099789  0.073795  0.093230  0.122463   \n",
       "...                               ...       ...       ...       ...   \n",
       "BaselTMA_SP43_115_X4Y8_4927  0.000000  0.036476  0.142845  0.384770   \n",
       "BaselTMA_SP43_115_X4Y8_4928  0.000000  0.100622  0.089643  0.091315   \n",
       "BaselTMA_SP43_115_X4Y8_4929  0.144029  0.070877  0.203246  0.503439   \n",
       "BaselTMA_SP43_115_X4Y8_4930  0.036400  0.099307  0.118155  0.447235   \n",
       "BaselTMA_SP43_115_X4Y8_4931  0.054069  0.118611  0.053894  0.484753   \n",
       "\n",
       "                             Cytokeratin 14  Cytokeratin 19  Cytokeratin 5  \\\n",
       "BaselTMA_SP43_115_X4Y8_1           0.046076        0.076746       0.113790   \n",
       "BaselTMA_SP43_115_X4Y8_2           0.059064        0.056487       0.098873   \n",
       "BaselTMA_SP43_115_X4Y8_3           0.026254        0.126815       0.165966   \n",
       "BaselTMA_SP43_115_X4Y8_4           0.071382        0.052757       0.022484   \n",
       "BaselTMA_SP43_115_X4Y8_5           0.000000        0.238924       0.193627   \n",
       "...                                     ...             ...            ...   \n",
       "BaselTMA_SP43_115_X4Y8_4927        0.046530        0.000362       0.139654   \n",
       "BaselTMA_SP43_115_X4Y8_4928        0.000000        0.098437       0.000000   \n",
       "BaselTMA_SP43_115_X4Y8_4929        0.051248        0.200929       0.138156   \n",
       "BaselTMA_SP43_115_X4Y8_4930        0.285294        0.270529       0.107387   \n",
       "BaselTMA_SP43_115_X4Y8_4931        0.340787        0.133814       0.150674   \n",
       "\n",
       "                             Cytokeratin 7  Cytokeratin 8/18  E-Cadherin  \\\n",
       "BaselTMA_SP43_115_X4Y8_1          0.073571          0.110530    0.800019   \n",
       "BaselTMA_SP43_115_X4Y8_2          0.019242          0.152825    0.972369   \n",
       "BaselTMA_SP43_115_X4Y8_3          0.310404          0.253835    1.108996   \n",
       "BaselTMA_SP43_115_X4Y8_4          0.093607          0.062275    1.204820   \n",
       "BaselTMA_SP43_115_X4Y8_5          0.000000          0.125397    1.225161   \n",
       "...                                    ...               ...         ...   \n",
       "BaselTMA_SP43_115_X4Y8_4927       0.000000          0.000000    1.730274   \n",
       "BaselTMA_SP43_115_X4Y8_4928       0.000000          0.155443    0.908461   \n",
       "BaselTMA_SP43_115_X4Y8_4929       0.045556          0.057138    2.136403   \n",
       "BaselTMA_SP43_115_X4Y8_4930       0.423265          0.234452    1.884903   \n",
       "BaselTMA_SP43_115_X4Y8_4931       0.365944          0.163528    1.843372   \n",
       "\n",
       "                             Fibronectin      Her2  Vimentin  pan Cytokeratin  \n",
       "BaselTMA_SP43_115_X4Y8_1        0.324533  0.613138  0.044827         0.388409  \n",
       "BaselTMA_SP43_115_X4Y8_2        0.384034  0.940250  0.019424         0.567159  \n",
       "BaselTMA_SP43_115_X4Y8_3        0.256832  0.628524  0.040905         0.849462  \n",
       "BaselTMA_SP43_115_X4Y8_4        0.661950  0.530076  0.093848         0.418222  \n",
       "BaselTMA_SP43_115_X4Y8_5        0.623664  0.493992  0.000000         0.296923  \n",
       "...                                  ...       ...       ...              ...  \n",
       "BaselTMA_SP43_115_X4Y8_4927     0.233316  1.276169  0.171234         0.396967  \n",
       "BaselTMA_SP43_115_X4Y8_4928     0.411734  0.365833  0.110548         0.789458  \n",
       "BaselTMA_SP43_115_X4Y8_4929     0.841147  1.730263  0.214340         0.948885  \n",
       "BaselTMA_SP43_115_X4Y8_4930     0.767309  1.646707  0.134634         1.923792  \n",
       "BaselTMA_SP43_115_X4Y8_4931     1.161388  1.426493  0.494433         1.612622  \n",
       "\n",
       "[4931 rows x 14 columns]"
      ]
     },
     "execution_count": 44,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "celltype_data.get_exprs_df() # Return a pandas DataFrame corresponding to the expression data used"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "outputs": [],
   "source": [
    "ast.normalize()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>CD20</th>\n",
       "      <th>CD3</th>\n",
       "      <th>CD45</th>\n",
       "      <th>CD68</th>\n",
       "      <th>Cytokeratin 14</th>\n",
       "      <th>Cytokeratin 19</th>\n",
       "      <th>Cytokeratin 5</th>\n",
       "      <th>Cytokeratin 7</th>\n",
       "      <th>Cytokeratin 8/18</th>\n",
       "      <th>E-Cadherin</th>\n",
       "      <th>Fibronectin</th>\n",
       "      <th>Her2</th>\n",
       "      <th>Vimentin</th>\n",
       "      <th>pan Cytokeratin</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>BaselTMA_SP43_115_X4Y8_1</th>\n",
       "      <td>0.008424</td>\n",
       "      <td>0.010409</td>\n",
       "      <td>0.000347</td>\n",
       "      <td>0.008586</td>\n",
       "      <td>0.009215</td>\n",
       "      <td>0.015349</td>\n",
       "      <td>0.022756</td>\n",
       "      <td>0.014714</td>\n",
       "      <td>0.022104</td>\n",
       "      <td>0.159329</td>\n",
       "      <td>0.064861</td>\n",
       "      <td>0.122322</td>\n",
       "      <td>0.008965</td>\n",
       "      <td>0.077604</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BaselTMA_SP43_115_X4Y8_2</th>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.006954</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.016915</td>\n",
       "      <td>0.011813</td>\n",
       "      <td>0.011297</td>\n",
       "      <td>0.019773</td>\n",
       "      <td>0.003848</td>\n",
       "      <td>0.030560</td>\n",
       "      <td>0.193268</td>\n",
       "      <td>0.076731</td>\n",
       "      <td>0.186959</td>\n",
       "      <td>0.003885</td>\n",
       "      <td>0.113190</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BaselTMA_SP43_115_X4Y8_3</th>\n",
       "      <td>0.003640</td>\n",
       "      <td>0.015719</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.008586</td>\n",
       "      <td>0.005251</td>\n",
       "      <td>0.025360</td>\n",
       "      <td>0.033187</td>\n",
       "      <td>0.062041</td>\n",
       "      <td>0.050745</td>\n",
       "      <td>0.220020</td>\n",
       "      <td>0.051344</td>\n",
       "      <td>0.125376</td>\n",
       "      <td>0.008181</td>\n",
       "      <td>0.169086</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BaselTMA_SP43_115_X4Y8_4</th>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.003740</td>\n",
       "      <td>0.003136</td>\n",
       "      <td>0.021657</td>\n",
       "      <td>0.014276</td>\n",
       "      <td>0.010551</td>\n",
       "      <td>0.004497</td>\n",
       "      <td>0.018720</td>\n",
       "      <td>0.012455</td>\n",
       "      <td>0.238691</td>\n",
       "      <td>0.132006</td>\n",
       "      <td>0.105818</td>\n",
       "      <td>0.018769</td>\n",
       "      <td>0.083547</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BaselTMA_SP43_115_X4Y8_5</th>\n",
       "      <td>0.019956</td>\n",
       "      <td>0.014758</td>\n",
       "      <td>0.018645</td>\n",
       "      <td>0.024490</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.047767</td>\n",
       "      <td>0.038716</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.025077</td>\n",
       "      <td>0.242644</td>\n",
       "      <td>0.124412</td>\n",
       "      <td>0.098638</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.059350</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BaselTMA_SP43_115_X4Y8_4927</th>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.007295</td>\n",
       "      <td>0.028565</td>\n",
       "      <td>0.076878</td>\n",
       "      <td>0.009306</td>\n",
       "      <td>0.000123</td>\n",
       "      <td>0.027927</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.339496</td>\n",
       "      <td>0.046690</td>\n",
       "      <td>0.252541</td>\n",
       "      <td>0.034240</td>\n",
       "      <td>0.079310</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BaselTMA_SP43_115_X4Y8_4928</th>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.020123</td>\n",
       "      <td>0.017928</td>\n",
       "      <td>0.018262</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.019686</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.031084</td>\n",
       "      <td>0.180707</td>\n",
       "      <td>0.082254</td>\n",
       "      <td>0.073101</td>\n",
       "      <td>0.022108</td>\n",
       "      <td>0.157243</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BaselTMA_SP43_115_X4Y8_4929</th>\n",
       "      <td>0.028802</td>\n",
       "      <td>0.014175</td>\n",
       "      <td>0.040638</td>\n",
