ReducedPositiveSamples.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Generate reduced positive samples\n",
    "\n",
    "Briefly:\n",
    "1. Transform protein and RNA sequences into vectors using prot_BERT and rna2vec\n",
    "2. Embed using optimized SAE\n",
    "3. Concat vectors\n",
    "3. Cluster using HDBSCAN\n",
    "4. Sample n samples from each cluster to esablish data  \n",
    "\n",
    "Headers:\n",
    "1. Train hyperparameter optimized RNA and Protein SAEs\n",
    "2. X"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/bobby/.local/lib/python3.8/site-packages/sklearn/linear_model/least_angle.py:30: DeprecationWarning: `np.float` is a deprecated alias for the builtin `float`. To silence this warning, use `float` by itself. Doing this will not modify any behavior and is safe. If you specifically wanted the numpy scalar type, use `np.float64` here.\n",
      "Deprecated in NumPy 1.20; for more details and guidance: https://numpy.org/devdocs/release/1.20.0-notes.html#deprecations\n",
      "  method='lar', copy_X=True, eps=np.finfo(np.float).eps,\n",
      "/home/bobby/.local/lib/python3.8/site-packages/sklearn/linear_model/least_angle.py:167: DeprecationWarning: `np.float` is a deprecated alias for the builtin `float`. To silence this warning, use `float` by itself. Doing this will not modify any behavior and is safe. If you specifically wanted the numpy scalar type, use `np.float64` here.\n",
      "Deprecated in NumPy 1.20; for more details and guidance: https://numpy.org/devdocs/release/1.20.0-notes.html#deprecations\n",
      "  method='lar', copy_X=True, eps=np.finfo(np.float).eps,\n",
      "/home/bobby/.local/lib/python3.8/site-packages/sklearn/linear_model/least_angle.py:284: DeprecationWarning: `np.float` is a deprecated alias for the builtin `float`. To silence this warning, use `float` by itself. Doing this will not modify any behavior and is safe. If you specifically wanted the numpy scalar type, use `np.float64` here.\n",
      "Deprecated in NumPy 1.20; for more details and guidance: https://numpy.org/devdocs/release/1.20.0-notes.html#deprecations\n",
      "  eps=np.finfo(np.float).eps, copy_Gram=True, verbose=0,\n",
      "/home/bobby/.local/lib/python3.8/site-packages/sklearn/linear_model/least_angle.py:862: DeprecationWarning: `np.float` is a deprecated alias for the builtin `float`. To silence this warning, use `float` by itself. Doing this will not modify any behavior and is safe. If you specifically wanted the numpy scalar type, use `np.float64` here.\n",
      "Deprecated in NumPy 1.20; for more details and guidance: https://numpy.org/devdocs/release/1.20.0-notes.html#deprecations\n",
      "  eps=np.finfo(np.float).eps, copy_X=True, fit_path=True,\n",
      "/home/bobby/.local/lib/python3.8/site-packages/sklearn/linear_model/least_angle.py:1101: DeprecationWarning: `np.float` is a deprecated alias for the builtin `float`. To silence this warning, use `float` by itself. Doing this will not modify any behavior and is safe. If you specifically wanted the numpy scalar type, use `np.float64` here.\n",
      "Deprecated in NumPy 1.20; for more details and guidance: https://numpy.org/devdocs/release/1.20.0-notes.html#deprecations\n",
      "  eps=np.finfo(np.float).eps, copy_X=True, fit_path=True,\n",
      "/home/bobby/.local/lib/python3.8/site-packages/sklearn/linear_model/least_angle.py:1127: DeprecationWarning: `np.float` is a deprecated alias for the builtin `float`. To silence this warning, use `float` by itself. Doing this will not modify any behavior and is safe. If you specifically wanted the numpy scalar type, use `np.float64` here.\n",
      "Deprecated in NumPy 1.20; for more details and guidance: https://numpy.org/devdocs/release/1.20.0-notes.html#deprecations\n",
      "  eps=np.finfo(np.float).eps, positive=False):\n",
      "/home/bobby/.local/lib/python3.8/site-packages/sklearn/linear_model/least_angle.py:1362: DeprecationWarning: `np.float` is a deprecated alias for the builtin `float`. To silence this warning, use `float` by itself. Doing this will not modify any behavior and is safe. If you specifically wanted the numpy scalar type, use `np.float64` here.\n",
      "Deprecated in NumPy 1.20; for more details and guidance: https://numpy.org/devdocs/release/1.20.0-notes.html#deprecations\n",
      "  max_n_alphas=1000, n_jobs=None, eps=np.finfo(np.float).eps,\n",
      "/home/bobby/.local/lib/python3.8/site-packages/sklearn/linear_model/least_angle.py:1602: DeprecationWarning: `np.float` is a deprecated alias for the builtin `float`. To silence this warning, use `float` by itself. Doing this will not modify any behavior and is safe. If you specifically wanted the numpy scalar type, use `np.float64` here.\n",
      "Deprecated in NumPy 1.20; for more details and guidance: https://numpy.org/devdocs/release/1.20.0-notes.html#deprecations\n",
      "  max_n_alphas=1000, n_jobs=None, eps=np.finfo(np.float).eps,\n",
      "/home/bobby/.local/lib/python3.8/site-packages/sklearn/linear_model/least_angle.py:1738: DeprecationWarning: `np.float` is a deprecated alias for the builtin `float`. To silence this warning, use `float` by itself. Doing this will not modify any behavior and is safe. If you specifically wanted the numpy scalar type, use `np.float64` here.\n",
      "Deprecated in NumPy 1.20; for more details and guidance: https://numpy.org/devdocs/release/1.20.0-notes.html#deprecations\n",
      "  eps=np.finfo(np.float).eps, copy_X=True, positive=False):\n",
      "/home/bobby/.local/lib/python3.8/site-packages/sklearn/decomposition/online_lda.py:29: DeprecationWarning: `np.float` is a deprecated alias for the builtin `float`. To silence this warning, use `float` by itself. Doing this will not modify any behavior and is safe. If you specifically wanted the numpy scalar type, use `np.float64` here.\n",
      "Deprecated in NumPy 1.20; for more details and guidance: https://numpy.org/devdocs/release/1.20.0-notes.html#deprecations\n",
      "  EPS = np.finfo(np.float).eps\n"
     ]
    }
   ],
   "source": [
    "import sys\n",
    "sys.path.append(\"..\")\n",
    "\n",
    "import pandas as pd\n",
    "from sklearn.model_selection import train_test_split\n",
    "import numpy as np\n",
    "import hdbscan;\n",
    "\n",
    "from Utils.model_utils import DenseSAE\n",
    "from Utils.model_datasets import SAEDataset\n",
    "from Utils.train import SAETrainer, dense_sae_splitter\n",
    "from Utils.encoders import SkipGramEmbedder, ProtBERTEmbedder\n",
    "\n",
    "import torch\n",
    "\n",
    "from fastai.basics import Learner, DataLoaders\n",
    "from fastai.callback.all import *\n",
    "\n",
    "\n",
    "from transformers import AutoTokenizer, AutoModel, pipeline\n",
    "import sentencepiece"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Load sequence data\n",
    "# Encode sequences\n",
    "# Train SAE\n",
    "# Embed RNA and protein data\n",
    "# Concat vectors\n",
    "# HDBSCAN \n",
    "# Sample 5k data from "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Train hyperparameter optimized RNA and Protein SAEs\n",
    "\n",
    "### Train hyperparameter optimized RNA2Vec SAE"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "def find_optimal_hyperparams_from_experiment(experimet_csv_path : str) -> dict:\n",
    "    hyperparam_opt_results = pd.read_csv(experimet_csv_path, index_col = 0).dropna()\n",
    "    min_loss               = min(hyperparam_opt_results['final_metric'])\n",
    "    min_val_exp            = hyperparam_opt_results[ hyperparam_opt_results['final_metric'] == min_loss ].iloc[0,:]\n",