       "      <td>0.100518</td>\n",
       "      <td>0.010249</td>\n",
       "      <td>0.040175</td>\n",
       "      <td>0.027628</td>\n",
       "      <td>0.009111</td>\n",
       "      <td>0.011427</td>\n",
       "      <td>0.415244</td>\n",
       "      <td>0.167446</td>\n",
       "      <td>0.339493</td>\n",
       "      <td>0.042855</td>\n",
       "      <td>0.188656</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BaselTMA_SP43_115_X4Y8_4930</th>\n",
       "      <td>0.007280</td>\n",
       "      <td>0.019860</td>\n",
       "      <td>0.023629</td>\n",
       "      <td>0.089328</td>\n",
       "      <td>0.057028</td>\n",
       "      <td>0.054079</td>\n",
       "      <td>0.021476</td>\n",
       "      <td>0.084552</td>\n",
       "      <td>0.046873</td>\n",
       "      <td>0.368578</td>\n",
       "      <td>0.152866</td>\n",
       "      <td>0.323661</td>\n",
       "      <td>0.026924</td>\n",
       "      <td>0.375847</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BaselTMA_SP43_115_X4Y8_4931</th>\n",
       "      <td>0.010813</td>\n",
       "      <td>0.023720</td>\n",
       "      <td>0.010779</td>\n",
       "      <td>0.096799</td>\n",
       "      <td>0.068105</td>\n",
       "      <td>0.026760</td>\n",
       "      <td>0.030130</td>\n",
       "      <td>0.073124</td>\n",
       "      <td>0.032700</td>\n",
       "      <td>0.360796</td>\n",
       "      <td>0.230238</td>\n",
       "      <td>0.281564</td>\n",
       "      <td>0.098726</td>\n",
       "      <td>0.317179</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>4931 rows × 14 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "                                 CD20       CD3      CD45      CD68  \\\n",
       "BaselTMA_SP43_115_X4Y8_1     0.008424  0.010409  0.000347  0.008586   \n",
       "BaselTMA_SP43_115_X4Y8_2     0.000000  0.006954  0.000000  0.016915   \n",
       "BaselTMA_SP43_115_X4Y8_3     0.003640  0.015719  0.000000  0.008586   \n",
       "BaselTMA_SP43_115_X4Y8_4     0.000000  0.003740  0.003136  0.021657   \n",
       "BaselTMA_SP43_115_X4Y8_5     0.019956  0.014758  0.018645  0.024490   \n",
       "...                               ...       ...       ...       ...   \n",
       "BaselTMA_SP43_115_X4Y8_4927  0.000000  0.007295  0.028565  0.076878   \n",
       "BaselTMA_SP43_115_X4Y8_4928  0.000000  0.020123  0.017928  0.018262   \n",
       "BaselTMA_SP43_115_X4Y8_4929  0.028802  0.014175  0.040638  0.100518   \n",
       "BaselTMA_SP43_115_X4Y8_4930  0.007280  0.019860  0.023629  0.089328   \n",
       "BaselTMA_SP43_115_X4Y8_4931  0.010813  0.023720  0.010779  0.096799   \n",
       "\n",
       "                             Cytokeratin 14  Cytokeratin 19  Cytokeratin 5  \\\n",
       "BaselTMA_SP43_115_X4Y8_1           0.009215        0.015349       0.022756   \n",
       "BaselTMA_SP43_115_X4Y8_2           0.011813        0.011297       0.019773   \n",
       "BaselTMA_SP43_115_X4Y8_3           0.005251        0.025360       0.033187   \n",
       "BaselTMA_SP43_115_X4Y8_4           0.014276        0.010551       0.004497   \n",
       "BaselTMA_SP43_115_X4Y8_5           0.000000        0.047767       0.038716   \n",
       "...                                     ...             ...            ...   \n",
       "BaselTMA_SP43_115_X4Y8_4927        0.009306        0.000123       0.027927   \n",
       "BaselTMA_SP43_115_X4Y8_4928        0.000000        0.019686       0.000000   \n",
       "BaselTMA_SP43_115_X4Y8_4929        0.010249        0.040175       0.027628   \n",
       "BaselTMA_SP43_115_X4Y8_4930        0.057028        0.054079       0.021476   \n",
       "BaselTMA_SP43_115_X4Y8_4931        0.068105        0.026760       0.030130   \n",
       "\n",
       "                             Cytokeratin 7  Cytokeratin 8/18  E-Cadherin  \\\n",
       "BaselTMA_SP43_115_X4Y8_1          0.014714          0.022104    0.159329   \n",
       "BaselTMA_SP43_115_X4Y8_2          0.003848          0.030560    0.193268   \n",
       "BaselTMA_SP43_115_X4Y8_3          0.062041          0.050745    0.220020   \n",
       "BaselTMA_SP43_115_X4Y8_4          0.018720          0.012455    0.238691   \n",
       "BaselTMA_SP43_115_X4Y8_5          0.000000          0.025077    0.242644   \n",
       "...                                    ...               ...         ...   \n",
       "BaselTMA_SP43_115_X4Y8_4927       0.000000          0.000000    0.339496   \n",
       "BaselTMA_SP43_115_X4Y8_4928       0.000000          0.031084    0.180707   \n",
       "BaselTMA_SP43_115_X4Y8_4929       0.009111          0.011427    0.415244   \n",
       "BaselTMA_SP43_115_X4Y8_4930       0.084552          0.046873    0.368578   \n",
       "BaselTMA_SP43_115_X4Y8_4931       0.073124          0.032700    0.360796   \n",
       "\n",
       "                             Fibronectin      Her2  Vimentin  pan Cytokeratin  \n",
       "BaselTMA_SP43_115_X4Y8_1        0.064861  0.122322  0.008965         0.077604  \n",
       "BaselTMA_SP43_115_X4Y8_2        0.076731  0.186959  0.003885         0.113190  \n",
       "BaselTMA_SP43_115_X4Y8_3        0.051344  0.125376  0.008181         0.169086  \n",
       "BaselTMA_SP43_115_X4Y8_4        0.132006  0.105818  0.018769         0.083547  \n",
       "BaselTMA_SP43_115_X4Y8_5        0.124412  0.098638  0.000000         0.059350  \n",
       "...                                  ...       ...       ...              ...  \n",
       "BaselTMA_SP43_115_X4Y8_4927     0.046690  0.252541  0.034240         0.079310  \n",
       "BaselTMA_SP43_115_X4Y8_4928     0.082254  0.073101  0.022108         0.157243  \n",
       "BaselTMA_SP43_115_X4Y8_4929     0.167446  0.339493  0.042855         0.188656  \n",
       "BaselTMA_SP43_115_X4Y8_4930     0.152866  0.323661  0.026924         0.375847  \n",
       "BaselTMA_SP43_115_X4Y8_4931     0.230238  0.281564  0.098726         0.317179  \n",
       "\n",
       "[4931 rows x 14 columns]"
      ]
     },
     "execution_count": 46,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "ast.get_type_dataset().get_exprs_df()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "source": [
    "## 6. Saving models <a class=\"anchor\" id=\"six\"></a>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "source": [
    "After fixing the models, we can save the cell type/state assignment, the losses, the parameters (e.g. `mu`, `rho`, `log_sigma`, etc) and the run informations (e.g. `batch_size`, `learning_rate`, `delta_loss`, etc) to an hdf5 file."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "outputs": [],
   "source": [
    "ast.save_models(\"data/astir_summary.hdf5\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The hierarchy of the hdf5 file would be:"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<img src=\"img/hdf5_schematics.png\" style=\"width:1000px;height:300px;\">"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Only the model that is trained will be saved (`CellTypeModel` or `CellStateModel` or both). If the functioned is called before any model is trained, exception will be raised. Data saved in the file is either `int` or `np.array`."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 7. Plot clustermap of expression data  <a class=\"anchor\" id=\"seven\"></a>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "After fixing the cell type model, we can also plot a heatmap of protein expression of cells clustered by type. The heatmap will be saved at the location `plot_name`, which is default to `\"./celltype_protein_cluster.png\"`"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAfcAAAFzCAYAAAAjVEDpAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4xLjMsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+AADFEAAAgAElEQVR4nOzdd3xUVfo/8M8zfSa995BCJp0QCEGaKMoCqyAsICiuigtYvhZ0bavufl2+uj/XtWJZZVdRbLgLCAgKFqpGgQSBkE5CGul9JpmZTDm/PzLBGEMSIAnJ8Lxfr3ll5t5z7z33zmSeOefeex4SQoAxxhhjjkNyqSvAGGOMsYHFwZ0xxhhzMBzcGWOMMQfDwZ0xxhhzMBzcGWOMMQfDwZ0xxhhzMLJLXQHGBkJGRoavTCb7N4AE8I9WxpjjswE4abFYVowfP76m+0wO7swhyGSyf/v7+8f6+Pg0SiQSHryBMebQbDYb1dbWxlVVVf0bwLzu87mFwxxFgo+PTwsHdsbY5UAikQgfH59mdPRW/nr+ENeHscEi4cDOGLuc2L/zeozjHNwZYw6ttLRUdv3110eEhIQkREZGxk+fPn30iRMnlL0to9FokgEgLy9PERUVFd913uHDh9UxMTFxMTExcW5ubmODgoISY2Ji4iZPnqwdzP1g7HzwOXfGmMOy2WyYN2/e6Jtvvrl+x44dRQCQlpamrqiokI8ZM8Z0IetMTU015ObmZgPAwoULw66//vrm5cuXNw5kvRm7WNxyZ4w5rB07drjIZDLx6KOP1nZOmzx5smH27Nl6APjzn//sl5CQEKvVauMefPDBwIvd3ty5c8M3btzo1vn6uuuui/j000/dXnrpJe+ZM2dGTp06NSosLCzh0UcfDegs89prr3klJibGxsTExN1yyy2hVqsVZrMZ8+fPD9dqtXFRUVHxzzzzjO/F1o1dXji4M8Yc1okTJ9RJSUltPc3bsmWL66lTp1QnTpzIycnJyT527Jjmyy+/dL6Y7a1YsaLuvffe8wKA2tpa6fHjx50WLlzYDADHjx93+uyzz4pOnDiRvXnzZs+0tDT1kSNHVNu2bXM/evRoTm5ubrbVaqV//etfngcPHnRqaGiQ5efnZxcUFGTddddd9RdTL3b54W55xthladeuXa4HDhxwjYuLiwOAtrY2SW5urmrOnDn6C13n3LlzdQ899FBoVVWV9N133/WcN29eg0zW8TU7ffr0Fh8fHysAzJkzp2nfvn3OFouFTpw44ZSYmBgHAEajURIcHNw+f/785qKiItXy5ctDrr/++uYFCxa0DMAus8sIB3fGmMNKTEw0bN261aOneUIIrF69uvKRRx6pG6jtSSQSLFq0qP6dd97x+vjjj70+/vjjos55RPSLuzmICEII3HTTTXWvvvpqRfd1ZWVlZW3evNnttdde8920aZPHJ598UjJQ9WSOj7vlGWMOa+7cubr29nZ68cUXvTun7d+/X7Nz507nOXPmtHzwwQfezc3NEgA4ffq0/MyZMxfd4Ln77rvrX3/9dT+lUimSkpLOXrR34MABt7q6OqlOp5Ps2rXLffr06fo5c+botm3b5llZWSkDgKqqKmlBQYGioqJCZrPZcMcddzSuWbOmIjMzU3Ox9WKXF265M8YclkQiwfbt2wvvueeekFdeecVfqVSK4OBg02uvvVaWmJhoysrKUk2YMCEGADQaje2jjz46HRQUZLmYbYaFhZnDw8NNixYtaug6fcKECbqFCxeGl5SUqG688cb6yZMnGwDg8ccfr7j66qu1NpsNcrlcvPnmmyVSqRQrV64ME0KAiPDss8+WX0yd2OWHhOBxP9jId/z48eKkpKQB615l7EK1tLRI4uLi4jMzM7M8PDxsAPDSSy95nzx5Uv3uu++WXer6Mcdy/Phx76SkpLDu07lbnjHGBsjmzZtdo6Oj4+++++7qzsDO2KXALXfmELjlzhi7HJ2r5d7rOff/XDF10CP/jT9+R4O9DcYYY+xy0mtwlyrlQ1UPxhhjjA2Q3oO7QjFU9WCMMcbYAOkjuHPLnTHGGBtper1aXqpUDPqDMUchlUrHx8TExEVHR8fFxcXFfv31104Xu861a9d63XrrraEA8NBDDwX+5S9/8bv4mg5fPaVY7Wu/ux4jxi6l4fQ/2mvLXcItdzZCrV3iOn4g13f/py0ZfZVRKpW2zlSgmzdvdn3iiSeCZ86cmTeQ9RhKhsb8AT2Gag9tn8eQjUwzP1g3oJ+Vr3+/6pJ/ViwWCzrzAoxEvbfcFYpBfzDmiJqbm6Vubm49jnT2+uuve2m12rjo6Oi4+fPnhwNARUWFbNasWZEJCQmxCQkJsV999VWvrf5nnnnGNzIyMl6r1cZdf/31EYOxD8NNampq9N133x2UmJgYGxYWlrBr165fZXDbuHGj29ixY2MqKytlCxcuDLv99ttDkpOTY4KDgxPXr1/vAXTkeL/zzjuDo6Ki4rVabdy//vUvDwC45ZZbQj/66CM3AJg5c2bk4sWLwwDg5Zdf9r7//vsD8/LyFBEREfFLly4dNXr06PgpU6ZE6fV6vtvnEsnLy1OEh4fHL1myZFRUVFT8vHnzwrdu3eoybty4mFGjRiXs3btXs3fvXk1ycnJMbGxsXHJycszx48eVQEfgXrVqVbBWq43TarVxzz77rC8ABAUFJT788MMB48ePj3733Xc90tLS1ElJSTFarTZu5syZkbW1tVKg47N4xx13hCQnJ8dERUXF79279+zwwDk5OerU1NTo4ODgxK6peq+99trI+Pj42NGjR8e/8MILZ4dDfvnll73DwsISUlNTo5cuXTqqsxfqfL8TuuNz7owNEJPJJImJiYkzmUxUV1cn/+KLL/K7l0lPT1e98MILAT/88ENuQECApbq6WgoAd955Z8hDDz1UPWvWLH1BQYFi1qxZUUVFRVnn2tbatWv9S0pKMtVqtairq5MO5n4NJxaLhTIzM3M+/fRTtzVr1gTOnj377DHesGGD+6uvvur39ddfF3RmX6uurpanp6fnHjt2TLVgwYLRy5cvb9ywYYN7ZmamOicnJ6uyslKWmpoa+5vf/EZ/5ZVX6g4cOOCybNmy5qqqKkVNTY0AgO+//975pptuagCA0tJS1Ycfflg0efLkkt/+9rcRGzZs8Ljnnnsaeq4tG2xlZWWqTz/9tGj8+PElY8aMif3oo4+80tPTcz/++GP3Z599NuA///nP6cOHD+fK5XJs3brV5dFHHw3evXt34YsvvuhTUlKizMrKypbL5ej8PwQAlUply8jIyAMArVYb9/LLL5ded911+tWrVwc+9thjgZ2jDLa1tUl++umn3C+//NJ51apV4QUFBVkAcOrUKVVaWlpeU1OTNDY2NuGRRx6pVSqV4qOPPir28/Oz6vV6Sk5OjrvlllsajUaj5IUXXgg4evRotru7u23y5Mna+Ph4A3D+3wnd8dXyjA2Qrt3y33zzjdPy5cvD8/PzsySSnzvIdu/e7Tp37tzGgIAACwD4+flZAeD77793LSgoUHeW0+v10sbGxnP2rEVHRxsWLFgQPm/evKZly5Y1DdpODTGinhvCndMXL17cCACTJ09ufeSRR85+QaWlpbkcP35cs3fv3nxPT8+zI8PNmzevSSqVYvz48cb6+no5ABw8eNDlxhtvbJDJZAgJCbFMnDhR/91332lmzpypf+ONN/wyMjJUWq3W0NTUJC0pKZFnZGQ4/etf/yqtqamRBQUFmTrHhE9OTm4rLi5WDt7RYH0JCgoypaamGgBAq9UaZsyY0SKRSDBu3Li2Z555JrChoUG6ZMmS8OLiYhURCbPZTACwZ88e17vuuqtWLu9owHb+HwLArbfe2ggA9fX1Up1OJ73uuuv0ALBy5cr6xYsXn+0lu/nmmxsAYM6cOXq9Xi/p/JH9m9/8pkmtVgu1Wm3x9PQ0l5eXyyIjI81///vf/Xbu3OkOAFVVVfKsrCxVRUWFfOLEibrO7S9YsKAxPz9fBZz7O6G/Ix/2cc6dgztjF+Laa69tbWxslFVWVsq6JiKxJwL51eBQQgikp6fnODs792vgqL179xZ8+eWXLlu3bnV//vnnAwsKCk52flGNZH5+fpbm5uZf9EQ0NDRIw8PDTQCgUqkEAMhkMlit1rO/BEJDQ02lpaXKkydPqq688sq2zumd5YGOY9z1b3fh4eHm5uZm2eeff+42bdo0XUNDg2zDhg0eTk5ONg8PD1tNTQ0UCsXZhaVSqTAYDDyE9yXU9f2QSCRn32+pVAqr1UqPPfZY0PTp03Vff/11YV5enmLGjBnRwLn/DwHAxcWlX8Gz+w/RztdKpbLrZwQWi4V27Njhsn//fpf09PRcFxcXW2pqarTBYJD0NkLs+X4ndNfHOXf5oD8Yc0Q//fSTymazwc/P7xfn3WfPnt2yfft2z6qqKimAs92BU6dObfn73/9+9vxcWlqaGudgtVpRWFiomDt3ru7NN98s1+l00u4BcaRyc3Oz+fr6mrdt2+YCdByfffv2uc2YMUPf23LBwcHtmzdvPrV8+fLw9PR0VW9lp0+frtu0aZOnxWJBRUWF7PDhw87Tpk1rBYDx48fr3377bd9rr71Wf9VVV+nfeOMN/4kTJ/a6bTZ8tbS0SIODg9sB4O233z57nvvaa69teeutt3zMZjMA/KJbvpOXl5fV1dXV2nltxzvvvOM1adKks5+FTz75xAMAdu/e7ezi4mL18vKydl9Hp6amJqmbm5vVxcXF9tNPP6mOHz/uBADTpk1rPXTokEttba3UbDZj27ZtHp3LnM93Qk+45c7YAOk85w50/Or+5z//Wdz9atuUlBTjH//4x8pp06bFSCQSkZCQ0LZ58+bidevWla1YsSJUq9XGWa1Wmjhxom7y5MmlPW3HYrHQzTffHK7T6aRCCLrzzjurvb29z/nFMtK8//77p++5557Qxx57LAQAHnvssYr4+HhTX8slJSWZNmzYULRkyZLI7du3nzpXud///vdNaWlpzrGxsfFEJP7617+Wh4aGWgBg6tSp+oMHD7omJCSYTCZTe3Nzs/TKK6/UDdzesaH02GOPVa1YsSJ87dq1/tOmTWvpnP7ggw/W5ufnK2NiYuJlMpm47bbbap944ona7suvX7/+9N133z3q/vvvl4SGhpo++eST4s55Hh4e1uTk5Bi9Xi9dt27d6d7qsXDhwuZ169b5aLXauMjISGNSUlIr0NFb9OCDD1ZOmDAh1tfX16zVag1ubm5WADif74Se9Jo45sia/x30seUn/OWvfLUpu2icOIYxNlRSU1OjX3jhhbKup4AuVHNzs8TNzc1mNpsxa9as0bfffnvdrbfe2u/raC4ocQy33BljjLHB88gjjwQeOHDA1WQy0fTp01tuueWWAblAtver5eUc3BljjLGuDh8+PGCDU61bt658oNbVFbfcGWOMMQfD97kzxhhjDqbX4E7ySz+uLhGFANgAwB+ADcA6IcSr3cpcBWAbgM4rFrcIIdYMZT0ZY4yx4WIktNwtAP4ohDhKRC4AMojoayFEdrdyB4UQ11+C+jHGGGPDSq+D2EgUykF/9EUIUSmEOGp/rgOQAyBoQPaesQFWWFgov+aaayJHjRqVEBISkrB8+fIQo9FIaWlp6k8//dSts9xwSg05Uq1evTpw69atLgCwZs0aX51Od/b7TKPRJJ/PurqmjX3++ed9Xn/9da/eynd//+64446QL7/80hnoSD5SWVk5aN2enftWUVEhmzZtWtRgbWck6v45GCrn+3kbCiOh5X4WEYUBSAZwqIfZk4joOIAKAA8LIfo9wD5zPGkvXzmgKSgnP3igzxSUNpsN8+fPH71ixYqaBx54oNBiseDmm28e9cADDwTFx8cb0tPTnZYsWdI8EPUZinSUp5rqB/QYjnb3GtA0nq+88kpF5/O3337bb+XKlQ39HTq0N48++uivBjPpTXV1tTQjI8OpM6HIUAkMDLT4+fmZv/rqK6ff/OY3rUO57e6U964Y0M+K6fV/X9BnpbfPwUhP4Xq++mi5Kwb9QUSriCi9y2NVT3UhImcAmwGsFkK0dJt9FMAoIUQSgNcAbB2Ig8PY+fj8889dlEql7YEHHqgHOsY/f+utt8o++eQT76effjr4888/94iJiTmbYvRcqSHffPNNz8TExNiYmJi4m2++eZTF0jGCrUajSV69enXgmDFjYr799ttfpTt1BD3tu0ajSV65cmVwXFxc7KRJk7QVFRUyAFi4cGHY+vXrPZ555hnfmpoa+fTp07UTJ07Udq7rvvvuC4qOjo5LSkqKKSsrkwH9S6PZtVX+4osveickJMRGR0fHzZo1K7KnVuEHH3zgcc011/ziO2nNmjV+iYmJsYmJibEnT55UAsDHH3/sNmbMmJjY2Ni4yZMnazvrtHPnTueYmJi4mJiYuNjY2LjGxkZJc3OzZNKkSdq4uLhYrVYb9+GHH7r3dLzmz5/ftGHDhl57GRxVS0uL5KqrrhodHR0dFxUVFf/HP/4xoPvnoPv/zLZt21xiY2PjtFpt3OLFi8MMBgMBHb0t9957b9DYsWNjEhISYr/77jvN1KlTo0JCQhKef/55H6BjsJn+vCfDRe9jyysVg/4QQqwTQqR0eazrXg8ikqMjsH8khNjSfb4QokUIobc//wKAnIi8u5djbDBlZmaqk5KSfjFilaenpy0oKKj9oYceqpw7d25jbm5u9sqVKxuBjtSQ+/fvzz9y5EjOCy+8EGgymejo0aOqTZs2eaanp+fm5uZmSyQS8dZbb3kBgMFgkCQkJBhOnDiRO2vWLIcb7/xc+24wGCTjxo1ry87OzpkyZYru8ccfD+y63FNPPVXj6+tr3r9/f/6hQ4fygY5jNWnSJH1eXl72pEmT9K+99poP8HMazZMnT+Z89tlnhXfddVdYb3VatmxZ48mTJ3Py8vKyo6OjDWvXrv3V90paWppzSkrKL1rOrq6u1szMzJw777yz5r777gsBgJkzZ+qPHTuWm5OTk71o0aKGNWvW+APAiy++6L927dqS3Nzc7B9//DHX2dnZptFobDt37jyVnZ2ds3///vwnnngi2Gb7dafElClTWg8fPuyQP/T6smXLFld/f39zXl5edkFBQdYTTzzR4+eg839m2rRprXfeeWf4p59+Wpifn59tsVjwj3/8w6dzfSEhIe3Hjh3LnThxov6OO+4I+/zzzwsPHTqU+9xzzwUCQH/fk+Gij275S5/NkDpS7bwDIEcI8dI5yvgDqBZCCCJKRcePlvohrCZjvWZ86ymVaU+pIXft2uVy8uRJTVJSUiwAGI1Gia+vrwXoyDB1++23Nw76jlwi59p3iUSCFStWNADAHXfcUf+73/1udF/rksvlYunSpc0AMH78+NZvvvnGFTj/1LoZGRnqv/zlL0E6nU7a2toqnT59+q9Oq1RXV8u7Jwi67bbbGgBg5cqVDU899VQIAJw+fVoxf/784NraWnl7e7skJCTEBABXXHGF/uGHHw658cYbG2666abGyMhIm8lkotWrVwf/+OOPzhKJBDU1NYry8nJZ5xj4nQIDAy01NTXD6/zpEBk3bpzhySefDLn77ruDbrjhhubZs2f/6gdv1/+Z48ePq4KDg01jxowxAcDtt99e/8Ybb/gCqAGAG2+8sQkAEhMT21pbWyUeHh42Dw8Pm1KptNXV1UldXFxs/XlPhouRMIjNFAC/B5BJRMfs054AEAoAQoi3ACwCcDcRWQAYACwVvQ2az9ggSExMNHTN6gQADQ0NkqqqKoVUKv3V57Gn1JBCCFq8eHH9G2+8caZ7eYVCYXPkc4bn2vdXX301oOvrc+V870omkwmJRNL5HBaLhezbOK80mqtWrQrftGnTqUmTJhnWrl3rtX//fpfuZVQqla176tfObdvrKwDg3nvvDX3ggQeqli1b1rxjxw6XNWvWBALA3/72t6r58+c3b9u2zW3y5Mmxu3btyj948KBTfX29LDMzM0epVIqgoKDEntLLtrW1kVKpHL7Nx0E0ZswY09GjR7M3b97s9uSTTwZ988033U/X/uJ/pq+Q0JkuViKR/CqVrNlsprffftuzP+/JcNFHt7xy0B99EUJ8J4QgIcQYIcRY++MLIcRb9sAOIcTrQoh4IUSSEOIKIUTaAB0fxvpt3rx5OqPRKOm80tpiseCee+4JWbx4cZ2/v79Zr9f3+UUwe/bslh07dnicOXNGBnRcrJWfnz8sfmUPtnPtu81mw/r16z0A4L333vNKTU39VZY2Jycna3Nzc5/H93zTaLa1tUlCQ0PNJpOJNm7c6NlTmejoaGN+fv4vvsw2bNjgCQDvvPOOR3JycisA6HQ6aWhoqLlzPzrLZmVlKVNTUw3PPvtsVWJiYuvJkydVzc3NUm9vb7NSqRSff/65S0VFRY+fgZMnT6q0Wq2hr/12RMXFxXIXFxfbPffc07B69erqY8eOaXr7HIwdO9Z45swZRec1EBs2bPCaNm1avzP+9fc9GS5G1NXyjA1nEokEW7duPbVq1apR//jHPwJsNhtmzJjRvHbt2jMtLS2SF154ISAmJibuj3/8Y+W51jF+/HjjU089deaaa67R2mw2yOVysXbt2lKtVts+lPtyKZxr39VqtS0rK0sdHx/v7+LiYt2yZUtR92Vvu+22ujlz5kT5+vqaO8+39uR802g+/vjjFampqbFBQUHtsbGxbXq9/ld5v+fNm9f8z3/+0+ehhx46m5XQZDLRmDFjYmw2G23cuLEIAJ588smKm266KdLPz689JSWltbS0VAkAzz//vG9aWpqrRCIRWq3WsGjRouampibpnDlzRickJMTGx8e3hYeHG3uq39dff+0ye/bsAbkDY6TJyMhQ/+lPfwqWSCSQyWTizTffLDl48KDzuT4HGo1GvPXWW8WLFy+OtFqtSEpKanv44Yf7fWfEihUrGvrzngwXvaZ8bT5VMOhd226jozjlK7tonPLVcWk0muS2trafLnU9ejN+/Pjo3bt3n/L29rYO5XZTUlKiv/zyy1M+Pj5Dul02fFxQylepklvujDHWl3/84x/lhYWFCm9v7yHrIq+oqJA98MAD1RzYWU+G/dXyjLHL23BvtQPAjBkzhnwQmcDAQMvvf//7Acn9zRwPn3NnjDHGHEwfWeHkQ1UPxhhjjA2Q3m+alQzbW/gYY4wxdg69BncBvpCdMcYYG2l6bZpbrGLQH4w5gqqqKmln8g9vb+8kX1/fMZ2vjUbjBf1K3rFjh8vVV189GvhlSlLWYbimfE1NTY0+cOCA5ny2fy7Tp08fXVdX96t76/ujax1XrVoVvH379l+NrsccV68t93bL4N9hMSD/AYx1U3XsjQFNQek/9n96TUHp7+9vzc3NzQY6vlSdnZ2ta9asqR7IOgy1t3OPDugxvDNmHKd8PU/79+8/NRDrefjhh2uWL18+at68ef0ekY2NbL223NvNlkF/MHY52bRpk2tcXFxsdHR03KRJk7RAR+rKxYsXhyUkJMTGxsb2mUry3Xff9YiKioqPjo6OS0lJiR6amg8NR0n52qlrD8L69es9Fi5cGNZZ92XLloVOnDhRGxwcnLhz507nxYsXh0VERMR3lgE6UpFWVlbK8vLyFBEREfFLly4dNXr06PgpU6ZE6fV66m8dtVpte1NTk6y0tNRxkxOwX+g1uJsttkF/MHa5qKiokN17771hW7ZsKczLy8veunVrIQA88cQTAVdffXXLyZMncw4ePJj31FNPBbe0tJzzf/O5554L+Oqrr/Lz8vKyd+3aNSAtu+HAkVK+9kdzc7Pshx9+yH/uuefKlixZEvXII49UFxQUZOXm5qp7GvO+tLRUdf/999ecOnUqy83NzbphwwaP/tYR6Mh2tmfPnssyPezlqNdfcSZuWTM2YPbt2+eUmpqqi4mJaQcAPz8/q3266+7du93Xrl3rD3SMS37q1KlzDjKRkpKiX7ZsWdjChQsbly1b5jApYB0p5Wt/XHfddU0SiQTjxo1r8/LyMqemphoAQKvVGgoLC5WTJ0/+xWh3QUFBps5pycnJbcXFxcr+1hEAfHx8LGfOnOHBSy4TvZ9zb+dRDRkbKOfK6y6EwKZNm04lJSWZuk6vqKjocaCJjz/+uHTPnj1O27dvdxs7dmz8sWPHsvz9/Uf8P6sjpXztqa4Gg+EXFe9MMSqVSn+VYrSzvl11LSOVSkXnNvtTRwAwGo2kVqu5u/Qy0Wu3vMlsGfQHY5eLq6++uvXQoUMuubm5CqDjQiz79JYXX3zRz2br+N79/vvve01DmpWVpZwxY0brK6+8UuHh4WEpKipyiNaYI6V87eTl5WU+evSoymq1Ytu2bR591e9C9KeOAFBYWKhKSkq6LNPDXo56b7mbR3xjgLFhIzAw0LJ27driBQsWjLbZbPDy8jKnpaUVPPfccxWrVq0KjYmJiRNCUHBwsGnv3r3nPJf+4IMPBhcXFyuFEDR16tSWK664wiG+sB0p5Wunv/71r2duuOGG0QEBAeaYmBhDa2vrgI8M1p86mkwmKi4uVl555ZVDPgY+uzR6Tfn6Xfrgp3ydmtJ3ylcimg3gVQBSAP8WQjzXbb4SwAYA4wHUA1gihCge+Nqy4YpTvjouTvl68TZs2OCekZGhefXVVyv6Ls1GkgtK+TocWu5EJAXwBoCZAMoBHCGi7UKI7C7F/gCgUQgxmoiWAvg7gCVDX1vG2OXoUqR8PR8Wi4X+/Oc/j+hxF9j56WMQm2Fx7UUqgFNCiCIAIKKNAG4A0DW43wDgafvzTQBeJyISvXVLMMZGhOHeagcuTcrX83HHHXc4zF0VrH/6uBXu0rfcAQQB6DryUzmAiecqI4SwEFEzAC8A3E3LGGPsstNHt/zgt9yJaBWAVV0mrRNCrOtapIfFurfI+1OGOTabzWYjiUTC7ztj7LJgs9kIQI+BuveW+xCMLW8P5Ot6KVIOIKTL62AA3S8K6SxTTkQyAG4AGgaynmzYO1lbWxvn4+PTzAGeMebobDYb1dbWugE42dP8S95y74cjAKKIKBzAGQBLAdzcrcx2ALcB+AHAIgB7+Hz75cVisayoqqr6d1VVVQL6GL+BMcYcgA3ASYvFsqKnmcP+gjr7OfR7AexGx61w7wohsohoDYB0IfwsknEAACAASURBVMR2AO8A+ICITqGjxb700tWYXQrjx4+vATDvUteDMcaGgz4uqLv0wR0AhBBfAPii27S/dHluBLB4qOvFGGOMDUfDvuXOGGOMsfPTR8udT1szxhhjIw233BljjDEH08etcNxyZ4wxxkaaPoI7t9wZY4yxkaaP+9y55c4YY4yNNL0G9+ExtDxjjDHGzkevwd3I59wZY4yxEYcvqGOMMcYcTB/BfaiqwRhjjLGB0nu3vHmoqsEYY4yxgdJrcDdwy50xxhgbcTi4M8YYYw6m1+DeysGdMcYYG3FICL4injHGGHMkkktdAcYYY4wNLA7ujDHGmIPh4M4YY4w5GA7ujDHGmIPh4M4YY4w5GA7ujDHGmIPh4M4YY4w5mF4HsWEXp+rYG+c1iIDbqJmDVZVh4Qx5XeoqsH76tqrkUldh0G06kt7j9AOHDg9xTUYu0+v/pktdB9YzbrkzxhhjDoaDO2OMsUuOiG4notftz58moof7KBs4dLUbeTi4M8YYG2luB8DBvRcc3BljjA0aIrqViE4Q0XEi+oCIfIhoMxEdsT+mnOf6FgFIAfARER0jouuI6LMu82cS0Rb7cz0RvUhER4noWyLysU+PJKJdRJRBRAeJKGYg93k44ODOGGPsghDRKiJK7/JY1W1+PIAnAcwQQiQBeADAqwBeFkJMALAQwL/PZ5tCiE0A0gEsE0KMBfAFgNjOwA1gOYD19udOAI4KIcYB2A/gf+3T1wG4TwgxHsDDAN48rx0fAfhqecYYYxdECLEOHYHyXGYA2CSEqLOXbyCiawHEEZ290N6ViFwuog6CiD4AcAsRrQcwCcCt9tk2AJ/an38IYAsROQOYDOC/XeqgvNDtD1cc3BljjA0WAtD9lmAJgElCCMMvCtJF3VW3HsDnAIwA/iuEsJyjnLBvv8ne6ndY3C3PGGNssHwL4EaijkEuiMgTwFcA7u0sQEQXEmR1AM629oUQFQAqADwF4L0u5SQAFtmf3wzgOyFEC4DTRLTYvn0ioqQLqMOwxi13xhhjg0IIkUVEzwLYT0RWAD8BuB/AG0R0Ah0x6ACAu85z1e8BeIuIDPi5F+AjAD5CiOwu5VoBxBNRBoBmAEvs05cB+CcRPQVADmAjgOMXso/DFQlxXoOoMcYYY8OO/R75n4QQ73SZphdCOF/Cal0yHNwZY4yNaPaWeSuAmUIIU5fpHNwZY4yx4YSI3gDQ/T74V4UQ63sqz37GwZ0xxhhzMHxB3SBqb2sTusICuERGob2+Fpa2NjiFjEJxVROsVoGwQHc0thhhsdqw+UAhVl0VgncPluOO6SFY9s8czB/vDjcnOdRKGdrNNqiVMuzLrMWt10Zi/VenEOihxNHiVkyPdcOyqyJR02rBE++fQItRQCkD/nZrPDYfKMKeHAPiA2X43eQgqBQyvLGzCL+/Kgh6gxlymQT55ToAQICnCvMmhODvW7Kx7NrRKChrQHG1HlszdLgmToMdJ9owyoPQ0CYwJ9EZWeUGBHnIIZcRgrzUyC3Xo6jWjMWTfJBT1oL9eUYsnOCGk2V6FNRY8cjcYLy+qxxLrvDG9ox6PHB9ODYeKEVMkBMCPdUorNRjT44eEyNU+P6UEXdf6489mXXwcZVjYrQX2owWnKlvg4+bEk4qOY7kN8DTRY5wf2dIJRI8uekMlk92QYCnGsXVrThS1IbfJntAZzAju7wNo7yVGBPugcr6Nvh7qpFZ3ARtkAuqGo2ICnLFmbpWtLSZUVpnRHK4G1w0cqTl1CMqwAnuzgpkl7XATSODxSrQbrEhyEsNpVyCBl07XNRyvHewDi5KIHmUGgfyDZgZ5wRntQyHCnSYGOWCTw81I9CNUFgvcEOSGifLjbhhgjea9O3YdrQZM2Kd0G6x4XipEfWtAitn+OBfe2oR4S1FqLcCH6YbcFOyCgcLjPB2JuhNAoFuUtS3WmETwD2zgqFrM2NnRi3CfBSobDKjodWGP1wTBJlUApVChtd2nsakKGeE+TlD12aGNsQdG/eX4Oarw/D/Np3Coiu84apRICHCB3uOlp59T/3dpPjd5CBIJRJ8n1UNLxcFfshvQXK4E6KD3XDfR+VYNl6FsREeKDijw7o0PRaNUcJksUEb4IS0/BZE+CpQVm/GK6vG47ZXjmBcmBrHSw2I8JEjr6od7VYgwluGPacseGy2F/69rx4po+Q4ccaMW6d6wdNFibLaVrS0WSCTEpIiPFHd2Aaj2YqqRhPKG0wI81ZBKiGE+zuhpc2M7RkNSB6lhkYhhatGBgkRfshvxq0zQvHPXSUIdJdBISNMivHC0VONCPRUYdk1WrQ3NUEn02DLgVPwdVchKtgdcWE+EABaWk0gIrhoFIDJiHapAkopUNnQBte2eqi8vCGsVuzPb8CUMSE4mleJ2iYDrp8yGgVl9dAo5Ug7WYmlV2vxwTe5uGG0EhKZFHUKdwTIzZCpVMitMWJUgDuU7a1oLiqC99hxqKzXI8BDjcY2MzQmHdp1OrTV1CDD5oNrolzxTX4zZkZ7oPFUAQDAEhEPAPDzdEZ5TTNqm9oQEegBJ7UCZ2paEBnkcfYetZZWE1ydlEjLLEN1owG/jXWHTKWCzMUVpvpamJqaYGpqgkQuh8rLC04BgahqMiIk2Iezwg1TfCscY4wx5mA4uDPGGGMOhoM7Y4wx5mA4uDPGGGMOhoM7Y4yxQUFEVnta1s5HGBGlENFa+/yniejhYVBPdyK6p8vrQCLadCnrdLH4annGGGODxdBDgpZidKRs7RcikgohrANaq19zB3AP7Klf7WPVL+p1iWGOW+6MMcaGDBFdRUQ7ukxKIqI9RFRARCu7lNlLRB8DyLRPe4iITtofq+3Twogoh4j+RURZRPQVEant8yKJaBcRZRDRQSKKsU/3I6LPiOi4/TEZwHMAIu29C/+wr/ekvfztRLTFvq4CInp+6I7WheOWO2OMscGiJqJj9uenhRALeigzBsAVAJwA/EREO+3TUwEkCCFOE9F4AMsBTERHGtlDRLQfQCOAKAA3CSFWEtF/ACxER+72dQDuEkIUENFEdLTKZwBYC2C/EGIBEUkBOAN43L6tsUDHj4ZudRwLIBmACUAeEb0mhCi7qCMzyEZUcE9JSXkcgOpS16M/0tPTn77UdWCMscFERKsArOoyaZ0QYl2X1z11y3e3zZ7VzUBEe9ER1JsAHBZCnLaXmQrgMyFEq327WwBMA7AdHT8aOn9AZAAIIyJnAJMB/LdLnnil/e8MALcCgL27v5mIPPqo47dCiGb7trMBjALAwX0AqThoMsbY8GAP5Ov6LNjHas7xurXLtN5GwjN1eW4FoEbHKeemfvyw6K/u2xj2sZPPuTPGGLuUbiAiFRF5AbgKwJEeyhwAMJ+INETkBGABgIPnWqEQogXAaSJaDADUIck++1sAd9unS4nIFYAOgMtA7dBwwMGdMcbYpXQYwE4APwL4P/uV6r8ghDgK4D172UMA/i2E+KmP9S4D8AciOg4gC8AN9ukPALiaiDLR0Y0fL4SoB/C9/WK9fwzAPl1yw75rgTHG2MjUUy51IcQ+APvsz58+x3Jny3SZ9hKAl7pNKwaQ0OX1C12enwYwu4d1V+PnQN91+s3dJiXYp7+Hjh8WneWu76nOww233BljjDEHw8GdMcYYczAc3BljjDEHI3366acvdR36bd26dVetWrVq36WuR3+lHSt+2ickEMcLqqB0cUFunQkhHiq4qaTwdHMCSSSoaWxFfXMb5k2JwI8FdThR3IKJSaG4JtEbyVo/OKvk+HBfKRZMCYOHqwr7T9bgRFED/u/2CRg72hdHcqtgMttQWquDXEpoM5hwy1UhKK/VIyncAwez6rD4Ch80t5pxRawfiAgnSxqREuWFrYcqEeihxJVJQUiM8MZHe0tx7YRQ7DxyBlV1emSW6FCnM+Oe2aGYP200nCUG3DApCKXVLXhg4Ri4KAWKa9rg76FERpEOFhtwz28jsOWHCsxNDUJVgw7xoS4orDTA11UKg9EMuRQoqjHAWSVBTlkzBAB/dyWOnGpCQ6sF0f4KWKxAfJAKWWU6SCWEhlYLKusNOF3TBqVcgsnxAThR1ICSOhNqWswYN9oDVqsNwmyEq1qKjKIWNLZa4OkkRUmtEVargNFig85oQ2ltG0rqjFDKCIXVRlwR443ECG98vLcYTa1mnGlsh1xKaG41o7LBCJsQsNoEIIBT1Qa0Gq245ZooZJ5uwOkaAwoq26AzWGCx2jA91g2NehPMFgEvJwlqWyxQyQlGsxWTY71Q29gKX1cZPNRATLATHvxdIvYdq0BpnRGuagl0BitaTTZolBJ4OhHqWkxwUxMSQjQoqTMhzIMwPcEbkb5ymNotUMkILmoJ3NRSqOWEwspW3HVdLH7MqUaLwQYPjQxOSkJmqQ6V9QacLGmGTArcPTce735diJrmdhwtbEK7RSAlyhPCYsKxYj1M7RZUNejh5qRAUbUBSjmQOtoNx0834WRJC0xmG+p1ZjirOtoGOWV6LEpxQ3VTOyL8nfBDfhNifKVwd5Kh1WRDkJcKdbp2OCklkEsJO46cgZOSYDILKGSAn5sC7WYbgjxkcFHJ4KkSONNoRKCbFCaLgI+LFGX1RjTq2lFUY0RTmwXNbVZklbaguskEvcGKmhYzwn1UqGxqh1ohQUFlG6w2W8fDKtDQakFNczuIAKkEaNKbYBMdnwujWaCstg0miwAgkDjaFwqJgNJJgwlaH8SM8kKr0YyqBj1cnVRQyGUQNWegcHNDWX0bvFxVsILgqlHAqnaBXKWEVC5HoK8rMgurMT46EC2tRoS4KeAKE1w9PaBWSODUrsP4hFC0KZzgrFFCZdTjQEkbosL94eGqATXVorRNAjeZgEyphMKkh7DZYJUqUNJogquvL+TWdkSHeUPh4YWoEE+UN7fDfdQoyHwC4O6shrNGAaFvhrypFr6hQXCTCZiqKqAjJRpaDFA118BqaEOzTQZUliIi1Bfx4T6QOTtDKJQwtFuxP7sW0WG+ONokg0HthlHhQTDbCG5OCkjl8r9e0i9Zdk5DekHdAAxCEzZAVRkSceE+sNpsZ19PSghGc6sJrvajbrH+fHtnZX0rymp+vq3zzc9z8ODCs9eJYO22HNw7LwYA8MzyCaiq18Pb3QkAMHNcAHJLG8+W/eRAKWSSjttC3dRS7M+qh7umY6MtrV1v1wRCfH6+3kUpJ5TXNJ99bbLYzq6nrLoZx063IC7Ms2NeuwXT4/3xfXYdAOCFFSl47N0MAMCq2ZEI8nUDDnaM8RAbpEZepeHsel1UUrS1dxyXm68aBavVhuxyPQCgQW8BEcHNXl9zl2NksghI6Je3uyplP7+22n4u66SUoNVk61JO8ot15Z75+Viv+yIPv5scjC+OdFyk224RUMs75hnabdAZ2+HprAAA6I02vPRZNoI8Ol6r5D93fn2X24iu1ZNKASF+3qa8S11vSPDE3z75CX5uHesxWwQWTQnE54crAeAX6ymp7fgx0tbesa6jRTpIznHX738PnEJbuw1Se4HO49yp6/Ezmm2YkeiFPZn1AID8KsPZ+c4qGcrr2s6WzSnXw1kl/cWyGoXEfrxsKKn9+f3VKCS/2G52mf7sc73JhkhfNQprOsoLAVQ3m8/ON1ttXZ6Ls/sBdBw/qYTOvs9d5wnxy9dd32sACPRQoKKxHUDH52jV3ET85f1fD29e2WTGsfxKXBHhDgBoKSqES3DI2flKGcFsH+W8oLwRUX5drhez2aAmKwQ6jlPXz0ZqpAeE+HnfRmlskHv6A/Yh02tNErja59W3GOCpJHQOpn5cJ8e0CA3Mel1HHRQ/f22XkBtG1dZC7tpRX09XzS/2R9JuOLsetdWIk9UGhNsXjwzyhKFcd7asxtfv7PPs4jpEaKyAmzcAoLz157qbLDZU1+sQ6uf2q+PHho+hvlr+ogahSUlJueBlGWOMscsFn3NnjDHGHAwHd8YYY8zBcHBnjDHGHAwHd8YYYwOOiPYR0axu01YT0btEtGmQtx1GRDd3eZ1CRGsHc5vDDQd3xhhjg+ETAEu7TVsKYL0QYtEgbzsMwNngLoRIF0LcP8jbHFY4uDPGGBsMmwBcT0RKoKM1DSAQQDkRnbRPu52IthLR50R0mojuJaKHiOgnIvqRiDzt5SKJaBcRZRDRQSKKsU9/j4jWElEaERURUeePhucATCOiY0T0IBFdRUQ77Ms8be892GdfxiGDPgd3xhhjA86eae0wfk7eshTAp/h1/vYEdLSyUwE8C6BNCJEM4AcAt9rLrANwnxBiPICHAbzZZfkAAFMBXI+OoA4AjwM4KIQYK4R4uYfqxQCYZd/m/xKR/EL3c7gaaVnhjCPlXveLuZ+fMcZGAiJaBWBVl0nrhBDrurzu7JrfZv97Rw+r2SuE0AHQEVEzgM/t0zMBjCEiZwCTAfyXfh6ESdll+a2iY4SgbCLyQ//sFEKYAJiIqAaAH4Dyfi47Ioyo4J6env5c36UYY4wNBXsgX9dLka0AXiKicQDUQoij9u75rroOm2nr8tqGjhglAdAkhBh7jm10Xf4cYzf2uowVIywW9gd3yzPGGBsUQgg9OvKyv4uOVvyFrKMFwGkiWgwA1CGpj8V0AFwuZHuOgoM7Y4yxwfQJgCQAGy9iHcsA/IGIjgPIAnBDH+VPALAQ0XEievAitjtiOVxXBGOMseFDCPEZunSXCyGK0XERHYQQ7wF4r8u8sC7Pz84TQpzGzxfmdV337d1eO9v/mgFc0634Pvu8p7stkwAHxC13xhhjzMFwcGeMMcYcDAd3xhhjzMFwcGeMMcYczHlfUJeSkvI4ANUFbi/sApcbkXSHv0NtRjosZyrgvHo1Dv75YcidndFaVY2r1r+P/XfcBu/EeNRlZqEOwOTnX4OrUx1Mp3JQ02KBU3srVB7uIAL+3+3jsDu9BKO8lThV3oCC8kYUV7fC102BfSeqUF7fjkMFLfBzkyPYUwmZhPDhniK4qKW4VuuJ/6ZV4W//yUFiiAZquQSlNTrEBGrw1q5ihHop4OMqR9IoF/h5usDXVYZQXw1unxOHjd/m4sv0ClQ2WTAu3AkffHsa4T4qfPRNHuakhmJKnDdOV+mQVVyP5jYrXt9RhD/MDIWGrAAApVwKF7UciaFStBotEELgf+bF4z/7CiAhwsHMKkQGOGN8pBuOnW6BRiFBcV07lLKO629MZoHHlySg1dAOr7Za/DfPBHn2EUQFxUIpk6C62YS4cF9s2pcPqYQwbUwANMpaLJsZg7teOwQACPaUwGi2wdhuQ0qkO4ztVjipZFAppAj0cUVbuw1p2TW4KtEHB7PqcNUYX3i7qSGXSbH7SBm8XBU4kN0EbYAKTiop3DRyFFW1QSWXYNb4AAT5uOK1bdmYmeSNuHBvbP3uNHzdlXDTKLAzoxZmq0BtkxERvir8VNyKUC8FcptsaGy14k+zvLEurRZ/8CnFt2UapES6IrtMD09nOWpa2qFRSBAb5IzaFhOqm81w0Sjg6STDBK0HjuQ34nStCR5OUuiNNiy8wg+Jkb44U29AcW07JBpAISXEBTujqsmIGyaNwu6MM0jLLEdymAsmxQfgUHYVFk0OQk5JA2KDNNAoZAjy1mDjd5WI8FVAZ7DCZBGYpHVHVLA7fsyuhlwmQZifM44WNqLNZMMDC+JRUFYPXZsZOoMZ0+K84OOuRnGVDrEhUnx5tA7J4c6w2myQSgjJoz2hM9Tgt6nB2PRdKVbM0eJQdiXMVhv0BgueWhaPjd/mIeN0C7QBauScaUOghwJRgc5o0DdhtJ8Kni4KBHg54