    "\n",
    "    optimal_params = min_val_exp.to_dict()\n",
    "    return optimal_params"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "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>batchsize</th>\n",
       "      <th>hidden_layers</th>\n",
       "      <th>dropout</th>\n",
       "      <th>activation</th>\n",
       "      <th>final_metric</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>Bp2ga</th>\n",
       "      <td>64</td>\n",
       "      <td>[500]</td>\n",
       "      <td>0.2</td>\n",
       "      <td>None</td>\n",
       "      <td>0.000404</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>ouBb8</th>\n",
       "      <td>64</td>\n",
       "      <td>[500]</td>\n",
       "      <td>0.3</td>\n",
       "      <td>None</td>\n",
       "      <td>0.000535</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>dl8pb</th>\n",
       "      <td>128</td>\n",
       "      <td>[500]</td>\n",
       "      <td>0.2</td>\n",
       "      <td>None</td>\n",
       "      <td>0.000549</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>E7DqE</th>\n",
       "      <td>128</td>\n",
       "      <td>[500]</td>\n",
       "      <td>0.3</td>\n",
       "      <td>None</td>\n",
       "      <td>0.000676</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>gX67R</th>\n",
       "      <td>128</td>\n",
       "      <td>[500,250]</td>\n",
       "      <td>0.2</td>\n",
       "      <td>None</td>\n",
       "      <td>0.001038</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>aZ2Hh</th>\n",
       "      <td>8</td>\n",
       "      <td>[500,250]</td>\n",
       "      <td>0.5</td>\n",
       "      <td>nn.ReLU()</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>ljnAa</th>\n",
       "      <td>16</td>\n",
       "      <td>[800,300,200]</td>\n",
       "      <td>0.4</td>\n",
       "      <td>nn.LeakyReLU()</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>WNMiD</th>\n",
       "      <td>32</td>\n",
       "      <td>[800,600,400,200]</td>\n",
       "      <td>0.4</td>\n",
       "      <td>nn.ReLU()</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>aLMx4</th>\n",
       "      <td>32</td>\n",
       "      <td>[500]</td>\n",
       "      <td>0.3</td>\n",
       "      <td>nn.LeakyReLU()</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>XhuG0</th>\n",
       "      <td>8</td>\n",
       "      <td>[500]</td>\n",
       "      <td>0.6</td>\n",
       "      <td>nn.LeakyReLU()</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>219 rows × 5 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "       batchsize      hidden_layers  dropout      activation  final_metric\n",
       "Bp2ga         64              [500]      0.2            None      0.000404\n",
       "ouBb8         64              [500]      0.3            None      0.000535\n",
       "dl8pb        128              [500]      0.2            None      0.000549\n",
       "E7DqE        128              [500]      0.3            None      0.000676\n",
       "gX67R        128          [500,250]      0.2            None      0.001038\n",
       "...          ...                ...      ...             ...           ...\n",
       "aZ2Hh          8          [500,250]      0.5       nn.ReLU()           NaN\n",
       "ljnAa         16      [800,300,200]      0.4  nn.LeakyReLU()           NaN\n",
       "WNMiD         32  [800,600,400,200]      0.4       nn.ReLU()           NaN\n",
       "aLMx4         32              [500]      0.3  nn.LeakyReLU()           NaN\n",
       "XhuG0          8              [500]      0.6  nn.LeakyReLU()           NaN\n",
       "\n",
       "[219 rows x 5 columns]"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "rna_exp_csv = pd.read_csv(\"../nni/experiment_results/RNADenseSAE.csv\", index_col = 0).sort_values(\"final_metric\")\n",
    "rna_exp_csv"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Load model params\n",
    "rna_exp_csv = \"../nni/experiment_results/RNADenseSAE.csv\"\n",
    "\n",
    "rna_params = find_optimal_hyperparams_from_experiment(rna_exp_csv)\n",
    "\n",
    "BATCHSIZE     = rna_params.pop(\"batchsize\")\n",
    "HIDDEN_LAYERS = eval(rna_params.pop(\"hidden_layers\"))\n",
    "ACTIVATION    = eval(rna_params.pop(\"activation\"))\n",
    "DROPOUT       = rna_params.pop(\"dropout\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Initiate optimized model\n",
    "rna_sae = DenseSAE(1024, hidden_layers = HIDDEN_LAYERS,\n",
    "                   activation = ACTIVATION,\n",
    "                   dropout = DROPOUT)\n",
    "\n",
    "# Early stopping callback will cut off training when loss stabalizes\n",
    "trainer = SAETrainer(9999, 255, plot = True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'/data/Chapter_3/Notebooks'"
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "pwd"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "text/html": [],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: left;\">\n",
       "      <th>epoch</th>\n",
       "      <th>train_loss</th>\n",
       "      <th>valid_loss</th>\n",
       "      <th>time</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>0</td>\n",
       "      <td>0.048093</td>\n",
       "      <td>0.003112</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>0.021075</td>\n",
       "      <td>0.001481</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>2</td>\n",
       "      <td>0.010237</td>\n",
       "      <td>0.001292</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>3</td>\n",
       "      <td>0.006522</td>\n",
       "      <td>0.001180</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>4</td>\n",
       "      <td>0.004954</td>\n",
       "      <td>0.000850</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>5</td>\n",
       "      <td>0.004504</td>\n",
       "      <td>0.000749</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>6</td>\n",
       "      <td>0.004381</td>\n",
       "      <td>0.000657</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>7</td>\n",
       "      <td>0.004270</td>\n",
       "      <td>0.000953</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>8</td>\n",
       "      <td>0.004201</td>\n",
       "      <td>0.000584</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>9</td>\n",
       "      <td>0.004224</td>\n",
       "      <td>0.000652</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>10</td>\n",
       "      <td>0.004152</td>\n",
       "      <td>0.000574</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>11</td>\n",
       "      <td>0.003877</td>\n",
       "      <td>0.000633</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12</td>\n",
       "      <td>0.003930</td>\n",
       "      <td>0.000582</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>13</td>\n",
       "      <td>0.003850</td>\n",
       "      <td>0.000616</td>\n",
       "      <td>00:06</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>14</td>\n",
       "      <td>0.003815</td>\n",
       "      <td>0.000508</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>15</td>\n",
       "      <td>0.004123</td>\n",
       "      <td>0.000511</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>16</td>\n",
       "      <td>0.003736</td>\n",
       "      <td>0.000655</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>17</td>\n",
       "      <td>0.003786</td>\n",
       "      <td>0.000472</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>18</td>\n",