XBeHYztNgR7KtDUakFSmAuKqtqwfGYkdqeXYbSfGjXN7citMMLXVQZPZzla2iz45JscrF4Qj5ziWrQaLZg9MQKAQHFVM4oqmtBokeGpNw7hlSVh2PJjCXQGC2amhOKu1w7hrfuuwM7idjS1VsKpqBbPZ2swO9kbPu5q7DtRhapmM3QGG1bNCsOEmEA8uf4IfN0UiAtxRU2TEQdzCwEACybqcDC7AXU6C57/wzh8mamHk0oOuUyKu9cdBQBoFBIkhGpgtZUiLMANxiYLXnn/MP7nugg8+8lxRPiqsFNvhldOJm66NgYni2oQ6O2M7zMrER7gghNFjbhu0ig8+0U9rk6wIrNEj+hADdydLSiraUFLqxlVjS1YOTcA//P6IVyT6IGyOgP83ZXwSw3D8x/+BABoNZqRXabHuEg3RNgakKdTo6HFgAlJEUP7pcr67UKulldd6OhrI2V0ucHklZCA1qrqfpV95vPT+NNNyYNcowv3xo58zJ3g36+yeoMF1M/hJa5OcEf+mdZ+lfVzU/ZdaICdKNHBWSntV9mX7kzFfW8e6lfZw6f0mDnGo19lCyt0/So3kPYcq4RG0b/OvvXflODqBM9+lc0tqe93HeamBuBEUUO/yw+Er0804oqo/t0yfUWMN7YequxX2c8O1cDbpX9fwW0mS7/KMdaJu+UZY4wxB8PBnTHGGHMwHNwZY4wxB8PBnTHG2KAhIn8i2khEhUSUTURfEJGWiAz2vO05RHSYiG7rsswNRHTCno89nYimXsp9GIl4+FnGGGODgjpytH4G4H0hxFL7tLHoSLFaaM/bDiKKALCFiCRCiPUAvgWwXQghiGgMgP+gIwc76yduuTPGGBssVwMwCyHe6pwghDgGoKxrISFEEYCHANxvf60XQgj7bCcAAuy8cHBnjDE2WBIAZPSz7FF0aZ0T0QIiygWwE8Adg1A3hzYsu+UvcqCcYeFCxwJgjLGRgohWAVjVZdI6IcS6C11d1xf2bHKfEdGVAP4PwLUXuN7L0rAM7riIgXIYY4wNDXsg7y2YZwFY1M/VJQPI6WEbB4gokoi8hRB1F1DNyxJ3yzPGGBssewAoiWhl5wQimgBgVNdCRBQG4AUAr9lfj7ZfjAciGgdAAaD/QxmyYdtyZ4wxNsLZr3ZfAOAVInocgBFAMYDVACKJ6Cd0nILVAXjNfqU8ACwEcCsRmQEYACzpcoEd6wcO7owxxgaNEKICwI09zFL3sszfAfx90Cp1GeBuecYYY8zBcHBnjDHGHAwHd8YYY8zBcHBnjDHGHMxQX1BnTElJebof5cIGuR5Dwlhfh4ApU6Eek4KaXdvgFBQIfXk5ZGo1Mv73SXiPSYDP+BQ0nSqCZ4wWJVXNKKttw6zEMHg552LNbSl46r10WK3AY+uPIjpQjVAfDbJO10Mhl8BqE6htMcPHVQ4/NznaLTb4uSuQX2mAySwQE6iG0WxFekEjwn2UuH5iMD5LK4XVJtDSasZvJ4ahue0UjO02ZJUb8JuxGlBTLRr0FhRVtaL9x0LMSA7Ge1+fgoeTFEXVBqgVEjTozQjyUuHqBE/kljVDKZdg03elkBDhmkR3bD90Bv5ezgCAqLpsZLcFoanVAqmEMC7CFVmna9GgN2NClAdyyloQeGg7SlOuh5NSirZ2K5r0ZhjNNlwZ74XCSj3e/TIXM8cFYE+hCc1tFlhsBoyPCcSR/ExEB7ngv3vzEB/mgUZ9O2w2gWMlOpx4Jx1XRLnAbLEhMtAFmcXNGOWjQqvRgqpGE+JHueJIQRMSI32QEumKmBAPKBVSSAjIKKjHBK03fNw1+O3EUHz+QwmuindH3plWhPposPDKKHyfWY4zdW34+mglpsSZMSXWEydON2NPZj18XTv+rYqq9PDIq0C4jxJEQFVTOwLc5fB0loOIYBMAefkhwr8N37WOhpezHj+d1sFotqGt3YbEUGfkVbQi43QL1HIJhBCoqGtFdbMZEf4alNWbICFCTJATWo0W/JjXgEP5DfB2UWD+RD98c7wWja025Fe2wtdVgT3HKlBYbcLd80ah6EwDTp6ug9FsxbZDFZg9zg8Zhc0I8lRC2mhATKAKNgH4uMrg46qAVEII9Ow4fgCQW9aC2hYzvF1kaNIZ8M2xGowNd4UQgFIuxeQYN+Sd0aPdYoO3ixRF1W34TbIf9p6oga+HE6pbzNC1tcNqEzBbbMgq7dhvIsJf3s/AuAg3KGUSnK4xQiohRAc5Y5zWH1sOFsHXXQmzRcDYboGXiwL+HmqM8nfF+98UoazOAD93Bfy8nFHbYoabRorxo91x5dhQ7D5UBIVcCv92K1yd5PjLByegDVCiQW9B3hkdgr1UkEokMFttOHC8DF7OUhToJVgw0ReNuna4aJSI8lfipU3HMDbCHSaLFcqxE/HKlSq8vzsHZ+rbEOSlwfhwV+gMZlhtNmw9WIAQLxVc1DJkl7XA01mOCZHOqG5qx4OLkrD+yyzIpQRpYzWqGk1ICFPjn9uzEOmrhK+7EhNjffHPnYUI9NQgyMcNb247gbGjnFBU0YzfjvfDtkNV8HCSwmi24bMDBXDVyPHtsWpcmeADY7sFYyM94e3mhKVTA7A3sxYuKimsNgEnlQzuziqU1bbhilgftBrMeHlxKCqsauSWF6G4xoDvTpRjYpQbvjnRiLI6A8ZFuuGalHAcyT4DicSGD/eVYkJSxCX7fmW9G9Lgnp6e/lx/yvXzB8CI5D12LFpKSvtV9u8bfxrk2lyc6iZjv8sS9V2mk0za/8Jq5dDf8PHlj6fh6iTvV9maJkO/1xvqrUJZvalfZWub+1duIBVUtCLQU9mvsj/ktSDAvX/H6EhBY7/rkBDmfl7HdCDIpP3v4Jw9IQwb9+b3q+yBYyX9Xu/M1HDYbP0uzhh3yzPGGGOOhoM7Y4wx5mA4uDPGGGMOhoM7Y4yxQUNE/kS0kYgKiSibiL4gIi0RGYjoJyLKIaLDRHRbD8tOICIrES3qMs1KRMfsj+1DuzcjBw8/yxhjbFDYk798BuB9IcRS+7SxAPwAFAohku3TIgBsISJJ5/jyRCRFxxC0u7ut1iCEGDtU+zBSccudMcbYYLkagFkI8VbnBCHEMQBlXQsJIYoAPATg/i6T7wOwGUDNENTT4XBwZ4wxNlgSAGT0s+xRADEAQERBABYAcT5Q9QAAIABJREFUeKuHcioiSieiH4lo/sBU0/EM1275/g52M2ylp6c/fanrwBhjg4mIVgFY1WXSOiHEuv/P3n2Hx3XVif9/f6Y3adS7Jdly707kOHF6SEIIkEIIoSwQYAk8+4VdwvLlyxbAwC678NuF3ZBls14eCLCUbEgzJCSk9zi2Y8e9yE299xlNvZ/fHzOyR4rGlhLLdpTzeh49uufezz3n3JFmPnPrebPVZUz/G/D/VDUpb3xIRrWqtqYP5T8lIjtU9eCbbHPGOiuT+2QfdmMYhmGcOelEfqJkvgv44AmWZ1oF7ElP1wO/TSf2IuBaEUmo6oPpIWRR1UMi8kx6PZPcxzGH5Q3DMIzp8hTgFpHPjs4QkdVATWaQiNQC/wL8CEBVZ6tqrarWAr8D/kJVHxSRfBFxp9cpAi4Edp+G7XjbOSv33A3DMIy3P1VVEbkR+DcR+RoQAY4AXwLqRGQr4AGGgB+NXil/AouA/xIRi9TO6T+rqknuEzDJ3TAMw5g26cPoH5pgkXeS69+aMf0SsOzU9GxmM4flDcMwDGOGMcndMAzDMGYYk9wNwzAMY4Yxyd0wDMMwZhj7unXrprTC+vXrL7vtttuemY7OzDQPHkqsa0r6eGxLC23eUhqLFpAsqWR70XI+8H8+SsXaC/nyQ53EV5zPQOUCioIegn4nv3u5lT3tMa5ZVYiVtIglLdoGktSVejh3fimgNHaGaOgYYV6Zl5GYxabDEY70JCnNsbF2YQGWleSqcytJJJLsbwvTOZjggU091BQ6mFvm59k9A9iJseXQMNefX862w4P81Xvn8a+PHOLrNy+mMM/HirmlPLbpKDubR8j32znQGScWt4gmlHnlXp7fM0hlvotfbxzitisrWVabw/6WIcryPSSTSTYfHMJZUc7qBSU8tq2Ha84p5rWDA6yYU4CVTPLivj5GYhZ5K1dit9koDrr4w7YhCn1CTbEXh81Ga2+Uj181jz9uaqKuPEBloZdWVyEuhw2vS7DbBKfDxrM7uinOdZFIJPnMtYtZMiuH7oEw19U5Ka2ppLVriOGRJJ96z1LK8900dgyzuDqXjt4Qzd0j/GlbN4V+G6FIgubeGNddUIPTaWfP0R5K8z088XoPq+bkcOeTPcwvcbCvJQQCS2pyGQjFeGpHL0W5TmIJi8uXl7KopoClsws5Z3Y+//1kM+9aUcw58woZicQIeB1ctKKaulIP7T3DXLKimu0Hu+gLxVk1J8iRzhFyvXaurq8iPBJlR1OEy5bk0dAe4c/eNY99jb1cfW4lB1oG2dyU4NKFOURiSUqCbv7imrk8t7uLK1bNYmh4hP5wnHWfqOeRTc187cOriEdHCEeivNbQzc3nz6KpJ0xzT4QPXlLHz55q4XufPofob/6L2ZdfQsnv/4vc8y+kNN+LpcryOfmsXVpJW/cwkViS962pIhKLU1Oay+YDvVy/tpbkz3/IC4461i7MI5G0mFsR4NLl5QyGIly9eg7nzCskJ9TNE/tD2CRJYY4Lv8fOc/tH+FTLfdwXmc+tV1TSOxylssDNLbOibOkWFlblcrh9AJfDRmWRn4tXVtPVF8ImsKyumO0Hu7i5KkJBdSXRWJKNe9q5+bI5fP/3rdx+/Tzae4a4ZMUscn1u9jT2YxchnkyS63UCcO68AoYjCQ60DtM1GGPNwmKaOkNcEujn3l1hBsJxDrb0MxKzWLuoCI/LwX2vdFJT6OIXTzRwy+VzeXRzKyV5Lp7b1YXLYeOV/f08vWeY5r4YQ+E4eX4HdeU5PLy1l9tvXEhZ5wE85RXct6mP+bV5eFw2Ht7cRUHAgc9tZ3F1kF881UTnsMXLO9u4dnUlz77eRl8ojt9tZ3ldEb96oZNVNV7y/E6Kcj3sPDpIbYmPTQ39qFpcd2Edv3+xgVcP9HPxkiKOdIZ535pq7nzkKIPDI1QWehkaiTMcjtI0mGT5vFK8TqV7IEI0nqSlJ8qOljjFARvP7h2iobGbjv4RXA4bH/A2UrB46bfO7KeskY3Zcz/Nrqmvzrps0/6eMeXvP9yRNdZue8NTm7La3JiYdOxQJDnp2PesKpx0rKXZlxXnjb1otqEzljX23AXlY8prl8/KGnvO/NLJdQ74xBXZ6xmvIMc16dj2V1/JuuyHG8Y+d8OysteTk05Co+5/sTFrrMM++bd1VXFg0rHL6ib/ep7o790Q9Ywpl+V7skTCg21jly2sKcoa++87Jv+e+PC7Fkw6dkGFf9KxeTnZt2W8h3rysi7rGBj7Hmjoyv6+vPU9Sybd5sDI5N/fTodJD29n5q9nGIZhGDOMSe6GYRiGMcOY5G4YhmEYM4xJ7oZhGIYxw5jkbhiGYUwLESkTkd+KyEER2S0ij4jI/BPErxSRaydR7zoR+cqp7e0b2xeR69LPxJ9KHT8VkU4R2Zll+VdERNMD30wbk9wNwzCMU05S47U+ADyjqnWquhj4W+BEt12sBE6a3N9iv040psqY9lV1g6pOdQjyu4FrsrQ9C7gKyH67yylikrthGIYxHS4H4qp61+gMVd2mqs+LyC9F5PrR+SLyKxG5Dvg2cIuIbBORW0SkQEQeFJHtIvKKiCwf34iIfFZE/igiXhGpE5FHRWSLiDwvIgvTMXeLyA9E5GngeyJynoi8JCJb078XiIhrgvZvFZE7M+q4Ix1/SEQmHKdeVZ8DerO8Jj8Evgqc4GbRU8OMCmcYhmFMh6XAlizLfgLcDjwkIkFgLfBJ4BtAvap+AUBEfgRsVdUbROQK4Bek9q5JL/8CcDVwg6pGRWQ98HlVPSAia4AfA1ekw+cDV6pqUkRygUtUNSEiVwLfVdWbRGR8+7eO63c5cBGwENhAaqz5SUl/eWlR1ddTBzWm1xlN7vX19V8jNZbvjLN58+Z1Z7oPhmEY00lEbgNuy5i1XlXXn2w9VX1WRP5DREqADwD3pRPt+NCLgJvS6zwlIoXpLwMAHweaSSX2uIgESH1JuDejHndGXfeq6uhTfILAz0VkHqm96LFPicruQVW1gN0iMumnOomID/g7Ul9EToszvefuMUnQMAzj7SmdyLMl813AhIeu034JfAz4MPDpLDET7eKOHtLeSWovvgo4TOo0c7+qrpxgHYBQxvR3gKdV9UYRqQWeOUE/M0VP0rds6oDZwOheexXwmoicp6rtU6hn0sw5d8MwDGM6PAW4ReSzozNEZLWIXJou3g18CUBVd6XnDQE5GXU8R+oLACJyGdCtqoPpZVuBzwEbRKQiPf+wiNycjhcRWZGlb0GgJT19a8b88e2fEqq6Q1VLVLVWVWtJHXE4Z7oSO5jkbhiGYUwDVVXgRuCq9K1wu4B1QGt6eQewB/hZxmpPA4tHL2hLx9eLyHbgn0mdl89s4wXgK8DD6VvLPgZ8RkReJ3Xk4Hom9n3gn0TkRcB+gvanTER+A7wMLBCRZhH5zJup560604flDcMwjBlKVVuBD020LH0eeh7wm4z4XmD1uNA3JGhVXZcx/RjwWLrYzQS3oanqrePKL5O6wG7U10/Q/t1Z6phw1CVV/chE88fF1J4s5q0ye+6GYRjGaZW+Qn0v8CNVHTjT/ZmJzJ67YRiGcVqp6hNA9vGvjbfM7LkbhmEYxgwjqWseJq++vn7dqbp97VTWdTb6Pz94RmcVOjnUGcPrEuIJJeCxEYoqy6p99IfiNPXEsQm8b3UJy+tKKcj1cvtdG1m7MMjBtjA5XjvNvTGSlhJPKN/55Cqef72Rp3f2Ylnw1ZuX8P17d1EQsNPWn+DmtWU88Eo7X715MUfbB3jq9Q7a+hPMKnDS1h8nnoSiHDvhmMXHL6/hlT2dHOqMMByxuOMvzuMnf9jJx65aSCSa4F/v28l3bq3n9rtexVIIeGzk++wUB13sagoTTyrf/sQK7nhgFwUBB5GYRdJSOgcTRBPKl2+YT3WOELZ7ufuxvaxdXMy2hh4uXFrG/qZ+RKCtN8L719YSDHj44e+2c8nSYo50DFNV5GPjvl4uXlJEWWGAXYd72Hp4gAsWFGC3CaFIgotXVLFxdytzK/No6hyio2+ES1dWsXF3Gxctr+JAUw8VRTl4XA42vHSYOWUB8nLc9A5GmD+rgIHhCACPv9ZGdYmXZbML2d/UT47PSVmBH7/XSXtPiNcP9eJx2YnFk8yvCtI9MILHZaepa4RowqKq0MP8WfnYRGjvDTE8EmdWSQ4Br5OO3hCLaov51ZP7WTE7D6/bQSSWxOmw0dQ1zHsvqKOxvZ/BUIx4Ikln/whHOsNcU19FImlhswl7G/twOWzUluXS3hvC6bDh96Ruy31yWwcel40rVpTh97ro7AvR3hsmYSnLZheyt7EPW/qe3/dcUMfDLzXg9zho74tgS9/Is2x2AQOhGN0DI3jdDroGoty0yIcnvwAA9eXQ3R8iJ9RNIhwmp7qGzq2v4V+1hoDXRfdAmNiWFyleuZKBQ4fwFhcTbm8nHgohK86nMuhGVRlKCC6nnb6NL2Bfuhqnw0Zj+wD5OR6GwjHKCgMU5HqPvX96BkYo9AgdoSTSsIOS885Hkgm6d2wn2teHbfkaygv9JBUirc1Y8TgA3urZOGxC85OPU/muq7DCIew+PwCRznbILyZycB/3H7Zo6o5y6dIiXtnbQ1m+m/a+KH5P6j0XiVksqPAypyzAw1u6sAnEk+C0g90mJC1FRAj6bHQPJXHYIGFBvt9ONGERi6eWx5OKz516sZNJyPHaKMpxcrQ7SjSuWAqOcbtZloLHJYQiitMOIoKqUhp00jmYIGkpTruk1rVDIgkluQ46BxM47OC0C+GoRcKCa88pZF/LMIc7o9htgk3AbgeHTXA5hKDPweHOKHWlbpp7Y1gWxJOK3SZ4XcJA2OKC+Tm82jCEpeB2pLfFUn7815dP/9NYjDfF7LlPs7b++JhyLHn8y1RTz9hl3/rV9jHl3lDi2HQiOWYRs4uzP/vnrz+wcEz54oXBLJHQn9EGwKHOyLHpr35o2ZhlVQWuMWW77fj7ujBn7LITqSoZ25/2nuEx5c7+47eShqNj+1ec5x1T9riOX+i6pLZgzLKdR/qz9uFo+9jTfAtn5R+bHh4Z+3epnz92fAcr4/twbYlvzLKS/LHlpXUlx6brqvLHLAv6xr5mB1oHj003tIztu8Oe/TP02vOqxpRry3KzxlYW+ceUXc7sHwFDTWMff+30H1+37MKLxyzLv+jyMeXC8y44Nt383LNjlpWsOmdMubTw+J1HOw91jlkmzrHPFolnXNjsOLp3zLKRvDKySYaGxpTz5mUdu4SG9siY8qYDfcem7Sf5xBxNfKnYsX+z+Nh/5THv6fi497fPPbahzKoc9rGxJ3rYWa53bD3xZPadud7hsR3MbGfzwbGvn9tpcvrZziR3wzAMw5hhTHI3DMMwjBnGJHfDMAzDmGFMcjcMwzCMGcYkd8MwDONtRUTmi8gjItIgIntE5H9PNEqbiFwmImsnUe/d2cZpfyvGty8inxeRT5zqdjKZh9gYhmEYbxsi4gEeBr6sqr9Pz7scKAY6sqx2GTAMvDSN/XKoaiLL4jHtq+pd09WPUWbP3TAMwzjlRKRWRPaKyM9FZLuI/C79PHlE5BsisklEdorIekmPgyoiz4jI90TkVRHZLyIXT1D1R4GXRxM7gKo+rao7ReR5ETk25KuIvCgiy4HPA7enB4S5WERqROTJdL+eFJE3PC1PRL6T