       "      <td>0.003972</td>\n",
       "      <td>0.000616</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>19</td>\n",
       "      <td>0.003570</td>\n",
       "      <td>0.000524</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>20</td>\n",
       "      <td>0.003558</td>\n",
       "      <td>0.000450</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>21</td>\n",
       "      <td>0.003671</td>\n",
       "      <td>0.000516</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>22</td>\n",
       "      <td>0.003864</td>\n",
       "      <td>0.000433</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>23</td>\n",
       "      <td>0.003690</td>\n",
       "      <td>0.000572</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>24</td>\n",
       "      <td>0.004056</td>\n",
       "      <td>0.000722</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>25</td>\n",
       "      <td>0.003744</td>\n",
       "      <td>0.000352</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>26</td>\n",
       "      <td>0.003877</td>\n",
       "      <td>0.000436</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>27</td>\n",
       "      <td>0.003757</td>\n",
       "      <td>0.000528</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>28</td>\n",
       "      <td>0.003614</td>\n",
       "      <td>0.000522</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>29</td>\n",
       "      <td>0.003737</td>\n",
       "      <td>0.000442</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>30</td>\n",
       "      <td>0.003419</td>\n",
       "      <td>0.000553</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>31</td>\n",
       "      <td>0.003694</td>\n",
       "      <td>0.000547</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>32</td>\n",
       "      <td>0.003749</td>\n",
       "      <td>0.000476</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>33</td>\n",
       "      <td>0.003326</td>\n",
       "      <td>0.000369</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>34</td>\n",
       "      <td>0.003932</td>\n",
       "      <td>0.000495</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>35</td>\n",
       "      <td>0.003573</td>\n",
       "      <td>0.000462</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "No improvement since epoch 25: early stopping\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: left;\">\n",
       "      <th>epoch</th>\n",
       "      <th>train_loss</th>\n",
       "      <th>valid_loss</th>\n",
       "      <th>time</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>0</td>\n",
       "      <td>0.003543</td>\n",
       "      <td>0.000652</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>0.003611</td>\n",
       "      <td>0.000507</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>2</td>\n",
       "      <td>0.003846</td>\n",
       "      <td>0.000689</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>3</td>\n",
       "      <td>0.003180</td>\n",
       "      <td>0.000619</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>4</td>\n",
       "      <td>0.003765</td>\n",
       "      <td>0.000708</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>5</td>\n",
       "      <td>0.003632</td>\n",
       "      <td>0.000485</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>6</td>\n",
       "      <td>0.003513</td>\n",
       "      <td>0.000888</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>7</td>\n",
       "      <td>0.003890</td>\n",
       "      <td>0.000518</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>8</td>\n",
       "      <td>0.003527</td>\n",
       "      <td>0.000774</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>9</td>\n",
       "      <td>0.003626</td>\n",
       "      <td>0.000930</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>10</td>\n",
       "      <td>0.003705</td>\n",
       "      <td>0.000576</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>11</td>\n",
       "      <td>0.003926</td>\n",
       "      <td>0.001105</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12</td>\n",
       "      <td>0.003941</td>\n",
       "      <td>0.000457</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>13</td>\n",
       "      <td>0.003469</td>\n",
       "      <td>0.000904</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>14</td>\n",
       "      <td>0.003304</td>\n",
       "      <td>0.000859</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>15</td>\n",
       "      <td>0.003520</td>\n",
       "      <td>0.000726</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>16</td>\n",
       "      <td>0.003244</td>\n",
       "      <td>0.000313</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>17</td>\n",
       "      <td>0.003160</td>\n",
       "      <td>0.000307</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>18</td>\n",
       "      <td>0.003328</td>\n",
       "      <td>0.000557</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>19</td>\n",
       "      <td>0.003385</td>\n",
       "      <td>0.000699</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>20</td>\n",
       "      <td>0.003641</td>\n",
       "      <td>0.000602</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>21</td>\n",
       "      <td>0.003621</td>\n",
       "      <td>0.000426</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>22</td>\n",
       "      <td>0.003672</td>\n",
       "      <td>0.000612</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>23</td>\n",
       "      <td>0.003541</td>\n",
       "      <td>0.000583</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>24</td>\n",
       "      <td>0.003832</td>\n",
       "      <td>0.000811</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>25</td>\n",
       "      <td>0.003536</td>\n",
       "      <td>0.000620</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>26</td>\n",
       "      <td>0.003547</td>\n",
       "      <td>0.000657</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>27</td>\n",
       "      <td>0.003699</td>\n",
       "      <td>0.000510</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "No improvement since epoch 17: early stopping\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAY4AAAEWCAYAAABxMXBSAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjQuMSwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy/Z1A+gAAAACXBIWXMAAAsTAAALEwEAmpwYAABQrElEQVR4nO2dd5gURdrAfy/LsuSMZAUUJOekgmJCgooBBSOG+zCe+e7wzN4ZzhzAnCMiJpQkSJYMIjksObOA5LShvj+qZ7dntmd2endmd1ne3/Pssz3V1d1VE+qtelOJMQZFURRFiZZiBd0ARVEU5fhCBYeiKIriCxUciqIoii9UcCiKoii+UMGhKIqi+EIFh6IoiuILFRxKkUFEbhKRaQXdjoJARA6ISIN8fuaTIvJFfj5TKRyo4FDyjIh0EZHpIrJXRHaLyO8i0iGkTllncBvtcf06ETnsnA/8DQ7zrCdFJNWps8d57hnx6ptfROQfIrJYRPaLyFoR+UeYel1dfT0oIiak/yf7ea4xpqwxZo3PtnZ2nl3W49wfInK3n/uFXN9NRDbl9nqlcKOCQ8kTIlIe+AV4E6gM1AaeAo6GVL3SKbtQRGp43OoSZ/AL/EUatL4xxpQFqgHTgO9FRPLalxghwI1AJaAHcLeI9A+tZIyZGugr0Mwprujq/4bMG4oUj0dDjTEzgU1A36AOiDQHmgJfx+O5yvGPCg4lrzQCMMZ8bYxJN8YcNsb8aoxZGFJvAPAOsBC4PhYPNsakAp8CNYAqgXIReUlE/nJm/D1d5TeLyDJnNbBGRG5znasqIr84q5jdIjJVRIo552qJyHcikuLc854IbXrBGDPfGJNmjFkB/ASc5adfzqpquIh8ISL7gJtEpKOIzHDat1VEBotICdc1RkROc44/EZEhIjLS6essETk1zOM+xQo6NzcCo4wxu0TkdRHZKCL7RGSeiHT105cw/WsiIpOcviwRkUtd53qJyFKn3ZtF5CGnPOzno+Q/+sYreWUlkC4in4pITxGpFFpBRE4BugFfOn+hA1WuEJEk4CZgozFmp1PcCVgBVAVeAD50rUZ2ABcD5YGbgVdFpK1z7kHs7LsaUB34N2Ccweln4E/saup84D4RuSiK9gnQFViSi+71AYYDFbHvWTpwv9OvM5x23Bnh+v7YlV8lIBl4Jky9z4GzRaSu0+ZiwLVYgQIwB2iNXU1+BXwrIiVz0R+c+ydi389fgZOAvwNfisjpTpUPgduMMeWA5sAEp9zz88ltO5S8oYJDyRPGmH1AF+yP+H0gRURGiEh1V7UbgIXGmKXAUKCZiLQJudWPzmwy8Pd/ER57tYjsATYC7YDLXefWG2PeN8akYwe/mtiBBmPMSGPMamOZjB28AjPoVKfuKcaYVEeVZIAOQDVjzNPGmGOOHeF97MCcE09if2MfR1E3lBnGmB+NMRnOKm6eMWams5JZB7wLnBPh+h+MMbONMWlYwdPaq5IxZiMwCfsZgRVIScBI5/wXxphdznNfds6d7nWvKOkMlAWed97PCVhV5zXO+VSgqYiUN8b8ZYyZ7yr3+nyUAkAFh5JnjDHLjDE3GWPqYGeJtYDXXFVuxA5eGGM2A5Oxqis3lxljKrr+3o/wyGFOnZOMMecZY+a5zm1zteuQc1gWwFkRzXRUHXuAXtgZPMCL2Jn5r44aa5BTfgpQyy3UsLNdt2DMhmNYvhHobYwJtfdEw8aQ+zVyVDXbHPXVs662e7HNdXyIrPfg3y4D/DvO+U/JEhw3AEMdNSAi8pCj3tvr9L1CDs/NiVrYFWKGq2w9djUH1hbWC1gvIpMly/Eh3OejFAAqOJSYYoxZDnyCFSCIyJlAQ+BhZ9DbhlUnXRsvo68XjlrrO+AloLoxpiIwCmvMxhiz3xjzoDGmAXAp8ICInI8dwNeGCLVyxpheEZ51CzAION8Yk1vPotDZ9NvAcqChMaY8Vnj5dggwxjzrMsDf7hR/D9QRkXOBK3DUVI4945/A1UAl5z3bm5vnutgC1A2xT5wMbHbaN8cY0werxvoRGOaUh/t8lAJABYeSJ0SksYg8KCJ1nNd1sWqHmU6VAcA4rJdOa+evOVAK6Bl6vzhSAqtmSQHSxBrNuwdOisjFInKaY5fYi7UpZACzgf0i8i8RKSUiCSLSXELcjV33uQ67GrjQr3tsDpQD9gEHRKQxcEesbmyMOYi1p3yMVfXNdT0zDfueFReRx7H2oagRkZLuP+z7eQj4p4gkikg34BJgqIiUEJHrRKSCs+LZh/0MIn0+SgGggkPJK/uxK4hZInIQKzAWAw86A8XVwJvGmG2uv7VYo6xbXfWzBMcx/BDLRhpj9gP3YGewf2ENwCNcVRoC44EDwAzgLWPMRMdWcjFW4K0FdgIfYFU2XvwX6+E1x0MllBcectq8H2tj+SYG93TzKVYt95mrbCwwBusAsR44QogKLQdqA4dD/upiBUVP7Hv5FnCjs1IFqypb56jjbgeuc8o9Px9/XVRihah9SVEURfGDrjgURVEUX6jgUBRFUXyhgkNRFEXxhQoORVEUxRf55kdfkPTo0cOMGTOmoJuhKIpyPBE2XueEWHHs3Lkz50qKoihKVJwQgkNRFEWJHXEVHCLSQ0RWiEiyV24ZEUkSkW+c87NEpJ7r3MNO+Qp3JlKxm/4sEpEFIjI39J6KoihKfImbjUNEEoAhwIXYdMhzRGSEkyE1wK3AX8aY08RudvM/oJ+INMVmH22GTYo2XkQaOVG8AOe60mgriqIo+Ug8jeMdgeRAvh4RGYrdY8AtOPpgU0+DzZUz2MlF0webofMosFZEkp37zYhjexVFUQBITU1l06ZNHDlypKCbEndKlixJnTp1SExMjPqaeAqO2gTntdmEzWnkWccYkyYie7F5fmqTlSQvcG0g7bLBplY2wLvGmPe8Hi4iA4GBACef7Gv7ZkVRTnA2bdpEuXLlqFevHlJodiWOPcYYdu3axaZNm6hfv37U1x2PxvEuxpi22CRpd4nI2V6VjDHvGWPaG2PaV6tWLX9bqCjKcc2RI0eoUqVKkRYaACJClSpVfK+s4ik4NmMzYQao45R51nH2ZqgA7Ip0rbMREMaYHcAPWBWWoihKTCnqQiNAbvoZT8ExB2goIvVFpATW2D0ipM4IslJr9wUmONtBjgD6O15X9bEplWeLSBkRKQcgImWw+yksjlcHZqzexeqUA/G6vaIoynFJ3ASHs9fx3dic/suw230uEZGnReRSp9qHQBXH+P0Adtc0jDFLsPsmLMXuB3CX41FVHZgmIn9iN4QZaYyJW0j4Ne/P5PyXJ8fr9oqiKJ7s2bOHt956y/d1vXr1Ys+ePbFvUAhxTTlijBmF3Z7TXfa46/gIcFWYa58BngkpWwO0in1LFUVRCg8BwXHnnXcGlaelpVG8ePhhe9SoUWHPxZITIleVoijK8cSgQYNYvXo1rVu3JjExkZIlS1KpUiWWL1/OypUrueyyy9i4cSNHjhzh3nvvZeDAgQDUq1ePuXPncuDAAXr27EmXLl2YPn06tWvX5qeffqJUqVIxaZ8KDkVRlAg89fMSlm7ZF9N7Nq1VnicuaRb2/PPPP8/ixYtZsGABkyZNonfv3ixevDjTZfajjz6icuXKHD58mA4dOnDllVdSpUqVoHusWrWKr7/+mvfff5+rr76a7777juuvvz4m7VfBoSiKUsjp2LFjUJzFG2+8wQ8//ADAxo0bWbVqVTbBUb9+fVq3bg1Au3btWLduXczao4JDURQlApFWBvlFmTJlMo8nTZrE+PHjmTFjBqVLl6Zbt26ecRhJSUmZxwkJCRw+fDhm7TkeAwAVRVGKNOXKlWP//v2e5/bu3UulSpUoXbo0y5cvZ+bMmZ714omuOBRFUQoZVapU4ayzzqJ58+aUKlWK6tWrZ57r0aMH77zzDk2aNOH000+nc+fO+d4+FRyKoiiFkK+++sqzPCkpidGjR3ueC9gxqlatyuLFWbHRDz30UEzbpqoqRVEUxRcqOBRFURRfqOBQFEVRfKGCQ1EURfGFCg5FURTFFyo4FEVRFF+o4FAURSkClC1bFoAtW7bQt29fzzrdunVj7ty5eX6WCg5FUZQiRK1atRg+fHhcn6GCQ1EUpRAyaNAghgwZkvn6ySef5L///S/nn38+bdu2pUWLFvz000/Zrlu3bh3NmzcH4PDhw/Tv358mTZpw+eWXxyxflUaOK4qiRGL0INi2KLb3rNECej4fsUq/fv247777uOuuuwAYNmwYY8eO5Z577qF8+fLs3LmTzp07c+mll4bdN/ztt9+mdOnSLFu2jIULF9K2bduYNF8Fh6IoSiGkTZs27Nixgy1btpCSkkKlSpWoUaMG999/P1OmTKFYsWJs3ryZ7du3U6NGDc97TJkyhXvuuQeAli1b0rJly5i0TQWHoihKJHJYGcSTq666iuHDh7Nt2zb69evHl19+SUpKCvPmzSMxMZF69ep5plSPN2rjiECJhGI0qFYm54qKoihxoF+/fgwdOpThw4dz1VVXsXfvXk466SQSExOZOHEi69evj3j92WefnZkscfHixSxcuDAm7dIVRwRa1a1A8WIqWxVFKRiaNWvG/v37qV27NjVr1uS6667jkksuoUWLFrRv357GjRtHvP6OO+7g5ptvpkmTJjRp0oR27drFpF0qOCIgCAZT0M1QFOUEZtGiLMN81apVmTFjhme9AwcOAFCvXr3MlOqlSpVi6NChMW+TTqcjIWBUbiiKogShgiMCArreUBRFCUEFRwREJYeinLCYE0TdkJt+quCIgNo4FOXEpGTJkuzatavICw9jDLt27aJkyZK+rlPjeAREbRyKckJSp04dNm3aREpKSkE3Je6ULFmSOnXq+LpGBUcERFRTpSgnIomJidSvX7+gm1FoUVVVBAQp8ktVRVEUv6jgiICuOJSixPTknazdebCgm3HCsvPAUfYeTi3oZsQEFRw5UJQXHMfSMjh4NC3fn7tj3xE+m7Eu359b2Hn51xUs2bI3ZverN2gkAz/L2rTn2g9mce5Lk2J2f8Uf7f87nrb/GVfQzYgJKjgiICIY4ODRtAIZYONNo0dH0+yJsfn+3IGfz+Pxn5awcfehfH92tOw9lL8zw/QMw5sTkukz+PeY3vfXpdtjej8lb6RnFI2ZqAqOCAiAMTR7YmyBDLDhSM8wTFqxo6CbkWv2HDoGQGp6RgG3xJtfl2yj1dO/Mmfdbt/Xzlyzi8PH0nP97LQiMrAoRRsVHBHIycYxeMIq7vhiXr61J8A7k1dz08dzmLA8NrPJ/HYACLfpTGFh+updACzc5E9ttHH3Ifq/N5N/fZeVgXTZ1n3c8OEsjqRGFibqhKEcT8RVcIhIDxFZISLJIjLI43ySiHzjnJ8lIvVc5x52yleIyEUh1yWIyB8i8ktc2w/siaCyeOnXlYxevC2eTfAkoOLZvu9oTO73zMhlMblPtBR2A+3SrfsAGLlwC9e8NzPq6w47wmGZcz3AYz8uZuqqnb6FkKIUZuImOEQkARgC9ASaAteISNOQarcCfxljTgNeBf7nXNsU6A80A3oAbzn3C3AvEPfRTkQ4dKzw2TZ+W27VVLEagKeu2pnne8xdt7vIeIzMXmtVVPM37GGGh+rp0LE0zxVCQjG7knLrsRMT7E+ssKrlFCU3xHPF0RFINsasMcYcA4YCfULq9AE+dY6HA+eL1WP0AYYaY44aY9YCyc79EJE6QG/ggzi2HQA7DhQ+tUrKfrvSGLVoa0zuV6xY3vp4LC2Dvu/MYMBHs31dlxEH9cySLXu5+t0ZOaqG/LB8W9YKImX/UZo+PpZ3Jq/JVi/BUcG57RSJxe1PbP+RyBMQVVQVLowx3PXlfHYfPJYvz0tNz4jpdzbexFNw1AY2ul5vcso86xhj0oC9QJUcrn0N+CcQcQonIgNFZK6IzM192gCheB4H1Xiy6a/DMblPXvuYvMPuA7Bg4x5f18XDEPzgsD+ZvXY3M9bsytX1Xp5e7lZu32e36fz5zy3Z6gVWHBtc9wg4AjwwbEGu2qMUDD1fn8rIRVvzzX22/3szafzYmHx5Viw4rozjInIxsMMYk6NF2hjznjGmvTGmfbVq1XL5PNi2L/b7+Y5ZvJWte2Mz6MeCcxpF9/6c8dxv/HP4n9nKt+3LXV/S0nMnOI6kpnPLJ3PYvCf7c5dv2w/Aex4rgpyYs243XV+YmK3cS5gsddkxAiR4CODALPJQDp5W+WkbP3A0jaNpwe05fCyd18av5Fja8a1SS9l/lBXOdyAv7Mtnteu89X/l6/PySjwFx2agrut1HafMs46IFAcqALsiXHsWcKmIrMOqvs4TkS/i0XjInZLKGMNnM9aFXeJu2HWI27+YzxnPTQh7j+QdBziQj3EjpUok5FwJ2Lr3CMPmbspWvmLbgVw9N9yKIz3D8NzoZTz981IyPOqMXLiVCct3cNbz4d/DnOwtxhiGz9sUZI8Yv8zbSy2peAJ/bLA/7GIRPMJCBcfCTXtyFBgFQfMnxnL6o8Gz2w+mruG18av4clbkPaxDmbZqJws37Ylh6/LGOS9O5KLXpuT5Plv2Zk0Yb/hwFq+MW+k5UYkWv15zqekZhdpmGE/BMQdoKCL1RaQE1tg9IqTOCGCAc9wXmGDsOzwC6O94XdUHGgKzjTEPG2PqGGPqOfebYIy5Pl4dyI3X6KLNe3n8pyVhl7gvj1uR4z0ueGUy170fnTfPjhisiOpUKuWr/paQH9An09d61pu4fAf7j4T/8i/abD2NNu4+xGVDfs9U6/y0YDPvTl7DR7+vZVpydsP9b1G4IXutCNwMmZjMQ9/+yYWvTM4sSwjzgd/+xTwuf2s6EPk7sc012BxNS+fSwb/HTJ0YDSu37+eSN6cxZaV/1eyhKFdGoVz/4SwujXHQYjgyMgzTk3dGHITjIainrtrJG7+t4s4v50d9zR1fzOPFscszX/tdUd791XxaPfWrv4vykbgJDsdmcTcwFusBNcwYs0REnhaRS51qHwJVRCQZeAAY5Fy7BBgGLAXGAHcZY/J96hb6YUcT9fnoj4sjno80Y3Xzp4f75pVvT6fFk8GBiO6ZUW5JKu7vaxDqouzlFrx5z2Fu/mQO9w1dEPY+jznv1b9/WMSCjXt46Fsb/7DrQNZq7T+/LM123ahFeXeBDgiWNT4804wxEQXHrZ/OyTyO5ruSkWF4a1Iyew+n8uMfoYtx//R9ezqLNu/lRp9OCuB/db1t7xF27I+9GjcSb09ezbUfzPI1gEfDmpQDXPzm1ByzBRzxIZRGL97GkImrM1/7dQQZu6RwR/zHNa26MWYUMCqk7HHX8RHgqjDXPgM8E+Hek4BJsWhnOELTNRxLy8hRrZOTv36NCv42THHjpQcdtWgrretW9HWftTsPBumy/XqKRrO51VFnBus1MJcoXizo+QGjekBVdNjlXbJqR7AaLK+eJ6/8uoIPp63lli7ZU2b/vDC7wdtNarpBIgyxO10CL9I4cSQ1nZKJCUxL3skLY1awfOv+mLh9p/q0GaVnmEz1mt/Vdefnfouq3v4jqSzdso9ODar4e4AHL461q/VoYqeGzdnI1R3q5lgP4JEfFrN48z7GLdtO33bh96XIzaZuN388m49v7sgPYSYG89b/xa9LsvpjJycS9nVh4bgyjhc0qRl5NxyGU4d4MT15J2e/MDHbYFmhVGLm8ZqUyDNmYwyfz1zPXy6by7kvTQrSA+c0G1q/62DQbCyayVNgZeU18w41wIbaMUpEWAHl1T3yjQnJHDyWzgiXV1TA+L1xd2S1UloOn39p16Qi3Hs6PXknjR8bw4zVuzJddEf8uYVjuXQUWL5tH4/+uIilW/YFCdxoCLRx697DmbPjwOAcK+74Yj793pvJvggqy3jwT1f0fiiLN+8N+j0EPPCSd0S21RkDM1bvYuGmPUFqyUhMXJHC9n1HWL/LOy/blW9P590pWY4cocI/L46HN308mx4xsPd4oYLDB7n1AnLTKofVgXtgfPqXpWzYfYjFm4NXMeVKZi0Uwxl0AyzZso/HflzM/Y47qJd+OCfBcc6Lk+jxur8voFcwnNdrgJZ1Kma1JcMECcZYkbzjAE/8lKVGLF8y6xkTlkeX9ys13ZAQ4Rdz5qlZs+pwMmamE1w4a+2uoAnB+l1ZE4BAQs2MDMPgCasiGkl7vDaVL2ZuoNcbU4PKozHGBj6Lq96ZkWPd3DLbyfdl8jDn+mXhFt+u3pG4+M1pXPn29GzlOXmUZRjDNe/P5NLBv9P5ud+iFoaHj6VHvaJLTc9g54Es1e+p/x4VoXZ29h62KzyASStSMr0MY40KDh+kuXQ67h+9l+dPKGtSDnDey5MyjcUnVy7tWe+nBVlL2sCH/sSIJUF1/BhcA26Xk1ZYg6nXeBLw4Fq1fX/YLMBbXTMs9w8mXEK/wA8ldADzElLumIt0Y4IEY7j7AtSvWiZsvVAueGUyn87I8hhqXKNcxDZ5MWnFjqCZ45qU4Bnq+GVZAmjTHu8ZZsDxKi3dUCYpq5/u+wZWDpNXpvDSryt5MuTzj4ZouhT47KK1u+WGwGC882Du0+Pc/dUfXDYkvAHeGMOmv/xlWvZSoX4zZ0PEa1aHrO6jzZid7sO+kZqeQfv/jg8qS94R/eB/zXszs00i4oEKDh+kZhheGLOctTsPBg2YoXr4AFv2HGbo7A3c9vlcznt5MmtSDvLAMBsHEW6weurn7MbgJVv2ZXocRWL66p18Ny/YXdatd39u1DJeHb8y23WP/LCYTX8d4sJXp3DzJ3OynQ/l2vdnZR5f94G391dAL+vu5ZjFW7OlSXn4+0VBr