3pO3ici5IvKsiGwRkcdEpDyjv98VkWeBvxKR94vIRhHZKiJPiEhpekjZ8e2vE5GvTGGbp8wkd8MwDGO6LADWq+pyYBD4i/T8O1V1taouBbzA+zLWcajqeaSGg/3mBHUuBbZkae8npIdwFZH5gFtVtwN3AT9U1ZWq+jxwJ/CLdL9+BdyRWYmIfB8oAT5FatS4HwEfVNVzgZ8C/5gRnqeql6rqvwIvAOer6irgt8BXVfXIBO2Pd7JtnrIZd1i+vr7+a0D2J7ycJjP5yXuGYRgAInIbcFvGrPWquj6j3KSqL6an/wf4S+BfgMtF5KuADyggNTzr6J74/enfW4DaKXbpXuDrIvJ/gU+THtFtAhcAH0hP/5LUELCjvg5sVNXbAERkAakvFI+nDzDYgbaM+HsypquAe9J79i7g8CT7/Va2eUIzLrkDHpNYDcMwpl86ka8/Ucj4cvqc+Y+BelVtEpF1jN0hG31EZZKJc9Qu4NIs/QmLyOOkhon9EFB/0o14Yz83AeeKSEF6CFgBdqnqBROvSihj+kfAD1R1g4hcRmo8+sk42TZPmTksbxiGYUyXahEZTYofIXXYejSRd4tIAJjq1em/BtaKyHtHZ4jINSIy+rzsn5A6zL4pnZwBhoCcjDpeAj6cnv5Yul+jHgX+GXhYRHKAfUDx6HaIiFNElmTpWxBoSU9/MmP++PannUnuhmEYxnTZA3xSRLaTOvz+n6raD/w3sAN4kNSe8qSp6gipc/RfFJEDIrKb1Hn2zvTyLaTO7/8sY7XfAzeOXtBG6vTAp9L9+jjwV+PauDfdxw2kDsN/EPieiLwObAOy3Va3DrhXRJ4Huk/Q/rSbiYflDcMwjLODpaqfHz9TVf8e+PsJ5l+WMd1NlvPPqroXuGaiZSJSQWrH9U8Z8fuB5eNCr5ig3lszpn9K6uI5SCX0S07U33T5IeChCeLGt/98xrLLMqazbvNUmT13wzAMY0ZIPxhmI/B3qmqd6f6cSWbP3TAMwzjl0reALT3Nbf4C+MXpbPNsZfbcDcMwDGOGMcn9NHPZZdKxBf6z+8BK0hp/l8upUZLnnnRsJJaclj6ciG3yf0J2HuycdOy8itw30ZvTJx4KnTworefVlycd29EzNOlYJ6f/7716Xv6kY6OJyb8nHPbJ92Ga3mrGDGaS+zTqCSVp6U/yeptFf9iiO2RRV+YlllCqinwsr/HRF7YYjlqsmFtK26DFSPNRDnRZdA/EaOyJsaQmyIuHE2xuThJPQmgkxryqfOw2oX/E4khbP5bC9uYYvWGLSCzJvsZeBoYj7G8eoKrQw77WEMVBF4d7LCxVKgvcxBOKqtLaF8NSpW3Qom8wQvtAko6eYZwOG/uO9hCOxNnZbhGOKbk+Jx6XDbtNiMSVUEzZ3tDJcDRJcdCD3S60DyTY25FkXpmbnYe7sdntRONJbAIuh51cvxO/18n5SypYXFvItWuqKSsMAHDzJTWsnFfKstkFNHeH+eDFNRTn+SgvyqGuMsifv2c+K+aVsXb5LK46bzbO7lauWVNHcb6fORV5tPeG8LgcXLJiFm6nnWvPr8HrdlJkj1M/vwhLlfqFFeT4XMypyCNpKWWFOVyzupL8HA9zKgtYNqeI6y+YTVVJLkV5flYvrmBOZT6ffPdi6ipyqCgK8O41c1gyu4h3nVPJmsUVfOD8GpwOG36vk2TSYtW8EsoLA7icdi5eVkk0nuDa82bR1RfG6bAT8DopLwyQtCwCHifzCpwA5OV4WFhdwEVLSoglLHL9bv7n6aNUFvmZV5XPnqN9FAW9BP1uygsD5Od4WLOggIHhKPGkxbyKPOoXVeJxOZhVHGDXkV7mlOeybE4RN6ydjfZ143E5WF5XwkevXMQN8z28Z80ccnxuaspyKS/0c8W5tVxxThX2slkMFc4iGYvS3R8iEktgFVWgVXNoD1s8u/UoTR0DWKrEE0m6+oZJBPKx6paSUzuH9qI6ButWUpIfYGegjlcburHZhEMtvdj8Odzx4G7yA262H+rB53HxuxcaufOh3Ty5tZWu/jChSJyfPbqXxzYf5Z5nGiisXwPA4NGjtOVWkVNTS7J2EXsbe4jFLZxeH/c8fYBA7Ry27G1lIBRlYM5KfvrITmIODz1xO8RjhDxBDjb38tKedkrzPDT1xnl4Swc7WqK82jDM/o4YAPGE0j5kkbSUfS1DtA1atA1aRBNKy4BFjtfGvk6LCxbkEopaVOY7yfPbWVzppXMwQVHAid0OhQH7sYSftJR4UtnREqcs30M0riyd5cPrEgYiFiNxpanfoqnfIui143WmPp6b+i38bsHlFHpDCYYjFjaBLc1JmvuTJJMQiVvEk8rAiMXgiIWlqS8aFfkuWntGUFV8bhsdQxZBn52+kMWRngSRuBKJWQxGlGhCWVWbw3DUomPIIpGEg10JtrenPrsO91gc7LZo6EqSsNR84TjLmeRuGIZhGDPMtB/3PcnjYGunu33DMAzDeKc5HSd1sz4Otr6+fsL5hmEYhmG8eeawvGEYhmHMMGfN5dincDS32lNQh2EYhvEWiUiS1GNmR/1WVf95grj5wL8B84F4ep0vqmrHJNu5G/iDqv5ORI6QGpSm+8RrZa2rArhDVaf6zPuzylmT3DlFo7mZQ/2GYRhnjRFVXXmigPQocQ8DX1bV36fnXQ4UA5NK7qeKiDhUtZWpD2Zz1jGH5Q3DMIwz6aPAy6OJHUBVn1bVnSJSKyLPi8hr6Z+1AJJyp4jsFpGHgZJxdX4xHb9DRBam1/GLyE9FZJOIbBWR69PzbxWRe0Xk98Cf0m3uzFh2v4g8mh6k5vu8TZxNe+6GYRjGzOIVkW0Z5X9S1XvGxSwFtmRZvxO4SlUjIjIP+A2pMdpvBBYAy4BSYDfHB3kB6FbVc0TkL4CvAH8O/B3wlKp+WkTygFdF5Il0/AXAclXtFZHacX1YCawiNeb6PhH5kao2TXL7z5h3XHI/hef2T+hUnGIwDMM4m4nIbcBtGbPWq+r6jPJJD8ufhBO4U0RWAklS5+QhNULbb1Q1CbSKyFPj1rs//XsL8IH09NXAdSLylXTZA1Snpx/PGPt9vCdVdQAgPbxsDWCS+1nolJzbNwzDeKdLJ/L1Jw3MICJrgP9KF78B7AIuzRJ+O6nz7itInUaOZDZ/gmai6d9Jjuc5AW5S1X0T9OdEz1aOZkxn1ndWM+fcDcMwjNNGVTeq6sr0zwbg18BaEXnvaIyIXCMiy4Ag0JYevvXjwOgT+Z8DPiwidhEpBy6fRNOPkToXL+k2Vp3CzTrrmORuGIZhTBeviGzL+HnDbXCqOgK8j1TiPZA+9H0rqfPtPwY+KSKvkDokP7qH/QBwgNQtc/8JPDuJvnyH1GH+7ekL5r7z1jbt7Pa2OLxgGIZhvP2o6qTGvlPVvcA1EyzqAJZnlP8mHa/AF7LUVZsxvRm4LD09Anxugvi7gbszykdIj0M/wbL3nXRjzhJmz90wDMMwZhiT3A3DMAxjhjHJ3TAMwzBmGJPcDcMwDGOGmYnJPVJfX78u2w+ncWAZt0MIeoQSPxQG7PhdQt9wnNJg6jrGvuE4Jbk2QjHF7XJQV+zAVVHN4nI78ytzyPHaeHxbF4tLhMUlgscplBf62bS3E0uVHLfwyt4u4knFbgOPQ9jc0MeRjmE6+8J0DcQIRZIc6Y7jsNvI8ZBqy2mjO2SxfG4ZlkIsrsQS4Pe68LuEHL+beMKioihAR+8wxX7wOoVQJEE8oXicdopzU9sTiyc52JWkdzjKliMjVOY7OafaSSRmAWB3uWnuHKSy0MtINMHciiDxhEVoJMY9zx4mEksAMByO8fz2Npo7B7HbhKDPSd9QhF1Hemlo6iHod/PSzlb6hyKERmI47DbanXnEkxa5tiS15Xn0h2IUeGzY7ULQJWzZ20EBEcjJ40DzAG6nneauQVbNLyfH7+ac2QUUOuIcbBmgOD9Ae88QSctCk0mCfhd+m0WktZl8n5NoIrU9cwrcbHihAVXo7Aux92g3iWiEQ60DOB12+oej7DnaA8DAcJS+cJwntjRRkOsl6HcTTyTp6h9hdnmQaNxib2M3jtw82nvD7G3sY29jL26nnYpCH6/t7+AL1y3gcNsQvYMRhkbiVBbnUhj00js4gt/rIhpP4nE5iCcs7nv+AG3dg7T3hth1pI/FNQXsa+4nL8fDb59poM/mw+2yIyKEIzG63QV09A7z4EtHAGjqGiYWT+K027EN9TKrNEjcn8fOQ12UFATI9bnweVx4XA5EBL/XyXPbjlJemIPdbiORSFLoiBNqaaK6LEhNeT53bdjOZefUpOoN9ROLJxHgfavLSYbD1JYGiCeSvH9NJbdcWsNly8uIxhMEPA78Hjt5ATfLaoMkerro6g+zL+alsWOQdmcepQV+CnJ9uJ12joaFc+YWEG46iipsePEgS2oLuWR5Ba54mKKgjxG14xno5LUDXaxeVImlkLSgqsBNXbGTulIXXqdwsDv1t7YJnLegiIKAk/JcG+EYxJNgt0FJrovqfGF/S4jSXCc7WqI4bMJLDamLuY92x8j3Odh4OIalMDRi0Tmcqtfvgn0tw9TX5dA5EGNnaxJLU58RQY8QiUNvKEHPcJLSoIOhKDT3JQl67cQTit0mDEaUc6vsDERgMGJRGnQQ9NkpyrHROay83pzAbhN2NY9QUeilutjH0IiVes8mLLpDypIKF0lLSVqK2wEep7C/LUTQZyNhgdspnFPtxm2HoQiU5QpVeUKhX2jtT1KUY67HPpvNuL/O5s2b33CrRaYzPbBMY3eMgsDkXvaeYYvS3EldbHpGlORO/t/n9SODVBT6J1dvvp/DbYOTirWlblk9rT5w6XzauocmFZvvd0263vecP4fHNh6aVOyR9iGK8yf3ep4qoZYmHKWVk4rdur+dNUuqJhXb0NyL3zu51ynod9PZF55U7KnS2hs9eVBaXbGT9oHEpGI3HhiiPG9y76GA+/T/nxtvbzNxz90wDMMw3tFMcjcMwzCMGcYkd8MwDMOYYUxyNwzDMKaFiAyPK98qIndOsY6rRGRLemz2LSJyxant5cw04y6oMwzDMGYGEXEA3cD7VbVVRJaSGgBmcld2voOZ5G4YhmGcdiJSDNzF8THVv6SqL4rIOqCC1G3L3ar60YzVdgEeEXGr6uRvY3gHMsndMAzDmC5eEdmWUS4ANqSn/x34oaq+ICLVpPbIF6WXnQtclB7sJdNNwFaT2E/unZjcI6fjXvfNmzdPexuGYRhnkojcBtyWMWu9qq7PKI+o6sqM+FuBUEwwAAAgAElEQVSB+nTxSmCxHH9WRa6I5KSnN4xP7CKyBPgecPWp24KZ6x2X3E/2kBvDMAxjctKJfP1JAydmAy6YIInD8XHbR+dVkRrD/ROqevBNtveOYq6WNwzDMM6EP5ExJruIrJwoSETygIeBv1HVF09T3972THI3DMMwzoS/BOpFZLuI7AY+nyXuC8Bc4Osisi39U3Laevk29Y47LG8YhmGcHqoaGFe+G7g7Pd0N3DLBOuvGlf8B+Ifp6uNMdaaTe+bFbbVnsB+GYRiGMWOc0eSeeXHbmR6tzTAMwzBmCnPO3TAMwzBmGJPcDcMwDGOGsa9bt25KK6xfv/6y22677ZlTHT/Vet8OjjZ1rCvOdfBnl1bR0RtmWXWAgXCcBRV+Xm0YYPW8fGLxBLG4RWN7H3k+B+cvLuPxrS209UW4cEGQfa0jeJw28nw2LlyYTzJpcf8r7cwv9+GwwycX2WkYhNVzclheE2BoJE5Tb4ziHCd5fgevHw3xycur2HaonxU1Pgr9duJJZXGVDyuZAE3w/vOrqMoT2rqHSFpJBobCvLq3kwdeaccpSRxiMafEQzRhcagzRlmei7b+1O+BcJxFFW5ae6JcubwAUHI8dnpDCS5ZVsZrB7tZW5fH4a4QXQMRNu7vo3dgmIrCAI9v62LHkT7C4Qgv7OxkJGaxr3mA/c39lBd4ePS1TvxuOxcsqeDVPW3sbhrmlb2d9PSHeOTVJg62DvLb5xqpKffz7V/v5OpVZTz0ShMbNrbQ1DXIi/sGqa7I4ed/OkDPcIxwJMET2zp4dnsbHT1DPLerk+b+CD2DMXoHQ0SiCR58uYUHNrbR0tHHuYvK+YcHDtDZP8yGlxvpGYyys2mA3Ue62bCxnXg8QTSe5IFNHexuDnPh4mIef62FwXACp12JJy1+8/RhioIuNu3ror03TDxhsf3IIPULSnlsUzMBj51XdrfROxyjvMDLi7u66B+OUpznJZaweHFXB92DcTr6Rzh/UQk7D3ezuLaI1w50UlUU4BdPN3Hx0mKe2tbG1aur+cPLR/F57AyPJFhQnc+jW9rpGwhxdcEwe4fsvLy7k9ULS/nVE/t4dV83+5oGuP2Dq3jqtUZ2NQ3T1TvEho2tXL12Hs2dAxTneqhwRNh0ZICCsmJyXDb+tLmRNUsq+PWTB1hUnU9bzxDNX/8ym4JLKDy6jUTtIjwuO8OPPUh1/Sr67/sf7Ls2sdFVy4q5pfQNRTjSPoDX56GjN8ySZAcvt1vUluby4s52Ll9RxSMbD3PpyioW1hTR1DGIw++j2BYjjJ3yogCxxx4gb9lKNu1ppTjPR+89P2OLllEb62DOisXMtQ+x5Z++S+3yhewftlOe58FhJbDlFbKs1MPe1kEWVBfikhg2gb98Tx2VZUG8DotrVhVSGHAwPBLn3fVV9A+NMLfcR12Jk1yvDbcdBKgpdmOzCTabcNXyQrwuGyVBN/Vzg+xrDTESs/j2RxdCMkpRjpP5ZW6qizzMKnRhtwkrZ+cwFE5w5dJc2vui1NflkOezUZZr4+pVJVy+opSDrUM4xMJhgwUVPopyHIRjSSyFklwHq6o9XL6sEFWLSNzCLiAoRX7B77Fx+ZICOvoi7GwK8YkrqllY4eHGi+awZV8nc0s9vKe+ArdD8DihKNdFPKF8/c/OZdOeNq49t4RXGwZZVe3mvDo/Qa+Ndy0vIjQS5Ws3L8bzn99g3oc+9K0z/TlrTMzsuZ9mK2fnTjr29SND09iTt87vsU86tqM/NunYzoHJx9YWuSYde6p0DcQnHXtNfcWkY/+0uWnSsTUlgZMHnWKeoe5JxzZ3Tf5/1+HzTTp2X2PvpGNPlY6ByT/ptL0/MunYkWhy0rFLZ03+NTIMMMndMAzDMGYck9wNwzAMY4Yxyd0wDMMwZhiT3A3DMAxjhjHJ3TAMw5gWIlImIr8VkYMisltEHhGR+SeIXyki106i3nUi8pVT29s3ti8i14nI16ZYx+0isktEdorIb0TEk7HsIyLydyKyUEReFpHo+O040fpTYZK7YRiGccpJauzWB4BnVLVOVRcDfwuUnmC1lcBJk/tb7NeJnsw6pn1V3aCqkx4mXEQqSQ+Io6pLATvw4YyQa4BHgd503L9Mcf1JM8ndMAzDmA6XA3FVvWt0hqpuU9XnReSXInL96HwR+ZWIXAd8G7glPfLbLSJSICIPpkeOe0VElo9vREQ+KyJ/FBGviNSJyKMiskVEnheRhemYu0XkByLyNPA9ETlPRF4Ska3p3wtExDVB+7eKyJ0ZddyRjj8kIh/Mst0OwJv+EuEDWtPrC6kvD6+paqeqbgImuq92wvWnyiR3wzAMYzosBbZkWfYT4FMAIhIE1gKPAN8A7lHVlap6D/AtYKuqLie11/+LzEpE5AvA+4EbVHUEWA98UVXPBb4C/DgjfD5wpar+NbAXuERVV6Xb/K6qxiZof7xy4CLgfcAb9uhVtYXU3ngj0AYMqOqf0otXAa+rqmZ5TU62/pSc6VHhTqq+vv5rwJs653Ambd68ed2Z7oNhGMZ0EpHbgNsyZq1X1fUnW09VnxWR/0iPy/4B4D5VTaR2bse4CLgpvc5TIlKY/jIA8HGgmVRij4tIgNSXhHsz6nFn1HWvqo4+OSgI/FxE5gEKOCe5yQ+qqgXsFpE3nF4QkXzgemA20J/uy5+p6v+QOiT/xxNVfpL1p+SsT+6AxyRKwzCMs086kWdL5ruAbIeuAX4JfIzUOeVPZ4l5Q7YnlYwBdpI6zF0FHCZ1JLpfVVdmqSuUMf0d4GlVvVFEaoFnTtDPTJmPK5yob1cCh1W1C0BE7if1heN/gKtJf1E5gROtPyXmsLxhGIYxHZ4C3CLy2dEZIrJaRC5NF+8GvgSgqrvS84aAnIw6niP1BQARuQzoVtXB9LKtwOeADSJSkZ5/WERuTseLiKzI0rcg0JKevjVj/vj2p6oROF9EfOlz7O8C9qSPNjhUtefNrP9mOmKSu2EYhnHKpc8t3whclb4VbhewjvQFYqraQSpx/SxjtaeBxaMXtKXj60VkO6lz3J8c18YLpM6tPywiRaS+CHxGRF4ndeTgeib2feCfRORFUlekZ2t/qtu8Efgd8Bqwg1SOXQ9cBTwxGpe+RbAZ+DLw9yLSLCK5J1h/yt4Oh+UNwzCMtyFVbQU+NNEyEfEB84DfZMT3AqvHhb4hQavquozpx4DH0sVuUue2x8ffOq78MqkL7EZ9/QTt352ljglHb1LVbwLfzJwnIteQuohwNKad1OmESa3/Zpg9d8MwDOO0EpErSV2x/iNVHTjT/ZluqvrnqvrK6WzT7LkbhmEYp5WqPgFUn+l+zGRmz90wDMMwZhiT3KeRx2XjizcsY39TP93DCV7aP0g4mqS1N8KKmhyi8SS9wwlWzQ5w8dJSqkt8NHYM0NyXxOu0MbfMy9UrCrAJrF2Yj8Mu9A6OcMXSPDoHorT1x/jvHQmKgy4OdYTJC7iZWx6gpsjN6kVl3HTpApbX+HnglVZeOhhj+9Ew25siPLsvzBM7BxgKx8j1OensC/Pc7n4uWjELVeX6i+dz40Vz+MxVNZQX+mgfSNA1GONgR5TVdQF6hmIMjVhsOxIiaSktPVF2tkToGYxSHHQTTypNvQl+/cxRtjT088vnG+keStDaG6Usz0VBwM3zO1q5eGEQh0248ZJ5zC33EU9aFAddiAgPvdrF6rlBHHZh8942QpEEi2cFmFvmY05ZgPevqeRz9QHOmZuPx+Xgs1fXcMfDR6gt9bOkys/CWUE+ekkF2w72sKDST9dgArsNVs8NUlPkIT/gwu+2UxBwk7SU7YcHWDy7mIKAg4KAnSW1+exv6uW9q8vZfmSI96+pJOBxsGZhMfGkxcpaP7PL/OT6nNx+01IuWZzH6w2drJyTTziapKU7TDxhsetIL3l+F7OKfaxdUsbGA/209MXZ8OIB5lUEKMn3cdNlC2jqiTIcSTAUSdLYHeFg6yC9QzGuX1vLRy6fw+uNEQZCUY50htm8t42u/gh3/fEAS6q83Pf8EVbPK+CRV44iAnUVQboG4zz00hFW1Obg9zj45ssWxXleKgo8/PSPu1m7uITSoIurziln054W1iwqI5FULlpWQThqsf1fv0dejpeG392Do7ic4ZEY7twgTU8+wQ3n1xBPWMwtD1BeGKA4z8f8D9/CtUUhaq56N66WBrz/+2MGDx6ivTdE/4ED9L7rw5QV+Gho7iXy7B85PyfMvsZeLFWe7HJRUeijqjTI2iVlPL7lKHkBN+UBOy+83kh1aS7J5x9lR1eMRTWFVOW6CM6bh9OeuhOpdzBM4S2f4iOLPQRnz2HL3laGGo/i8HrZFvKysq6E4bgiThd/evUQD2xpY9msPH73bAM3XDyf2tIAhzY8hPfIbq6Z6+el3d1UFPopDToJR+IU57r4w2u9PLlrkCXVuQxHLDoHE2w7GqY06GZBZQAR6BmKUV7g4Zmd3VyyOEhp0Mk//u9eDneO0DUY48KlZcQSFs09UfweO7l+J2X5bjYfHKS2yIXP7SCeVF5oiNLZP8LvX27isuUlzClxket30tEfA8DnsrGyxkfSgqGRBOFoAq/LjsdpY06pj9veXct7zinC7RBe3NdHWYGX0qCTPUf7eHZnNw3NvdSVuukdTvD8zg76QzEGwgkCXgefumYBL+9owmaDl/b0UFfq5tn9EXxuO4/vCtPRN0JxrovP/ngHvpLiM/nxapyESe7T7K7f7xpTXlyTlzV204H+MeWdTaEskdA1ONFTC1POnVc4phxwZ/8zD4bH1pPrzX6mpqrANaa8avaE15MAcOXS4Jiy1zXRLaEpL+9oHlOeV+7LGutx2ceUn9vVfWz6g+eP/bDZ8Gp71noiseSYssuZ/TW6amXJmPLiWdnvlMkLuMeUL15ROek+JJJZAoGbLxi7bYW5x9ux28a+tg579m0Z378TxXoiQ2PKhwcS2eu99Kox5cZ3Z7ttGWz2sX/DeZXBLJFQWzH2/XK4Lfvp2dy6eVmXed1j/6/jtuzPLHE6xr6ec8pzs7fpH1uPw5b9//wnjx0aU97dOJQlEpq6R8aUD7aFs8buOjqYdVlJztj+uR3Z/96bGsbW8+5zy7PGfvPmmqzLjLODSe6GYRiGMcOY5G4YhmEYM8ybuVo+Ul9fv24K8bVvog3DMAzDMN6kKSf3zZs3T3psW4ApfBHI9qWhdirtGYZhGMY73Vlzn3u2Lw1TPEpgGIZhnCVEpAz4N1JPfYsCR4Avqer+LPErgQpVfeQk9a4DhlX1X05xf8e0nx5jfrGqTnqnVkR+SmpI2E5VXZoxfwVwFxAg9Tp8LOM5+aecOeduGIZhnHLpgU8eAJ5R1TpVXUxqTPY3DJWaYSVw7TT360Q7tWPaV9UNU0nsaXczwSNwST1+9muquozU6/J/p1jvlJjkbhiGYUyHy4G4qt41OkNVt6nq8yLySxE59sx4EflVei/528AtowO3iEiBiDwoIttF5BURWT6+ERH5rIj8UUS8IlInIo+KyBYReV5EFqZj7haRH4jI08D3ROQ8EXlJRLamfy8QEdcE7d8qIndm1HFHOv6QiEw4nK2qPgf0TrBoAalR7gAe5+TDv74lJrkbhmEY02EpsCXLsp8AnwJID4e6FngE+AZwj6quVNV7gG8BW1V1Oam9/l9kViIiXwDeD9ygqiOkRlD7oqqeS2q0uB9nhM8HrlTVvyb1XPtLVHVVus3vqmpsgvbHKwcuInXYfap79DuB69LTNwOzprj+lJw159wno76+/muA50z3YzI2b9687kz3wTAMYzqJyG3AbRmz1qvqSYcoVdVnReQ/RKQE+ABwn6omUkfyx7iI9B6uqj4lIoXpLwMAHweaSSX2uIgESH1JuDejnsynNt2rqqOPigoCPxeReYAC2Z9qNNaDqmoBu0XkRKcXJvJp4A4R+QawAYhNcf0peVsld8BjkqZhGMbZIZ3IsyXzXcCEh67Tfklq/PUPk0p8E5nokX+a/r2T1DnyKuAwqSPR/aq6MktdmY/8/A7wtKreKCK1wDMn6Gem6En6lpWq7gWuBhCR+cB7p7L+VJnD8oZhGMZ0eApwi8hnR2eIyGoRuTRdvBv4EoCqjj6newjIfL7zc6S+ACAilwHdGVeYbwU+B2wQkYr0/MMicnM6XtJXqE8kCLSkp2/NmD++/VMmfZQCEbEBf0/qyvlpY5K7YRiGccqpqgI3AleJyEER2QWsA1rTyzuAPcDPMlZ7Glg8ekFbOr5eRLaTOsf9yXFtvEDq3PrDIlJE6ovAZ0TkdVJHDq5nYt8H/klEXgQyBzsY3/6UichvgJeBBSLSLCKfSS/6iIjsJ3W+v3Xcdp9yb7fD8oZhGMbbhKq2Ah+aaJmI+IB5wG8y4ntJ3ROf6Q0JWlXXZUw/BjyWLnYzwW1oqnrruPLLpC6wG/X1E7R/d5Y6Jhw5S1U/kmX+vwP/PtGy6WD23A3DMIzTSkSuJLUH+yNVzT7Un/GmmT13wzAM47RS1SeA6jPdj5nM7LkbhmEYxgwjqWsepk99ff26t3L7Wub6b7Wu0+3H/7tRF9fkseNwPyKQSCrRhIWqkh9w0jccx24TLIXCHBdl+R46+yP0Dse5dFkpL+3uxOOy0TccP1bn8to8bDZh26E+yvM92EToHY5htwmhaOoWTp/LzuwyPwD7W4aIJSxyvQ4GRxIAeF12RmJJllQHaekJ0x+KU1vi40hnmNoSH7k+F4Ph2LFya2+EeFJxO2yIQJ7fSVtfBIDzFxbxyt5u/G4HqxcUMRCKsuvoALGExQcums3WA51EYkmSllKQ48Jus9HUFSJppeoB8LjsOB3C3uYhAJbWBOkeiNLeH8HjtFNZ6OVoV6ovhztCFOe6yQs42ds8RHm+hxyfk8bOMNGERdDnoD8UpzzfQ38ozkgseWx7RYSLl5aw+2g/kGr/cEcIv9vOstkFWKo4HTZ2HenDSr8vKgt97G0ewibgc9vJD7gYGkmQSFrUzy/mia2tFOa4yPE68LodNHWFGY4k8DjtOOxC0OekYyBKjtdO33CcwpzU+rGERVmeh4DXQUPbMJUFXpwO4WjXCHl+B33DcSoLvAD0h+L4PXZyfU4Gw3E6B6Lkeh3MKc/hUNsQkbiFx2kjHLNYWJVDY2fqjp+r6mfx2KamdN8dWKp0DkTxOO1E4kmqirw0d4+wsCoHj8tB71CUZFJxOoT3XTiP1q5BRn7/Wwo++HEKgz5irY20PPMMR//0BK6An5pv/jMtXUMssvXz7Bf/mosfeIiRjc9SvHIlf/zwJ6i+8mIW3fppjsRcLJ6VT/tAlPaeIQ5/7hNc9P/9I1q3hJJcN4+8eoSLls8i1+vgydcamV9dwKziHBRh6/52ytr3YsXjlJ9/PnaXmwfe/T4ArnvyMe5/dj+LawqwPXk/4YuvY55jGP+cuXT2htj7N7eT+OzfUvDYL+l998fxuOzUlzoI+/Lpuu9XPFO4hlnFXhraQly0pAS/18mTr7WSsJRE0gKgPN+D3+PgcEeIpJX6n3A5bMQSFgGPg0g89X4GSFpKnt/JcCRJImnhctiIJxVVxW4TkpbisNtQVSwFj9PGSCz1nnXYbSSSFoU5LsLR5LH1kpYe+3uNyix7XXY8LhuqEI4miSWsN3wO5QecRGLWsbb87tQB21A09X8K4PfYCUeT+Nx2eoZix+JEIBRN4nbYiMSTBDwOhiMJHHYbOV47n7vpvCndDmacPm+HPfdIfX39uvQAMrVnuC9TMhRJsuNw/7GyCMc+CPqG48feWADdg1F2Hj1+6mlfc/+YN36mbYf6js0ryfcQTx7/guZx2li9oOhYefybPfMhEbl+FwHv8TMzK+fk0zWQemMX5nq4YFHxmHUzYzM57DZC0cSxssthw+s6vm02EeLpD8sDrUPMKT9+p0nvcIxwxroAuxqPj6WQ+aHW0DZ87AN2b/MQ584tOFb2uGwsqc49FjuvKnhs+sIlpce+SAAkLevYBxjA4EiCF3d3AvDcjg5K8o4/J6lnMMr5C4+/npFYkr7h1LpPbG0l4Dn+mrR0h6ko9Ga0o7T0jgCpv/eS6uN9EhG6Bo/fMhuKJmjvS5WTSR3TX6/bTufA8diqIi+xRGq7L15eTl358et6dh4dYH56219v6Dr2+gA47TbyA8frjcQsltfmAdDeO8LQyPEvkYfu/m8Gf/uTY+WB4Qj+ytQDteZc916CX153vH/zFnPd7+87VvaUlFH9nz8HoE197Pv0x44tczntXHj/g8fKcbURT/+PxgcGmF9dcGzZ3p/8FzVlx1+zFw70gf346733ruMPHztw/x9YWJv6f913tJsntjQeW3bOV/8fteXH62lo7sXzntTt10PhBEtrji9bNbfgWGIHaOuLcLD9+O3Rme+f4UiCzJ0jj9POQPj4/3Lmey/z7wBgt8mY921mmyOxVDIdlfkeGF8eiSUZiVrH2svNeI9mvgcBcr2OY3//UDRBeX7G//lQbEx/UzsLqW2xyRv7kNlf4+x01p9zzxwtzowQZxiGYRgn93bYczcMwzAMYwpMcjcMwzCMGcYkd8MwDMOYYUxyNwzDMKaFiJSJyG/Tj5/dLSKPpAdNyRa/UkSunUS960TkK6e2t29sX0SuE5GvTWH9BelH147+DIrIl051PyfjrL+gzjAMw3j7kdStBQ8AP1fVD6fnrQRKgf1ZVlsJ1JMa2326+uVQ1USWxWPaV9UNpIZnnRRV3ZeuAxGxkxqc5oG31OE3yey5G4ZhGNPhciCuqsdGP1PVbar6vIj8UkSOPTNeRH4lItcB3wZuGR24RUQKRORBEdkuIq+IyPLxjYjIZ0XkjyLiFZE6EXlURLaIyPMisjAdc7eI/EBEnga+JyLnichLIrI1/XuBiLgmaP9WEbkzo4470vGHROREw9kCvAs4qKpH3+oL+WaYPXfDMAxjOiwFtmRZ9hPgduAhEQkCa0mN+PYNoF5VvwAgIj8CtqrqDSJyBfAL0nvG6eVfIDVG+g2qGhWR9cDnVfWAiKwBfgxckQ6fD1ypqkkRyQUuUdVE+jn331XVm0Tk/2/vzoPjqu5Ej39/6pZaS2uXtcuSZUu25T2RbewYbNYYwkDMYpxMFioBJ/OKyZCQV4+pmpl4JvNeMjAzjxkCD1wQIGQGhj1MbCAYG2PABrzbsrxKtqzNUmvtRb2f90e3QttIdkMsGTe/TxXFvef+7jmn/Yd+99zlnDPbv/2MfpcAS4BpREb0L5zl968iZlGc8XaxJXfvxfKt+8U0k55SSn0WIrIaWB1TtNYYs/Zc5xljNovIQ9E1zm8CXowm2jNDlwA3R8/ZKCL50YsBgG8DrUQSe0BE7EQuEp6PqccWU9fzxpjh2XiygadEpAYwQDLxecUYEwYOiEjRaEHRuwA3AH8dZ73n3UWV3GMntFFKKXVhRRP5aMm8ATjbreuniay/vgr43igxI01vOzy1334io/hyoJnIY+Z+Y8zcEc4BcMds/xzYZIxZISJVwNtn6WcsX8z22abevRbYGV2z/oLQZ+5KKaXGwkbAJiJ3DheIyHwRWRrdfRK4G8AY0xAtcwKZMXW8Q+QCABFZBjiMMcPzU+8CfgC8KiKl0fJmEbk1Gi8iMmeUvmUTedkN4PaY8jPb/6y+wQW8JQ+a3JVSSo0BE5l4fwVwdfRTuAZgDdAePX4KaASeiDltE1A3/EJbNL5eRPYCvyTyXD62jXeBnwLrRKSAyIXA90VkD5E7BzcysvuAX4jIe0DsJPxntv+piUg6cDXw0mc5/3y5qG7LK6WUungYY9qBlSMdiybBGmJGuMaYXmD+GaGfSNDGmDUx228Ab0R3HcDyEeJvP2N/K5EX7Ib97Vnaf3KUOuyMwBjjAfJHOjaedOSulFJqXEXfUD8IPGiMGThXvPr0dOSulFJqXBljNgATL3Q/EpmO3JVSSqkEo8ldKaWUSjCa3MdQnj0yL8LGRhcfHHMz6Aly62XVZNis3HpZNauqw8ybnEdRto0Ftfm81ehl0BPkld0e9h53csm0Au5/y8W+k0N0DQYREXYc7WPIH6bfHeTRzf3MnlxE92CA7kE/6SkWAiFDRVE2ydYk/MEws6tyeOwDH//6tpumLj/dgwFmVObw6l4vz7zTxvpd/UwpsbP8ksl4fEFae31UFGWSkZZMw/E+Fs8sw+0NMeAJ4vYG2X9yiFDY0NDmo8Xh50jbIDWldpbXlzMp6MAfCPPAZjfOoRCvbj3B4pll5NhTcHtDdPZ5SRLBNRQgFDY0nfLwjStrOdjmJs1mZXeLlwFPCEuSYE+zMr0iiwybla8urGZKSQbXzi9ndlUOHl+Q714zjazXf8sVKZ3MrSlk4oQMDrc5mVRs556bZzOzuojmbi997hCNJ/rodQZYODWfL1VkcqLbyzfLXNRV5XPlvFK2H/eSYbPypakl9LlD3PNCN9uOuLh0VinXTRTaHG6unFdKhs3KB0cGKcy2UZBl46+WT2bJzGKSLUkcaXdTU5bFYrsLjy9M16CfGZU5OJxBBj1B5k3OJT87lXcPOQkbyLBZ6HMHsVqEmZXZiIDVInzUPMShjiE8vhCv7eln9uR81u0eYEqJncNtLiYW2nnxwz6WzSmh4uA7rPj3o8yaPIG9LR6mlGSwaPoEGk70U5qfxuNvOzjc4WXWpDz+8Y0BVl05lbcanNjTrNxzyxxWzon8Wzo9QexpVvLsNjLTrVw9fxJp199G/+Ur2VF1GZv3nOR37x1j/YfN/NvgNN60zWDzvlM0tfdTU5GP0+PjV2800dHjpLd6Huu2HuXEKSfeP7udo219LHrhZVocbu5au5e3drUTCoX59RErrV2DPL5uPxv29rLrcCf/8loz+451I3u28sGBdg7ULOOx9Qe570A6qfVLWFBspaXbyZTHfov9nx5lc+ml3LS0loy0ZIb+5/288PZhsiZW8sSGZhb178H2o5+xYa+D9TtOcqS1j7CBXScIZ/cAABO3SURBVI4wP3u+hbZuJzMqs7ksuZOKwkzkpV+TtGU9kz2tTCrKYHKxncZ2H/fcOpfrFpRTkpeKyxvmL1fMoXswQFqKhdRkC/Nr8/nhDbOoKbVTkpdKjzNAe5+fn9xYR0Obj353kPzMFHpdQaaVZ1Jfk0eLw095fip/ddMcHM4gMyZmU5BlIz8zBa8/TEGWjcY2D7OqcrCnWhnwhHA4gwx4QnylrpBp5Zksm11Mi8PPyZ4A3kCIDQ1OvrawgqyMZG5YVIklScjPSmFBbT5LZ5fxvetmUFOWhdsb4palU3A4g8yqzudbV0+lMMfGsVN+SvJS+dY1dcytzqV7MMCkogySBKaVZ1FVmM6HzV6KclJpcfgpzknllqWjrv+iPgc0uY+zLXvb4459+I3WMezJn24oYM4dFNXQOhR3bF1FVtyx71MWd+z58tL2/rhj19xYFXfsV+dXxB2780h33LHny8K60rhj//P9nrhj77+zPv5O7H4v/tjzZMDlO3dQ1DeXVcUd++DLe+KOzUnX16PUp6PJXSmllEowmtyVUkqpBKPJXSmllEowmtyVUkqNCREpFpFno9PPHhCR9SIy6pt4IjJXRK6Lo941IvLT89vbT7YvIjeIyL2fso7jIrIvOoXt9vPdx3jpWxpKKaXOO4msu/oy8JQxZlW0bC5QBBwe5bS5QD2wfgz7ZTXGBONp3xjzKpF12z+ty40xjs/YxfNCR+5KKaXGwuVAwBjzyHCBMWa3MWaLiDwtIn+cM15E/kNEbgD+AbhteOEWEckTkVdEZK+IbBOR2Wc2IiJ3ishrIpImIpNF5HUR2SEiW0RkWjTmSRH5VxHZBPyTiCwQkfdFZFf0/1Oja7Cf2f7tIvKrmDr+PRrfJCJnW872gtORu1JKqbEwE9gxyrHHgB8DvxORbGAxkRXf/g6oN8bcBSAiDwK7jDFfF5ErgN8QGV0TPX4XcA3wdWOMT0TWAj80xhwRkYXAw8AV0fBa4CpjTEhEsoDLjDHB6Dz3/8cYc7OInNn+7Wf0uwRYAkwjMqJ/YYTfZoA/iIgBHo2ueT/uxjW519fX3wukjmebF8r27dvXXOg+KKXUWBKR1cDqmKK18SQzY8xmEXlIRAqBm4AXo4n2zNAlwM3RczaKSH70YgDg20ArkcQeEBE7kYuE52PqscXU9bwxJhTdzgaeEpEaIsk4Oc6f/IoxJgwcEJGiUWK+Yoxpj/62N0XkoDHmnTjrP2/Ge+SeqklPKaUSQzSRj5bMG4Cz3bp+msj666uA740S84lsTyQZA+wnMoovB5qJPGbuN8bMHeEcAHfM9s+BTcaYFSJSBbx9ln7Gip3RaKS+DS9zizGmS0ReBhYA457c9Zm7UkqpsbARsInIncMFIjJfRJZGd58E7gYwxjREy5xAZkwd7xC5AEBElgEOY8xg9Ngu4AfAqyJSGi1vFpFbo/EiInNG6Vs20Bbdvj2m/