8MZldftPJjth1o+DyuT8xqflHkcrTrg3qELuPXTrP0tPpuxnnqDRnquCD6c6u1p9qczc35v6hrKJHnbywLDeOC9yI1bZjTC8MNpto3uGW5uafb4GOoNGhmkAnIzb12wfe6FMcu5NkrPwXAcTUtn9KKtPPLjYrr8b2JUsRChExn3av6gT4+snNTEAUInlpEmml6rnkAuuZT9R/l6thVuAz6aTbcXs8ccBZw+4h24qHuO++Ddyav5bMZ63pq0Oqg8XB6iMyPEGfh1z7virexL61ACA/qVLgNfmRJZH7FblxpKl//ZL2EgT1OAeoNGRnzm/A17PMsDAmzr3iMs37aPxjXKc/sX2b1hAj+EAJv3HOZZj6SL3V6aRMd6lXm1f+uswjxEzR1xBcDlduD8ZPo6AFanHKDtyZWCzn0fxhjaum4lJq5I4ZxG1Ugq7i04AjPU4gnZU5hES+glXkFxuw4eY8zirTFxYQ0Muo/8uIi3rmuX7fwLY5cHGatDf0PRcFGz6kHeRs+PXs7Hv6/LfB1N4J/7vXx21DLei/CbyAmvVc60VTu5/sNZQWV7DqcGGcePpWdQspj3Z3/MYywJaATu+GIec9f/RZfTqjI5B5freKdK0RWHDz5zqTrc5OSN44XXjDDSTMSP26ib6uWTcnVdXnllXNbKpsdr0S+dt+87EtbFePa63UHvUV4Mh+6Z3cTlO2j4iD9dspv3oxh8yjiG84AarnbFUrwz2XvwTM8wDJ29IfM7khZmYhIq5N2Efr/e9XjWuKXbGTwxOce2+yGcq3RuNzDa4FLhhQqGdSG/iUjOC0MmJlNv0MigVWs4oRFu1RTK6+NXZSt7b2r2e171zowg9XKkiYDXgB/IXvGXo3UoDDmtVHDEgHcnr8k0SEWL+4e9Zc9hbv54Ni/9GhuPll2uGbQf/aoftocJPMzrvhIHj2b/Ubjv6R6AonGPTEvPwBhDuaTgxfV/f8la1Szftt+3K6ubaNxDL21tU625bT/hjPJjFm9j0PeLeM0ZmMINNJFUWKEfw4792VdVOw8cZfHmnL+3KR7X+iU9w/B78s5sAaHhhGKAC17NCtBcF+KZFBqpH27lNH/DX5neYjmtSnYfPEabKNU8f3nEfUST9i01ghF+4GfZsykFnEUC/b/rq6yV++SVKdQbNDLfd9NUwREjer0xlX98mz2PUzi27zvKase4+sq4lUxckZKr5bsXgS/05j2H+XjaOt/Xf+AxawolXCZct+ogN3hFmrvHTbcgDJ1ghka079h3hNMeGc3bk1ezP8Q+Evo6r8xbH3m3wIBKbqzjs78nwqAfcNMNBEIGhJ8fQicMfoM8AyzYuIcOz4zn+/lZtrM/c+HhtO9IGtd9MIu7v/ojqPxImEF08ea9jFu63VPnn5ggpGeYIGcEsKorN/8avpAJy7cHqXl/8khO6cZPUKOX+/PMKJJrRkqx75Ub78bO9YCsSdPK7Vk21YBKOL9zXangiCHfzsuexykSfR2XQK/ZYF4IzMSueW8m38zdmEPt7Pw3io2dwrn5Pe0R6Q3ReZ4Bmc4DQdcGCYus46Vb97Fye1Y77v9mQdB1Gx0d9AtjYhub4IXbCSEcQyYmM3ONFTCrU8LHDAQmrQFhMX/DHvqHbCi159CxiMIkdFZ7MEyAYcXS2R0MxizeysVvTsUYk/n+/p68i+37jrD/SGqO6fyHzg7vnRSqmw/09dVxK4ME0sVvTuP/PpsbVDfgyl4qMSFsQJ2bb+ZuDPIGhJzNYn5UalXLZlcDH0nN2e84YMeIdpWw8+DRsLbGgNdkfqV/D6CCIwJPXhK671RsCRi9cpNbKBJz1u3mqnemB6X3zk/cLsUB8qIycwuO0NxVbj3zrBCdf6SIW7/5uXIiVN/uhTu4LtLbEWi229bj7lt6hqH10+MYFOKR5sZtZH3656WZAiuUy1qH7nQAt38xn8Wb9zF2yXY+d+x6R9PS6fTsb1z06pQc4xcGfb+IGz6cRYdnxnuedw+YzZ6wKXRe/20VfSK43ELWSifDRJ+9dv76PUGvt+WQlXpWmPfJi50HjnIkNZ1Za3ZRb9DIqF2CU9MNP/+5xTMTsxeRJj4BZ4HnxywPWyceqOCIwHWdTwHgrnNPjcv9Y7UdxdMhLrxv/LaKOevyvnR1b2DkRbg0Gfd65KfKrXEUrDtygFB35ZERZr+R4hNinXzwudH+friRopRf+jW7yzTYzK9DJiZnqjoizTLdKp6Pfvd2DQZISgw/BNz+xbzMRJS/LLTv85a9R6LaeXLqqp1hbSOXv5W3pIgHjqaxbld0ziLfzQ/WAuTkZBJuxRyOr2ZtoJ+zGpyxOro9YEYs2MI/PLYnyAvhghcva10rps8JoIIjAokJxVj3fG/6dzg5Lvc3xuTZmAzZB4ZYDIqtn/41R2+omz7OOeYjwBu/ZfdAiZacXJHDGYkL8R5cHM3Fvhfrdx2KOiVItPdPLOZ/CAi3eomWULXep45bsx/CeTjmRLjUH7nFLWiinRx9O29jVCqtWBCv4E4VHFFQJik+4S4ZJnc/GoBTqnhHnseK0Ay4XkRyBw0lVoZ/L+4d+odneX5ujpSfRNOvaDdkmuqRtj6/Cc2McLzyo4eK1ot4pdr3siPGK0GiCo4oqFymRNzu/aRHpHg0bNt7hGa1yse4Nccn4fIY/RVFtP3xSKQ9TgJE8txxE62HVMf6laOqFwtCveOOFwpyonJNx7qZe6q4ideqWwVHHHi6T7O4P+NoWkbMbCTHO3sOpXoGRQ2eEJvgto9uah+T+8SKjs/mnNI8Gq+jwshLY1dEzLhQmAl1zshP2tSt5FmuqqpCyBVts3ukAFzaKj4GqVBy65tfFGn82JhsZX5zD7lpXtuu5ro2rEqdSvFVC8aDT3KpAg2H3wDX3BLrSPa80CePhuXi+WhkSzcmcw8cN343kIoWHXnywLOXt/AsTygmXNMx9wb1e89vGFW9WKaahuD9JIoCy3LYPjYSJ5WzG24lJhTLFqF8InIgxgGTxwOP9G5Ct9Or5fr6Yvn4vZmWvDMoMDCA39iyaFHBkUtuO6cBJRO9B9pyJRPzpFs867SqubquZh52FwTo0bxGVPVKRNqUIkY0qVmw9puGJ5UFrKdMuNTxStHlXz0ac1K5knxyc8dc36N+lejT/n93x5m5fg7AyIVbqVYu//LSqeDwyQc3tqdJzfIMOKNexHp5UVeFS7cdid+Wbc/ziuH+CxrlWGfGw+cR6sFZLqk4a57tladnh1IuqTiXt/FWBbasUyGmz3JzarUyTHjwHLo2tDPNtIyMiNlz/xayBW0lj0jswkRiQtFfPT1+cd4Dd3u3qJl5fEYut739emDnqOpNfKgb7U7xtlH4IV5qKS9UcPikS8OqjL63K7UqekceP3WpNYx3alCFGQ+fl21giYbq5cOvHML98G/9dG7U+5nPePg8z/K6YTaXclOzQqls+243r10h7LK8aZQrh29CfmRdG1YNqyboWK9y0EZMeaFf++B9qT8Y0IEG1cpmpjRPTTOeqSUChEZRtzsl/7yPcsOqZ3qx7vneBd2MuFLV58z7vguyq4bdmXY/vKk9/7jo9MzXk//RLar7RuONeXHLmr42JItEYIfSe6JUdecFFRw+yUnf7XZbrFmhFI9e3JSzG9kB8MYzbCR6uC9UQjFhwoPnRNyXPFIwYrRbolcslTf34tC34JV+rbyfUzqRG5w+50Sou2fxhGIkhlGJpRvD1/8X3WwuEl/9rVO2/c3rOfExgWenZmREdHtetzM4oOyNa1pnHt/qTBpe7Nsyqvbk1r167qMXMO/RC7j73NMiCtR6cY79KQxc0KQ6ftdU2/Ye4YMbgz3n3LP30iWKBw3up1Qpw2MxWNUADL62rWf56+59Z1y0jrD19PTVNiYnP1RWKjh8EmlQB+99yd+6ri0f39SBBs6XL9zSN/mZnjSoVpaECOqEwEyoTIkEvr8zWC86wyMz5yO9mgS9vqRVLUq60kwMCfPFDTD89jMyjwc4QiDUW6lmBe/VV0aGyTajP6NBFd68pk22uqGBSq3rVgwrOAAqxSC25qTyJbNNBALtCKzsUtMzIgZRhaa+KO3aOOuxi5uy7vnedDv9pNDLPHnpKm8BnBNVyyZRpWwSD110ekSnjKYuwRQ6AA08u0HEQSkaOtTzVreMvrdrnu7rh26nV4t6f+8AJRMTuKBp9aCy0KDf0MnSrVFqEtY+lzsVbh+PHGIAV4f8ntwEkpPOWxd/t2AVHD7JyVPCyz5RNqk45zY+id+dXDYjF20NmoWWLpHAO9e3yxygIgmnQO6gg8fSs+0658WQScHuja/3ax00EPZqUYMSCcXCDlrVy5fM/NE81ad5js9zY0z296tH8xpcEoX95/Qa5TLVRaGEqspCmf/YhWHPNXDNHBMThEbVs2bo7usC6SNy2q+i2+nVuKZj+B8zRHbLdKseY+HVFsmxwj2pefLS4Fijf/dqwo93nZWnZ/dsXjPnSi5a5VFQeVEhF9sJ1/ZQO4dOhro0jN67asi1bfnvZfa3kpfI7YVPdueGzlkr9reva0v/DnWpVi4pV/2MJSo4YsQzlzfnm4GdaVCtbNg6LWpbo+4lrWpxfpOsGc6Spy4K8miKpA5L8JlbaM+h1CAdbuhALiKsfKYnfV3bzQZ44cqW1K1cmjH3nR02U/ALV4ZXw3gZ6wLL6ZyoXKZEWO+tQHE4I2g4VeCiJ7sz+r6s2a8gQYO++7rQqPNws/FjaRk8d0VkVVRihHibt13brOYmEeS3rhUhwGknhf/+ue/fum7FPMcpBOjdoiYfDmjPLV3q89wV2V3Uw9nOXuvXOibPX/Rk98zjsknFPScWoe+Tm2hse2WTijPqnq78EWFSEqBn8xpc3zm7ivb9EHXY6dUj2+nKl0yktGsi2rNFTYoVE2Y+fD7zH7uQ7+88M5vWIZTzGke32vWLCo4o+fSWjp6Da4CzG1ajUw7eF4Fd8/YfSQ2aaYbOStyCI3TnusNhMtJGIpzbcE5c6CzfG1Uvx01neS/NE4uHF3IDz7ZZhd0z2cBMKdIAF8DtzfTr/WdnXrv7oDVI39KlfpD3S06UK5kYtM+3SPgZYSCOI0BgpRJqsP9xQfaNgWpXLMU9552W+bpsUnHevs5bJVgvaAXk/+fYoV70xviyJYO/S36uDdDm5IrZyoZc1zZzIuSlKisbJtdbtAFy5+cw+JUrmTX77tSgsufKLVxfB57dIGo39Ka1yueoIq1QKjHb5CwQ79WpQVYbPrqpfdgV3ic3d8g8PqVydsN5QjEhoZjQ9uRKnlqH4q7v0bs3ZN//PRao4IiScxpV81TnBDxuovFoCuz6tm3vkYiDhFtVdXKIQdP9I4mGxy5umjn/Gnh2g8zyH+48k+c9ZocBkooXi8qO8N6U8Cm7Az9It2GxvzOweBkXn7k8WBW2aNPezONG1ctlerK595qIlFY9JwKfQYvaFbJ5ojSoZtv8f12twAwImF4tarLg8axZZ93K2dUcvw86jwe6nx5U1jOMgMswhgWPX8jr/VtTt3JpXz90Py6cPZvX4OkQVWNuIpsfuDBnl+1occ/0I00khoQI3UiThcSEYpzTKHq1ktsh4dHeTSLUjA73byzAtZ1OZt3zvSlfMpHTq5ejYulEzmtcnVJhVJNum9iZp/p3Bb7FNcnLzWQkGlRw5JHv7ziTZy5vHtUH9PJVrbi208l8f+eZEX+07hnLp7d05J3r2/HDnWfSo1mNqJee5zoz41NcP053Cvc2J1fKHMS9aFWnYlTPiRSdHZB/7q4GbArnNKqWzS30uk6nsO753pnloZst9e9g1UoVY6TfDQj7n//eJduAWLqEjU15pLcVcIHZ4mknlaVi6RKZP87L29hV6P91rZ8rF+EMY6hYukSmMfSiZuFnv7VCJifRbGZ03wUN6dygMq/3b5NNL+5lr3OvGD6+qUO2vWi8nD9Cyc1g9+Ql4fO7ha6YvSYLAbVm8WKSox1y/APnZB6795q/ql1kW1Uk1j3fmx/uPJM7zom8d8/Y+89mwePdPc+9c31bXrk6eHJaLxeuuiUTi/HRTe3DxkHFAhUceeTkKqW5rlN0LqclExN49vIWlC5RPEe33ktb1eLTWzpStWwSPZrXoM3JlXjnhnbc2e1U7rugIcue7gF4J1T85e9d6Ov8CE6vUS5zAI8mPigQsX2vh2+7F6HurG4CNg63qiJUbeE1Yw/QMsSukNkPcu5ItGlbIuEegK5qV4fpg87LVA080rsJT17SlDu7neq8bsqY+86O6r6vXN0qM0C0TInoUvave7430x8+PzNOCOBgFGlALm1Vi6EDz/D8nLyCVN3C+tzGJ/GPixpn2ubAepmddlLZiBOfwOrh0d5NsgWGfuyoYW4JUX16bWHrJvA4rxk9wIi/n8UjvZpkrgy97hfI7eZe3aRFmUXY85l3W1VTYPXc5uRKeUoz0qN5Ta5oG14dHi0ZxnBe4+q8GiMbkhfx2WhCyZHAFzyc3eQND5fVwHX3uSK8bzyjHo//FLyfQfPaFWheuwK9W9qZextnsOscRQRswFW3ZISd4dxEcucNxJVE8iz55e9d2Rtm749Q/W3WXtzh2xNITtg2BpG4Qc8WCQr6TCgmYe0+4Rh1T1e+nLWey1rXpmfzmvTvWNfTMPv7oPN4ZuRSRi3alu1czxY1Mvev6NIwcmqavu3qRHTWKJmYQO2KpdjsSmPutQGU++NLyzCZM/Zw+2AH8Jr9N69VwTMA0UuwlSmRkOn6PXTgGVz97gwublmTX/7cwpa9RyhRvBjj77dtaVyjPI1rZKmdzjq1araVybzHLiQ9ZMV0suv9L1+qOAPOOCVz0pUTLetUzJdgyoe6N6JFBA3AOY2qBe3lnpYPabNVcBQgK/7bI1c7sPmlQ73K/PlE96hc+DID36JQSSx6sntEm4vbq+r6zifzxcwN2epUKJUYvWuhM4KFa9k95zfMDLJsEkFtVDKxWL7twOamaa3yPOMYSkuVSODMU70H/toVS3HXuad5Cg63x1BOTg9eHk6hTPvXuUGCfYPHlqzuod+9z8fga9t4rnoizblDg9OeurQZdSqV8lT1/vF498zVZcf6lTMH6XduaMelg39n8DVtstkAA7x4VctsgsO92h122xn8tGAzZ7rcl0XEt8t5fnD3eZFXz6EhANGoE/OKqqoKkKTiCTHPoHnTmfU8y6MdnANup5FUBy/0bUmdSqXCesuc6hiW3YLjP32ak/xMz6jaEI7SzkDpXnG4VSYPXNgo01mhfIT+TniwW7YUJ4WN8o5Abn9KpSAvG/fnGG7lNfjaNjStWT4q43foanD8sh3Z6riNuO7P/OKWtejnkckgMGM/12WPKxVGyA04sx7nN6nu2dYSxYsFecEFaFmnIkufvojuEexB7udd2yl7GzvWr5wpxI93uobEmBzLg/otWnTFUcQIDezyyz8vOp1eLWoGLftDubp93YgRrIGBxh03ICJhA/oiMfkf3TJTpHQ7vRpdTqsapOe+/8JGvDh2RbYI9ZKJCcx/7EL2HU6l20uTgma6tSqWCptrrLBQt3Jpvvq/TrSpWylo4C5RvBgf3Niev302N+z2wRe3rMXFLXMXo+G1c+Br/drwyfR1NKpeNirnjBZ1squjxt53Nku27A1zhf8U5KVzsA2JCM9e3oIzTq0Ss1xQhZX+Hery8PeLMl9Xi5BbLVao4CgCtKpbMeotQHOieEKxXKeeGHVPVyYs386UlTbILxb7WJziSk1dpWwSX/ytU9D5ck5swilVsw+ilcuUyBwIj8f9x8Opsi5oWp2v/tYpKpuVX7o2rJpt1VGjQkkG9Wycp/ueXKV0WLUS5JzKJzd4rTSKIqGrxmgCGvNKjqoqETlLRMo4x9eLyCsiEp0bkZIv3Oyop/7Z4/TIFeNM01rlufu8hrzavzV3dDuV5rXil/48wDUdT+bR3k34Wxdvb5vAbD0vG/IURs48rWpcNgoKBMqFU3nGi+rlk7j73NN4NUzCTKVwEc2K422glYi0Ah4EPgA+A86JeBUgIj2A14EE4ANjzPMh55Oce7UDdgH9jDHrnHMPA7cC6cA9xpixIlISmAIkOW0fbox5Ioo+FGkua1Oby+Los+2X2hVL8a8eeZuhRktiQjH+1tVbaIC1FUz5x7lRp5w/0QmsEuO0VXVYRISHnNTl5zQ6KSjmSIme3CbK9Es0giPNGGNEpA8w2BjzoYjcmtNFIpIADAEuBDYBc0RkhDFmqavarcBfxpjTRKQ/8D+gn4g0BfoDzYBawHgRaQQcBc4zxhwQkURgmoiMNsbM9NFn5QQjkopECSagpuxUP/ZqsGiJZh8LJZgX+rakTsVSQV5i8SQawbHfmf1fD5wtIsWAaFx0OgLJxpg1ACIyFOgDuAVHH+BJ53g4MFiswq4PMNQYcxRYKyLJQEdjzAwgsLFuovOnUxNFiRHt61VmweMXUrG0Dt7HE5GcVeJBNO64/bAz/VuNMduAOsCLUVxXG9joer3JKfOsY4xJA/YCVSJdKyIJIrIA2AGMM8bM8nq4iAwUkbkiMjclJcWriqIoHqjQUHIiGsGxH3jdGDPVURe1Br6Oa6siYIxJN8a0xgqwjiLiGbFjjHnPGNPeGNO+WrWiZRhVFEUpSKIRHFOAJBGpDfwK3AB8EsV1mwH3+qmOU+ZZR0SKAxWwRvIcrzXG7AEmAj2iaIuiKIoSI6IRHGKMOQRcAbxljLkKiCYufw7QUETqi0gJrLF7REidEcAA57gvMMFYd4oRQH8RSRKR+kBDYLaIVBORigAiUgpreF8eRVsURVGUGBGNcVxE5AzgOqwXFEQhcIwxaSJyNzAW6477kTFmiYg8Dcw1xowAPgQ+d4zfu7HCBafeMKwhPQ24yxiTLiI1gU8dj61iwDBjzC9+OqwoiqLkDcnJX1pEzsHGb/xujPmfiDQA7jPG3JMfDYwF7du3N3Pnzi3oZiiKohxPhI3myXHFYYyZDEwWkbIiUtZxrz1uhIaiKIoSW6JJOdJCRP4AlgBLRWSeiOQtk56iKIpy3BKNcfxd4AFjzCnGmJOxaqv349ssRVEUpbASjeAoY4yZGHhhjJkEFO08xYqiKEpYovGqWiMijwGfO6+vB9bEr0mKoihKYSaaFcctQDXge+A7oCpwczwbpSiKohReovGq+osQLyoR+Qabw0pRFEU5wcjtnuNnxLQViqIoynFDbgWHoiiKcoISVlUlIm3DnSK6/TgURVGUIkgkG8fLEc5pYkFFUZQTlLCCwxhzbn42RFEURTk+UBuHoiiK4gsVHIqiKIovVHAoiqIovggrOETketfxWSHn7o5noxRFUZTCS6QVxwOu4zdDzt0Sh7YoiqIoxwGRBIeEOfZ6rSiKopwgRBIcJsyx12tFURTlBCFSAGBjEVmIXV2c6hzjvG4Q95YpiqIohZJIgqNJvrVCURRFOW6IFDm+3v1aRKoAZwMbjDHz4t0wRVEUpXASyR33FxFp7hzXBBZjvak+F5H78qd5iqIoSmEjknG8vjFmsXN8MzDOGHMJ0Al1x1UURTlhiSQ4Ul3H5wOjAIwx+4GMeDZKURRFKbxEMo5vFJG/A5uAtsAYABEphe7HoSiKcsISacVxK9AMuAnoZ4zZ45R3Bj6Ob7MURVGUwkokr6odwO0e5ROBifFslKIoilJ4ibR17IhIFxpjLo19cxRFUZTCTiQbxxnARuBrYBaan0pRFEUhsuCoAVwIXANcC4wEvjbGLMmPhimKoiiFk7DGcWNMujFmjDFmANYgngxM0r04FEVRTmwirTgQkSSgN3bVUQ94A/gh/s1SFEVRCiuRjOOfAc2xgX9PuaLIFUVRlBOYSHEc1wMNgXuB6SKyz/nbLyL7orm5iPQQkRUikiwigzzOJ4nIN875WSJSz3XuYad8hYhc5JTVFZGJIrJURJaIyL2+eqsoiqLkmUhxHJGESo6ISAIwBGtg3wTMEZERxpilrmq3An8ZY04Tkf7A/4B+ItIU6I8NQKwFjBeRRkAa8KAxZr6IlAPmici4kHsqiqIocSRPwiEHOgLJxpg1xphjwFCgT0idPsCnzvFw4HwREad8qDHmqDFmLdYw39EYs9UYMx8yc2YtA2rHsQ+KoihKCPEUHLWxcSABNpF9kM+sY4xJA/YCVaK51lFrtcHGmGRDRAaKyFwRmZuSkpL7XiiKoihBxFNwxA0RKQt8B9xnjPG0txhj3jPGtDfGtK9WrVr+NlBRFKUIE0/BsRmo63pdxynzrCMixYEKwK5I14pIIlZofGmM+T4uLVcURVHCEk/BMQdoKCL1RaQE1tgdmv9qBDDAOe4LTDDGGKe8v+N1VR/r3TXbsX98CCwzxrwSx7YriqIoYYgYAJgXjDFpTpT5WCAB+MgYs0REngbmGmNGYIXA5yKSDOzGChecesOApVhPqruMMeki0gW4AVgkIgucR/3bGDMqXv1QFEVRghE7wS/atG/f3sydO7egm6EoinI8ETax7XFpHFcURVEKDhUciqIoii9UcCiKoii+UMGhKIqi+EIFh6IoiuILFRyKoiiKL1RwKIqiKL5QwaEoiqL4QgWHoiiK4gsVHIqiKIovVHAoiqIovlDBoSiKovhCBYeiKIriCxUciqIoii9UcCiKoii+UMGhKIqi+EIFh6IoiuILFRyKoiiKL1RwKIqiKL5QwaEoiqL4QgWHoiiK4gsVHIqiKIovVHAoiqIovlDBoSiKovhCBYeiKIriCxUciqIoii9UcCiKoii+UMGhKIqi+EIFh6IoiuILFRyKoiiKL1RwKIqiKL5QwaEoOZGRAccOFXQrFKXQoIJDUXJi/OPwZltIPVLQLVGUQoEKDkWJxIEdMPt92L8Vlv9S0K1RlEJBXAWHiPQQkRUikiwigzzOJ4nIN875WSJSz3XuYad8hYhc5Cr/SER2iMjieLZdUQCY+RakHYUyJ8H8Twu6NYpSKIib4BCRBGAI0BNoClwjIk1Dqt0K/GWMOQ14Ffifc21ToD/QDOgBvOXcD+ATp0xR4svhPTD7A2h2OXQaCGunwO41Bd0qRSlw4rni6AgkG2PWGGOOAUOBPiF1+gCBadxw4HwREad8qDHmqDFmLZDs3A9jzBRgdxzbrSiW2e/Dsf3Q9QFofR1IMfjji4JulaIUOPEUHLWBja7Xm5wyzzrGmDRgL1AlymsjIiIDRWSuiMxNSUnx2XTlhOfYQaumangR1GgB5WtBw+7wx5eQnlbQrVOUAqXIGseNMe8ZY9obY9pXq1atoJujHG/M+wQO74azH8oqa3sjHNgGyeMKrFmKUhiIp+DYDNR1va7jlHnWEZHiQAVgV5TXKkow2xbB+KfgyN683SftKEx/E+p1hbods8obdneM5J/l7f6KcpwTT8ExB2goIvVFpATW2D0ipM4IYIBz3BeYYIwxTnl/x+uqPtAQmB3HtirHM0cPwNhH4N1zYNorMPLBvN3vz6+t+23XkPskJELra2HlWNi3NW/PAFg9AWa9C+umwSE12ynHD8XjdWNjTJqI3A2MBRKAj4wxS0TkaWCuMWYE8CHwuYgkYw3e/Z1rl4jIMGApkAbcZYxJBxCRr4FuQFUR2QQ8YYz5MF79UKIkIwNG/9PO0FtenX/PXT4SRv0T9m2CtgOgZAWY/gacdiG06uf/fulpMO1VqNUWGnTLfr7tjfD7a/DnV9kFix/2bYGvr4W0w1ll5WpB9WZQvSlUbw4nNYWqjaB4idw/R1HigNgJftGmffv2Zu7cuQXdjKLNvE/g53shIQlumwInNY7v8/ZsgNH/ghWj4KRmcPGrcHInO/B/0hu2L4E7pkGlev7uu/Bb+P5v0O9LaHKxd52Pe8O+zfD3+VAsl4v2H26Hxd/BzaOt2++OJbbN25dCynLISLX1ihW3wqN6M2h6GTTuDSK5e6ai+CPsF00Fh5J3DuyAwe3tALdrNVQ8Gf423qp2Yk16qvV2mvS8fd3tYeh8R/Cz/loP73SBao3twJwQ5cI6IwPePtMe3zE9vFD48xv4YSAM+Bnqn+2/D5vnw/vnwln3wYVPZT+fngq7kh1B4vxtXQAHtkPtdnD+496rIUWJLWEFR5H1qlLykTEPQ+phuOxtuOQ1O8hNeSn2z9kwy9oxxjkD512z4ax7sguoSqfYFcim2TDlxejvv3I0pCyzcRuRVhJNL4WkCrkzkhsDY/8NpauGV3UlJMJJTaBFX7jgCbhuGNy/FC4dDPu3w2d94NNLYfM8/89X8oejB2DWe3CgaIYCqOBQ8saq8bB4uB0EqzaEpn2gxdV2wN48PzbPOLQbRtwDH3W3HlP9v4JrvoaKdcNf06IvtOwHU16ADTNzfoYxVthVqgfNrohcN7GUteMsHeHfqL30J9gwA857BEqWj/66hOLQ9gb4+zy46DnYvhjePw+GXgc7lvtrgxJfDqTApxfD6H/Au11h/fSCblHMUcGh5J5jh2DkA1ZF1eX+rPJeL0DZ6laPn3o4/PXRsGUBDOloI7bP/DvcNcvq+aOh10tQoS58/385u+iumQRb5lv1UTSqrbY3QvpRWPRtdG0B6+Y77nFr9G5zY/TXuUksCWfcCff+adV0aybD22fAj3dau49SsOxeayc4O5ZDzxcgsTR8cjFMfcWqQosIKjiU3DP5edizHi5+DYonZZWXqgR9BsPOFfDbf3J//61/WrVM8VJw22To/l9IKhv99SXLw5UfwN7NMPKhyHWnvgzlalp322io2RJqtoZ5n9rVSjTMese+Xxc9E73dJRxJ5aDbICtAOt8Ji4bDm+2sw0ARVY8UerYsgA+7w+G/YMAI6HQbDJxkV+G/PQVf9ysybtcqOE5EMjKiH+zCsW0RTB8MbW6AemdlP3/a+dD+VmvIXjctd/f/rI8dIG/6xab9yA11O8I5/4JFw2DhMO86G2fDuql2ReMWgDnR9kbrDbUlCpXcgRSrCmt4EZx6XvTPyIkyVawgumc+tOpv82u93sqqEJX8Y/VE681XPAluGZsVOFqyPPT9CHq/bFe173SxtrrjHBUchZm0Y7B1oVVx5BVjYNNcGD0IXmlsZ6cpK3J3r4x063pbqhJc+HT4et3/Y20GP94BR/dHf/9ti63xN7GMFRqVTsldOwN0fRDqdoZfHoC/1mU/P/VlKFXZxoH4oUVfuxqKxkg+6Vmb/6p7HlZgkahQBy5906ryKp4MP99jDbSFmbRjNgDyl/th7dS8T2YKikXD4cur7Pt+669Q7fTg8yLQ4W9w6zjr+PBJL5uZ4HjtLyo4Ci87V8EH51vj2v/qw5dXw8x3bLmfL9yO5VZd9EZre7+5H0GdDnYg/7B77gx3cz60Hj09noPSlcPXK1EGLn8H9m6ynkTRsH0pfHapNUDf9LP/OAwvEorDFe/ZH/D3A4OTFG5bBCvHWJdeP2owsMGGzS63A0ekQXrHMhvn0uHW7INKrKnaEC553caZTH4+vs/KLcbA8lHwVmcbNPrHl9aY/EYbuyrbt6WgWxg9M96C7261K4ybR9tkmOGo1drGOJ3eE359FIZea9VaxyEaxxELjh2CYTfCGXfBqefm7V7G2BnsmEFQvKRVs+xKhtW/Ze0FUaGufc6p50ODc+zM382eDTa4bNFw630jxaz7avO+NqitZAU78/6ir9W5X/4uNM/BkyjAvi0wuCPU7QDXfx9dMNq4J2y09bXDoNFF4evtWGYNiQkl7EqjyqnRtSlaAsF93R629gGAb2+GVePg/kXZ38doWD8DPu4BfYZAm+u963x+BWyeC/csiCxoY8mIv9sB+bYpUKN5/jwzGrYutJOIdVOtU0X3Z6BeF1g2AuZ/Duun2e/raRdaL7JGPeITD5RXMjJg/BM2S0GTS+CKD6zjQjQYY1davz5q7WpXfQJ12mWvd3gP7F4Nu9Y4/1fbscBk2HHh9J7xDgbVAMC4Co4dy+Grq+0gfHova8TNzaB3aLdVAS0bAfXPsQN6+ZpZ53evhTUTIfk3u6nQ0X32R1a7ndWbl6oMS36AjY77aZ2OVp3S7HIoe5L3876+xtbv/owVfDl9Eb+53g60d86Ayg2i61faUXjvXDi0E+6c6T147lhuZ52SADePir3QCPD9QOsJdfMYKF3FBi6eda93IF40GAODO9g+3fpr9vOrxsGXfeGiZ+37m18c2m37VuU029fcRrjHin1bYeJ/rTArVQnO/Te0uym7UNi12nrQLfjKZiIuU826Vbe9Mf6rtWhJT4Wf7oaFQ60dr9eLUCwh5+tC2TQPvr0pKy9aQqKdHO5Ktu/DoZ2uymLVkVVOtc4eu1bZ33yP5+P5vqjgiHvkeOoRmPW2XWqnHbUeFWf/A0pVjO76ddPsoHZgu40MPuPvkX/s6Wl2Frt6gv3bPM/ORE5qaoVF8yujU/OkHrFR0Et/gk632wEu3I9g+SgYeg2c/4QNkvPD1oU27qBxbzvDcguolJXWsCgCN4206pZ4cWSvNVCCFbgrRsN9i7wFa7T8/gaMe8wGJLp/xOmp8PZZkJFmBWZ+55z640v46U5r+2ibS/ffvHLsEMwYDNNeg/Rj0Pl26PpQzr+L9DS7yp7/mVUlZqTZidAZd9rUK7GaaW+cAyPutt+L8rXt4FyhjnPsvC5fxwqwYsWsSnLYjbZt5z5q0+7npS2HdltX6pWj7etyNaHyqVClgRX6lU+1wqJS/awVTXqqdYKY9DykHoSOt0G3f1lNQmxRwZFvKUf2b4cJ/7GzptKV4dxHrNE1nPtleipM/p8VOJXrw5UfQu22/p97aLfVl+Zmpp6RYZfNM4dA44utC2tiqeA6R/fDkE72y3nblNypD6a8ZN+bKz+0wg2szeaT3nbmftNIqNbI/339smEWfNwTTDp0HGhnjHnhQIp1OOh0u/VwCjD7fRj1kA1YjDb2JJYYAx/3stHwd8+zHlj5RUaG9WQb/xTs32JdUi94MvpVqpsDO2DhN1aI7FxpV9C9X8mb2i8jw6qZJvzH2iXqdbW2uH2b7YzenXwSrPq0XE07Odu32bqgt/PpTBEOY6zquEw1f3a2Aym2/fM/s6vnC56A1tfHcnWpgiPfc1Vt/dOm4lj/u10F9Hgue36h3WttcNqmOfYD7/k//wbaWDLjLat/rtMBrhkaPNCMHmTjEG79NXiPCj+kp1l7wM5VdgZ+7KAjNNIdoZGPqoipL1vPltumWG+YvPLNDfazfmC5XVkc3mONvdWb2ZxWBZWYcMcyu8Jq1d/aYfLlmcvhx9thyx9Qq42NdD/ljLzfNyMdfn8dJj5rB8rLhsBpF/i/z4EU+OE2u2po2gcueSN4BWSMnYjt22SFyN5NWceHdtkJwuk98t6fWLHlDxu/s3GWjS3q+YJN+Jl3VHAUSJJDY6y94tfHsts/Fg6z7qFSDC551aqWCgNLfrQqswp14Prv7Cpo83zrkdXuZrj4lbzdf2eyHchqt7X63PRUawg/qUlMmu+LtGOxUx+tGg9fXglXfQrNLrP7g8wYYgVTzZaxeUZuGfe4HXBvGQsnd47vs7Yttl5xkmC/6y2uir19Zeuf9juashw6/J91CS9ROrpr10yy1x7Zaydz7W4uGtmGjbHOMOMeszaTlv3ggqeCbaT+UcFRoNlxQ+0fdTrAhuk2tuDK92Mz440l62dYW0ax4lbNMvIBO0u7e3Zs9Kiz3rN5fEpXgQG/2P0njncy0uG1lnbV1OtFq9Zr1S//ZvmROHbQtiepXO7VjNGwdaET6V8yPl5xblKPWDXNjMHWFnD5e96eSQHS02DSc3alWbWhtbNVbxa/9hUURw/YzcymvwnFEuHsB629NHcTJBUchSKtesD+sXCY9aLo+mDeU0/Ei5SVdga9ZyNgsmbSsSAjA+a8b1V3hcVTJhZMfM7aq07ubAfRe+ZDuRoF3SrL8pE2buDC/9iMwrFmywL4/DInaPPn3NkycsOayda4vH8rnPPPLO8kN3s3wfBbrfdgm+utKqdEmfxpX0Gxe621W+5Khtun5XayoIKjUAiOABnpuXPfy2/2b4dvB1i11RXvF40lfTzZs8GuOjBw3qPWq64w8VV/68Z992z7mcaKLX846WHKW3tO5fqxu3c0HN5jAwkXfmM95S5/D6qeZs8tH2kFS0aaNWi3vCp/21bQHNmbFy2BCo5CJTiUosuXV1uD9N2zs3umFTR7NliV1annQf8vY3PPzfPgs8uhVAWrdsxrepi8sOQHm74k9Yi1e+xebR06araCvh/HV3VWNFHBoYJDyReOHrDxCvkVIe6Xaa/C+CdzjuKPhk1z4fPLbV8H/BJ5f5T8Yt9W+Oku6zEF0OkOG9zpJ3mlEkAFhwoORcF6kr3bFVIPwZ2zovdGCmXDLPjiSihT1RrCY6n6yivGWLVVmaq5c9dVAujWsYqiYL1rer9i1VZTc7m97/oZ8MUVNtr+5lGFS2iAtcW16q9CI46o4FCUE416Z0Gra22qFL+p9df9blca5WraoM1I2WCVIosKDkU5Een+H+uSOvJB2L/NRkofPWADMsOpr9dOtQkbK9S26qm8BZcpxzGFNIhAUZS4UqaqzR31y33wcmgsjdjcTMWTrP9/QpJVce3bal1tB/yct6SQynGPCg5FOVFpd5NVOe3fYo3m6Uez/qcfy15WoozNCFu2WkG3XClgVHAoyomKSOFK1qccN6iNQ1EURfGFCg5FURTFFyo4FEVRFF+o4FAURVF8oYJDURRF8YUKDkVRFMUXKjgURVEUX6jgUBRFUXxxQqRVF5EUYH0uL68K7Ixhcwob2r/jG+3f8U1h7t9OY4xnhOgJITjygojMNca0L+h2xAvt3/GN9u/45njtn6qqFEVRFF+o4FAURVF8oYIjZ94r6AbEGe3f8Y327/jmuOyf2jgURVEUX+iKQ1EURfGFCg5FURTFFyo4wiAiPURkhYgki8iggm6PH0RknYgsEpEFIjLXKassIuNEZJXzv5JTLiLyhtPPhSLS1nWfAU79VSIyoAD785GI7BCRxa6ymPVHRNo571eyc60Ugv49KSKbnc9wgYj0cp172GnrChG5yFXu+Z0VkfoiMssp/0ZESuRf70BE6orIRBFZKiJLRORep7xIfIYR+ldkPsNsGGP0L+QPSABWAw2AEsCfQNOCbpeP9q8DqoaUvQAMco4HAf9zjnsBowEBOgOznPLKwBrnfyXnuFIB9edsoC2wOB79AWY7dcW5tmch6N+TwEMedZs638ckoL7zPU2I9J0FhgH9neN3gDvyuX81gbbOcTlgpdOPIvEZRuhfkfkMQ/90xeFNRyDZGLPGGHMMGAr0KeA25ZU+wKfO8afAZa7yz4xlJlBRRGoCFwHjjDG7jTF/AeOAAtln1BgzBdgdUhyT/jjnyhtjZhr7q/zMda98IUz/wtEHGGqMOWqMWQskY7+vnt9ZZ+Z9HjDcud79XuULxpitxpj5zvF+YBlQmyLyGUboXziOu88wFBUc3tQGNrpebyLyF6GwYYBfRWSeiAx0yqobY7Y6x9uA6s5xuL4W9vcgVv2p7RyHlhcG7nZUNR8F1Dj4718VYI8xJi2kvEAQkXpAG2AWRfAzDOkfFMHPEFRwFFW6GGPaAj2Bu0TkbPdJZ1ZWZPywi1p/HN4GTgVaA1uBlwu0NTFARMoC3wH3GWP2uc8Vhc/Qo39F7jMMoILDm81AXdfrOk7ZcYExZrPzfwfwA3YJvN1Z0uP83+FUD9fXwv4exKo/m53j0PICxRiz3RiTbozJAN7Hfobgv3+7sKqe4iHl+YqIJGIH1S+NMd87xUXmM/TqX1H7DN2o4PBmDtDQ8WQoAfQHRhRwm6JCRMqISLnAMdAdWIxtf8ALZQDwk3M8ArjR8WTpDOx11Adjge4iUslZYnd3ygoLMemPc26fiHR2dMk3uu5VYAQGVIfLsZ8h2P71F5EkEakPNMQahj2/s85MfiLQ17ne/V7lC877+iGwzBjziutUkfgMw/WvKH2G2ShIy3xh/sN6dqzEejk8UtDt8dHuBlhvjD+BJYG2Y/WkvwGrgPFAZadcgCFOPxcB7V33ugVruEsGbi7APn2NXeqnYvW7t8ayP0B77I96NTAYJ6NCAffvc6f9C7EDTU1X/Uectq7A5T0U7jvrfCdmO/3+FkjK5/51waqhFgILnL9eReUzjNC/IvMZhv5pyhFFURTFF6qqUhRFUXyhgkNRFEXxhQoORVEUxRcqOBRFURRfqOBQFEVRfKGCQ1F8IiLpTrbTP0VkvoicmUP9iiJyZxT3nSQi7WPXUkWJDyo4FMU/h40xrY0xrYCHgedyqF8RyFFwKMrxggoORckb5YG/wOYqEpHfnFXIIhEJZFR+HjjVWaW86NT9l1PnTxF53nW/q0RktoisFJGuTt0EEXlRROY4CfNuc8prisgU576LA/UVJd4Uz7mKoighlBKRBUBJ7F4M5znlR4DLjTH7RKQqMFNERmD3mmhujGkNICI9sam1OxljDolIZde9ixtjOord9OcJ4AJsJPleY0wHEUkCfheRX4ErsCk3nhGRBKB0nPutKIAKDkXJDYddQuAM4DMRaY5NlfGsk404A5v6urrH9RcAHxtjDgEYY9x7cQQSAM4D6jnH3YGWIhLIVVQBm99oDvCRk2DvR2PMgpj0TlFyQAWHouQBY8wMZ3VRDZtnqBrQzhiTKiLrsKsSPxx1/qeT9fsU4O/GmGxJJh0h1Rv4REReMcZ8lotuKIov1MahKHlARBpjt/zchV0J7HCExrnAKU61/dgtRQOMA24WkdLOPdyqKi/GAnc4KwtEpJGTBfkUYLsx5n3gA+z2s4oSd3TFoSj+Cdg4wK4GBhhj0kXkS+BnEVkEzAWWAxhjdonI7yKyGBhtjPmHiLQG5orIMWAU8O8Iz/sAq7aa76TwTsFuHdoN+IeIpAIHsOnEFSXuaHZcRVEUxReqqlIURVF8oYJDURRF8YUKDkVRFMUXKjgURVEUX6jgUBRFUXyhgkNRFEXxhQoORVEUxRf/D8FtcW6gO5hNAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Load and process data\n",
    "\n",
    "if not os.path.exists(\"rnavec_encoded_aggregated_u_t_swapped_rnas_DenseSAE.pt\"):\n",
    "    rna_dense_embedded = torch.load(\"../Data/EncodedData/rnavec_encoded_aggregated_u_t_swapped_rnas.pt\")\n",
    "\n",
    "    train, val = train_test_split(np.array(rna_dense_embedded), random_state = 255)\n",
    "    train = SAEDataset(torch.tensor(train))\n",
    "    val   = SAEDataset(torch.tensor(val))\n",
    "\n",
    "    # Initiate learner\n",
    "    loaders = DataLoaders.from_dsets(train, val,\n",
    "                                     bs=BATCHSIZE,\n",
    "                                     device=\"cuda:0\")\n",
    "\n",
    "    ###################################################\n",
    "    # Train SAE & find and record min loss\n",
    "    ###################################################\n",
    "    rna_dense_learner = trainer.train_sae(loaders, rna_sae, dense_sae_splitter,\n",
    "                                          learner_cbs = [EarlyStoppingCallback(patience = 10)])\n",
    "    \n",
    "    torch.save(rna_dense_learner.model, \"rnavec_encoded_aggregated_u_t_swapped_rnas_DenseSAE.pt\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Train hyperparameter optimized ProtBERT SAE"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Load model params\n",
    "prot_exp_csv = \"../nni/experiment_results/ProtBERTDenseSAE.csv\"\n",
    "# d.sort_values('final_metric')\n",
    "\n",
    "prot_params = find_optimal_hyperparams_from_experiment(prot_exp_csv)\n",
    "\n",
    "# Best params with multiple layers\n",
    "BATCHSIZE     = prot_params.pop(\"batchsize\")\n",
    "HIDDEN_LAYERS = eval(prot_params.pop(\"hidden_layers\"))\n",
    "ACTIVATION    = eval(prot_params.pop(\"activation\"))\n",
    "DROPOUT       = prot_params.pop(\"dropout\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Initiate optimized model\n",
    "prot_sae = DenseSAE(1024, hidden_layers = HIDDEN_LAYERS,\n",
    "                    activation = ACTIVATION,\n",
    "                    dropout = DROPOUT)\n",
    "\n",
    "# Early stopping callback will cut off training when loss stabalizes\n",
    "trainer = SAETrainer(9999, 255, plot = True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "text/html": [],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: left;\">\n",
       "      <th>epoch</th>\n",
       "      <th>train_loss</th>\n",
       "      <th>valid_loss</th>\n",
       "      <th>time</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>0</td>\n",
       "      <td>0.047636</td>\n",
       "      <td>0.003239</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>0.020272</td>\n",
       "      <td>0.001582</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>2</td>\n",
       "      <td>0.010302</td>\n",
       "      <td>0.001289</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>3</td>\n",
       "      <td>0.006302</td>\n",
       "      <td>0.000913</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>4</td>\n",
       "      <td>0.005295</td>\n",
       "      <td>0.000860</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>5</td>\n",
       "      <td>0.004486</td>\n",
       "      <td>0.000660</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>6</td>\n",
       "      <td>0.004158</td>\n",
       "      <td>0.000735</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>7</td>\n",
       "      <td>0.004274</td>\n",
       "      <td>0.000756</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>8</td>\n",
       "      <td>0.004433</td>\n",
       "      <td>0.000643</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>9</td>\n",
       "      <td>0.004247</td>\n",