Mz2PxURyRCRzOFtIo8M9n/W+v4UmtyVUkqdd8YYA6wAro5+CtcArAGGR7angEbgiZjTNgF1wy+0RePrRWQv8Esiz+Vj23gX+CmwTkQKiFwIfF9E9hC5c3AjI7sP+IWIvAdYztL+p1UEvBtt/0NgnTHm9c9Qz59MX6hTSik1JqK3qFeOdExE0oEa4JmY+F5g/hmhn0jQxpg1MdtvAG9Edx3A8hHibz9jfyuRF+yG/e1Z2n9ylDrsI7TTBIx2t2Bc6chdKaXUuIq+oX4QeNAYM3Ch+5OIdOSulFJqXBljNgATL3Q/EpmO3JVSSqkEo8ldKaWUSjCa3MdQarIFty/ID64qYWqxjcYOP8sXVrFg2gSOtvWxL5RLdVkOc6cU8PK2TuZVWPnu8un8859XkZlq4Sd/VsVPr7CzqMbOFbPycQ6FmF+bx/yaXOoq7KyYm8bjrx0kPSWJI6cCbD3qxusPs3ZdI/uOD7DtUB+BUJgFpfDDxWnYrEJ7f4j3DjioyhX+ZdUUJmQm8dKHPdz7+A7qKvOZNymTDxq7SBJhxaVTcA35cfnCZKVZONrpJc9uYUeTi5JsC//vjln4AmHeP9hHTmYqr55MYsNeB9fWJFGen0p7n5+n3zxMjt2GzSpMn5jD4Y4hKgozOTXg51Cnn/ue20flhFQWz6qgtiiF5V8uZsuBHg62urhuQRUef4j3950kGDLYkq1kpqeQkWolLTWZY4tu4VCbi9KCTOomFdDS4yMQDPOfGw+TJiG8fsNVcybwpdoJhA3MmlyEb2CQiQWp2GbV8+aOVmZV5fM3t0yhrCCNky8+S3VRKncvzeB/+DdTlGcns2Yq82oK6RkY4minm3/4zlye+vF8XENBLDYbG3e3s2JGFhUFaWxpcLBlIAOXL8y9q+bx++2dWFNTOdoVIEmEV7a28Te3TaPXFaC+toAkEcoK7CyaWc7mRhcP/f4wJVkWUpOTGPAEWfujBby5s50J9iQKslM51uXDahG+t6wItzfECzKTBaXw6LrDLJiSSY7dxoeHHBgDTk+Qwkxh5ZJSMtKSue+mAgCuqMtkTnUe+5od+Pr7qSrOYs8JJ4tmltE94KUgO43uPjf+QIjUFAvfWFTO+l39AHj9QaoLU8jPSmHl0sk0tgwSCoVx9LsZ8oVwDwWYUV2IyxvkujmZHGodIN1m5aODHby7r51ffGsq/mCYdoeT8vxUbMkWAqEwj/73ft474GDV5dV09Xs5UVRHeWEWWw8PcN2CMqaXpeHx+ultbKT06mtxDwW4anYJV8wrZfOuExw80cusSfn0OP10DPpZMi2X/9tdzeKphdisQorVwsxJBUwqyWbxzDKWz0wj+PD/5rn3OmjMqKIoz46z5STe7m7aNm3keJcHi0VIsQjPvnWIfpePrn4fTd0B7n7kQ/7+u/VMLrGz76SHpg4nL759mIzUZLIzUlj/YStF2cls2NPO7IpUirJTmFiYwanBEEfaXXT2eki3JTGlLIc7Fk1gyfRcjp9yY7MmMel3D+INhDjeNUS6LYlN+xyEwgZf0GC1wF031uH71c/pHvCx/qN2vnPFRKwWEBG+Xp/L439oYt8JF/+9rQWPL8zOJidef4h2h5Nn3zrI+409tPb6ee7tIwAcaxtg6/42jrR//Pm3NRTgN5s7cfsMx7s8eANhnnmvi/ZeL7PKUgiFw1ROsHG4w81fPrJz/P+oqrhpch9nwQO74o595p2TY9iTP91vt5yIO7Z/yJw7KOrmr1TEHXvLpZVxx54v6bb4X1XZ1+4+d1BUbXFa3LGfnMhr7D3wSmPcsZsbXXHHFufazh0U9fi68f9kOOlT/GNf8+XSuGNfbRw8d1BUfmZK3LFKgSZ3pZRSKuFocldKKaUSjCZ3pZRSKsGMx3fu3vr6+jXR7apxaE8ppZT6Qhvz5L59+/ZfDm/HJHmllFJfACJSDDxAZFpXH3CcyIIxe4jMUpdKZMGWh4wxT0XP+XPgf0WrcAF/YYzZEz22HPg3InPCP2aM+WOOUR/TGeqUUkqNCYksrP4y8JQxZlW0bC6RBVaOGWPmRcuqgZdEJMkY8wSRJVyXGmP6RORaIsvKLhQRC/AQcDWRtdw/EpFXjTEHxv3Hfc7pM3ellFJj5XIgYIx5ZLjAGLMbOO073+iCKz8BfhTdf98Y0xc9vI3Imu0QWRv9qDGmyRjjB55l9JXfvtAu+Mi9vr7+XiK3ZRLK9u3b11zoPiil1FgSkdXA6piitcaYtTH7M4EdcVa3E5g2Qvn3gdei22WcfmHQCiyMs/4vlAue3IFUTYRKKXXxiSbytecMjM8nZgsSkcuJJPclo8UA8c+Q9QWit+WVUkqNlQbgy3HGzgP+OA2iiMwGHgNuNMb0RItbgdgpLMuB9vPQz4SjyV0ppdRY2QjYROTO4QIRmQ+cNm+0iFQB/ww8GN2fCLwEfNsYczgm9COgRkQmiUgKsAp4dSx/wMXq83BbXimlVAIyxhgRWQE8ICL3Al4+/hRusojs4uNP4R6MvikP8HdAPvBw5IV7gsaYemNMUETuAt4g8incr40xDeP6oy4SmtyVUkqNGWNMO7ByhEOjrpRkjLkDuGOUY+uB9eend4lLb8srpZRSCUaTu1JKKZVgNLkrpZRSCcayZs2acWts7dq1y1avXv32ucoSRcOxzjW+QJjD7S7uuG4a2bYQ3ZZMFs8qJ0mEPcd6aDnl5FiHk/opOUyvyGLdBy3UlGXz1t4e/vq2mWw90M2tS6dQPiELp9vDtkN9ZKVZOdDqQoC2vgApVkGAqSU2ZlVmk55iwRcIIyIYE8Y1FMSSBJmpFnIzkhj0hslKS6J6Yh4fHOqltjiFHleIuooMwmFDZVEmmekp7GvqprPHQ3P3EIEQ5GZY6HEFsSQJi6fmcrTDTfdggNx0K+09Tj465uayuhx8/hCleWmc6B6itiSDfpcP51CA4uxkmk556erzIALfXFpOOBTCHzD4/H5mVOaw7qMOOno9NHf5mVZu5/WdXQy4/SyuK6LN4eQ/3mmntjQdR7+HK6rSyMjLY9POE7Q7XHT1+5g9KYfmThfdg0NcMjUPx4CXubXFlOTZeHN7C6nZmdSU5bBpVys5Gckcv+eHtE9bzIpLa+natpVg5VRuWlpLyVeW0NTWR574Odrt4aTDTXFOKiX5djJTkmg82cdQMExWejKV1WV0OAbpcwVoOuWhNDeFN3a0kWwRXB4fbX0BMCFsycK2gz14/GHm1+Tj9HgpyE6l8biDFEsYhzPyb+vxhbFahI7uATy+EF+rL+Hlbe0IsGTGBN5v7MbtDdHrCVJgt2BLFpxDITr7hrBZLRRkp1BVlEmbw417KMD0ibn8ZlMrB5q7cfvCFOXY2Nvcy9QZ1ZzqdXOk3YmFIFsPD3LD4kqe3XSMG+aXc6LbRY8nRO+gh6x0Kysvm0RdZS75WalkZdjw+vzMqSkiNzOdP+xo4/pLKnlv70nKJ9ipK08nxZqEPxjG5w+BwJwphbR3O1k22c6TWzox4SDHTnkZcAfpc4fY3dSLYGhocbKorogdRxyc6HTiCxiumV9JqKCEr9TmktrZjMWWQq9f8HiD+AIh6ktTycjKYMPOVtJtFgY9AXIzk3lzby/pyeD0+Nh+2EFWupXXdnaTvuhS5j7/j8xeUMfJF5+j99ARXO0deDq7OVC+kOIcG9lpSdRV5dLS5SbZmoTLG2B6aTrPbTnBpMJ0irOT6XcHyLEns7d5gPzMZHYe68MYaHUMsXBqPjuaBiAcJjvdQrrNQmefj9I8G/uPDzC9Ips9zf24vCHmTcmj8vrryU6z0tzpIifDSm1pBs1dQ/R7wlw1t5BBt4+GgpnsbHZz4yUlVA2eYGt3EhhocfhIS0ni+1+dQlP7ILl2K1+bX07DiT4ml2VzzewSjnYOkJFi4ZLpE6ibmMVbe7pwe4NMLk5nQW0uyy+p5pmNR8i3W+h3B5lekcG1C6rY2thN2Bh8QUNnv4/sdCurr69jZX0+1gz731/gP7NqFOM9cvfW19evif0PXSnujzbs6Tpt/+k3D48SCXUV9rjr9QXjn+OhOD8j7tiZE7PijvX6R+/DZbOLT9tPT7GMGluWm3zafnN73yiR0NrtirN38NHB+D+V/a93jsYdm5Q00pwbEftbPKftV04YfaLGAU/4tP2uweCose82DsTZO8jKSIk7dvOetrhjf7/91KjHTgVPb7M0d/Q+9LhP/50drtCosc+/e3LUY2eaP7Uw7tj2Xm/csZfOLIo7dvaU+GM7ejyjHps5Mf6/BfW1+XHH3nb5SJPFqYvFuL4tH7tC3DBdKU4ppZQ6v/SZu1JKKZVgNLkrpZRSCUaTu1JKKZVgNLkrpZQaMyJSLCLPisgxETkgIutFpFZEhkRkl4g0isiHIvLdM85bJiK7RaRBRDbHlP84WrZfRJ4RkYRbMvx80OlnlVJKjQmJTAz/MvCUMWZVtGwuUAQcM8bMi5ZVAy+JSJIx5gkRyQEeBpYbY1pEpDAaVwb8CKgzxgyJyHNEFo95crx/2+edjtyVUkqNlcuBgDHmkeECY8xu4LTvFo0xTcBPiCRugG8CLxljWqLHY78TtgJpImIF0tElX0ekyV0ppdRYmQnsiDN2JzD8cX0tkCsib4vIDhH5DoAxpo3I0rAtQAcwYIz5w3nuc0L4PNyW9ybit+7bt29fc6H7oJRSY0lEVgOrY4rWGmPWftbqYratwJeBK4msHrdVRLYB3cCNwCSgH3heRL5ljPntZ2wzYV3w5D7SxDZKKaU+/6KJ/GzJvAG4Jc7q5gGN0e1WwGGMcQNuEXkHmBM91myM6QYQkZeAxYAm9zPobXmllFJjZSNgE5E7hwtEZD5QGRskIlVEbrc/GC36HXCpiFhFJB1YSCTxtwCXiEh69GW9K/n4gkDFuOAjd6WUUonJGGNEZAXwgIjcC3iB48DdwGQR2QWkAk7gQWPME9HzGkXkdWAvEAYeM8bsBxCRF4g8nw8Cuzj7nYMvLE3uSimlxowxph1YOcKhtHOcdz9w/wjlPwN+dn56l7j0trxSSimVYDS5K6WUUglGk7tSSimVYDS5f459++raC92Fs9rfMhh3bGqKnDsoyuMPxR07qTQ37tjz5bbLpsQdGw6buGNPdHvjji3MGv/XZZbOKYs79vr6orhj2/v8cceW2C1xx54vpXnxT12+Zf+puGP3Ho0/tiQ/Pe5YpQDEmPj/+CillFLq809H7koppVSC0eSulFJKJRhN7koppVSC0eSulFJKJRhN7koppVSC0eSulFJKJZj/D1gRFG2801akAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<Figure size 504x360 with 5 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "ast.type_clustermap(plot_name=\"./img/celltype_protein_cluster.png\", threshold = 0.7, figsize=(7, 5))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Note: `threshold` is the probability threshold above which a cell is assigned to a cell type, default to 0.7. "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 8. Hierarchical model specification <a class=\"anchor\" id=\"eight\"></a>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "In the marker yaml file, the user can also add a section called `hierarchy`, which specifies the hierarchical structure of cell types. Here's an example:\n",
    "```\n",
    "hierarchy:\n",
    "  immune:\n",
    "    - B cells\n",
    "    - T cells\n",
    "    - macrophage\n",
    "  epithelial:\n",
    "    - epithelial(basal)\n",
    "    - epithelial(luminal)\n",
    "```"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Some notes: \n",
    "1. The section would be accessed by key `hierarchy`.\n",
    "2. In the section, the higher-levelled cell type names should be the keys.\n",
    "3. The values in the section should also exist as the cell type names in the `cell_types` section. (e.g. if we have `\"B cells\"` in `marker[\"hierarchy\"][\"immune\"]`, we should also be able to get `marker[\"cell_types\"][\"B cells\"]`)\n",
    "\n",
    "This section could be used to summarize the cell types assignment at a higher hierarchical level. (e.g. a cell is predicted as \"immune\" instead of \"B cells\" or \"T cells\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>immune</th>\n",
       "      <th>epithelial</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>BaselTMA_SP43_115_X4Y8_1</th>\n",
       "      <td>0.435195</td>\n",
       "      <td>0.005911</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BaselTMA_SP43_115_X4Y8_2</th>\n",
       "      <td>0.385014</td>\n",
       "      <td>0.044718</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BaselTMA_SP43_115_X4Y8_3</th>\n",
       "      <td>0.188510</td>\n",
       "      <td>0.330741</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BaselTMA_SP43_115_X4Y8_4</th>\n",
       "      <td>0.444039</td>\n",
       "      <td>0.011956</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BaselTMA_SP43_115_X4Y8_5</th>\n",
       "      <td>0.468241</td>\n",
       "      <td>0.017085</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                            immune  epithelial\n",
       "BaselTMA_SP43_115_X4Y8_1  0.435195    0.005911\n",
       "BaselTMA_SP43_115_X4Y8_2  0.385014    0.044718\n",
       "BaselTMA_SP43_115_X4Y8_3  0.188510    0.330741\n",
       "BaselTMA_SP43_115_X4Y8_4  0.444039    0.011956\n",
       "BaselTMA_SP43_115_X4Y8_5  0.468241    0.017085"
      ]
     },
     "execution_count": 53,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "hierarchy_probs = ast.assign_celltype_hierarchy()\n",
    "hierarchy_probs.head()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "To make it more clear, here's a heatmap for the cell assignment in a higher hierarchy:"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<img src=\"img/hierarchical_celltype_cluster.png\" style=\"width:550px;height:400px;\">"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The way it is calculated is simply summing up the probabilities of the cell type assignments under the same hierarchy."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}