       "      <td>0.000587</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>10</td>\n",
       "      <td>0.003828</td>\n",
       "      <td>0.000562</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>11</td>\n",
       "      <td>0.004154</td>\n",
       "      <td>0.000732</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12</td>\n",
       "      <td>0.003801</td>\n",
       "      <td>0.000527</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>13</td>\n",
       "      <td>0.003836</td>\n",
       "      <td>0.000425</td>\n",
       "      <td>00:06</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>14</td>\n",
       "      <td>0.003869</td>\n",
       "      <td>0.000570</td>\n",
       "      <td>00:06</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>15</td>\n",
       "      <td>0.003705</td>\n",
       "      <td>0.000490</td>\n",
       "      <td>00:06</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>16</td>\n",
       "      <td>0.004118</td>\n",
       "      <td>0.000595</td>\n",
       "      <td>00:06</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>17</td>\n",
       "      <td>0.003650</td>\n",
       "      <td>0.000500</td>\n",
       "      <td>00:06</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>18</td>\n",
       "      <td>0.003501</td>\n",
       "      <td>0.000574</td>\n",
       "      <td>00:06</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>19</td>\n",
       "      <td>0.003590</td>\n",
       "      <td>0.000507</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>20</td>\n",
       "      <td>0.003459</td>\n",
       "      <td>0.000390</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>21</td>\n",
       "      <td>0.003744</td>\n",
       "      <td>0.000602</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>22</td>\n",
       "      <td>0.003332</td>\n",
       "      <td>0.000521</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>23</td>\n",
       "      <td>0.003751</td>\n",
       "      <td>0.000465</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>24</td>\n",
       "      <td>0.003524</td>\n",
       "      <td>0.000403</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>25</td>\n",
       "      <td>0.003463</td>\n",
       "      <td>0.000530</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>26</td>\n",
       "      <td>0.003721</td>\n",
       "      <td>0.000461</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>27</td>\n",
       "      <td>0.003554</td>\n",
       "      <td>0.000452</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>28</td>\n",
       "      <td>0.003659</td>\n",
       "      <td>0.000457</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>29</td>\n",
       "      <td>0.003731</td>\n",
       "      <td>0.000428</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>30</td>\n",
       "      <td>0.003836</td>\n",
       "      <td>0.000364</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>31</td>\n",
       "      <td>0.003637</td>\n",
       "      <td>0.000559</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>32</td>\n",
       "      <td>0.003645</td>\n",
       "      <td>0.000459</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>33</td>\n",
       "      <td>0.003335</td>\n",
       "      <td>0.000613</td>\n",
       "      <td>00:06</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>34</td>\n",
       "      <td>0.003517</td>\n",
       "      <td>0.000376</td>\n",
       "      <td>00:06</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>35</td>\n",
       "      <td>0.003405</td>\n",
       "      <td>0.000350</td>\n",
       "      <td>00:06</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>36</td>\n",
       "      <td>0.003867</td>\n",
       "      <td>0.000458</td>\n",
       "      <td>00:06</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>37</td>\n",
       "      <td>0.003710</td>\n",
       "      <td>0.000341</td>\n",
       "      <td>00:06</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>38</td>\n",
       "      <td>0.003933</td>\n",
       "      <td>0.000681</td>\n",
       "      <td>00:06</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>39</td>\n",
       "      <td>0.003276</td>\n",
       "      <td>0.000568</td>\n",
       "      <td>00:06</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>40</td>\n",
       "      <td>0.003786</td>\n",
       "      <td>0.000476</td>\n",
       "      <td>00:06</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>41</td>\n",
       "      <td>0.003323</td>\n",
       "      <td>0.000518</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>42</td>\n",
       "      <td>0.003508</td>\n",
       "      <td>0.000441</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>43</td>\n",
       "      <td>0.003534</td>\n",
       "      <td>0.000489</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>44</td>\n",
       "      <td>0.003969</td>\n",
       "      <td>0.000413</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>45</td>\n",
       "      <td>0.003606</td>\n",
       "      <td>0.000431</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>46</td>\n",
       "      <td>0.003802</td>\n",
       "      <td>0.000498</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>47</td>\n",
       "      <td>0.003431</td>\n",
       "      <td>0.000552</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "No improvement since epoch 37: early stopping\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYIAAAEWCAYAAABrDZDcAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjQuMSwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy/Z1A+gAAAACXBIWXMAAAsTAAALEwEAmpwYAAAqoElEQVR4nO3deZhcZZn38e9dS2/pdNJZyB4SmAgJgYEQEFwwCo4JKLiAwIgiOoMbA26jOPO+LryzMDrjjIy4gDKCC5GJolGW4AKCyhYQQwKBhCxkISRk7Syd7qpzv3+cpzunOtWdTtLVleT8PtdVV5393OdU1bnreZ6zmLsjIiLplal2ACIiUl1KBCIiKadEICKSckoEIiIpp0QgIpJySgQiIimnRCCHJDN7v5n9vtpxVIOZbTezY/p5nV80sx/05zrl0KFEICXM7HVm9kcz22pmm8zsD2Z2WpdpGsPB6p4y868ws11hfMfr692s64tm1h6m2RLWe2altm1/mdkbzez+sC9W9DDd6xPbusPMvMv2j9+f9bp7o7sv289Yzwjrbiwz7k9mdtX+LK/L/DPMbPWBzi+HPiUC6WRmTcAvgf8GhgBjgC8Bu7tM+q4w7M1mNrLMot4WDmYdr54OQj9290ZgOPB74KdmZge7LX1kB3AL8Pc9TeTuD3VsK3BCGDw4sf0vdkxrZrlKBOrujwCrgQuTw81sKjAFuL0S65UjgxKBJL0KwN1vd/eiu+9y9/vcfUGX6S4HvgUsAC7rixW7eztwKzASGNox3Mz+3cw2m9lyM5uVGH6FmT1rZi1mtszMPpQYN8zMfhlKGZvM7CEzy4Rxo83sJ2a2ISzz6h5ieszdvw/s17/zpFDqmWNmPzCzbcD7zex0M3s4xPeSmX3dzGoS87iZ/UXo/p6Z3Whmd4VtfdTMju1mdbcC7+sy7H3A3e6+0cy+ZmarzGybmT1hZq8/0O1KxDrZzB4I27LIzM5PjDvXzJ4Jca8xs0+H4d1+PlId2vmS9DxQNLNbzWyWmTV3ncDMjgZmAD8Mr64HngNiZrXA+4FV7v5KGPxq4DlgGPBl4LuJ0sJ64K1AE3AF8J9mNi2M+xTxv+PhwAjgHwAPB5tfAH8mLu2cDXzczN7SF9vQgwuAOcBg4n1WBD4RtuvMEMdHe5j/EuKSWTOwFPjnbqb7PnCWmY0DCNv718QJAuBx4GTi0t6PgP81s7oD3CbMLE+8P+8DjgL+DvihmR0XJvku8CF3HwhMBX4bhpf9fA40Djl4SgTSyd23Aa8j/lHeDGwws7lmNiIx2XuBBe7+DDAbOMHMTumyqJ+Ff3sdr7/tYbXvNrMtwCrgVOAdiXEr3f1mdy8SH8xGER84cPe73P0Fj/2O+GDU8Q+3PUx7tLu3h6obB04Dhrv7de7eFurhbyY+0FbSw+7+M3ePQinrCXd/xN0L7r4C+Dbwhh7mvzOUTgrEieTkchO5+yrgAeLPCOIEUwvcFcb/wN03hvX+Rxh3XLll9dIZQCNwfdifvyWuWrw0jG8HpphZk7tvdvcnE8PLfT5SJUoEUsLdn3X397v7WOJ/caOB/0pM8j7igxHuvgb4HXFVUdLb3X1w4nVzD6u8I0xzlLu/yd2fSIxbl4hrZ+hsBAgllkdC1cIW4Fzif9gAXyH+53xfqDa6Ngw/GhidTFLE/0aTia4SViV7zOxVoWpkXagu+pdE7OWsS3TvZM8++IdEg/S3wvhb2ZMI3gvMDtVumNmnQ3Xa1rDtg/ax3n0ZTVyCixLDVhKXtiBuSzoXWGlmv7M9JwJ09/lIlSgRSLfcfTHwPeKEgJm9BpgEfC4cxNYRV9/8daUaQcsJ1Ug/Af4dGOHug4G7AQtxt7j7p9z9GOB84JNmdjbxAXl5lyQ10N3PrXDIXf/tfhNYDExy9ybiZLTfDeTu/i+JBukPh8E/Bcaa2RuBdxKqhUJ7wGeAdwPNYZ9tPZD1JqwFxnWp3x8PrAnxPe7uFxBXG/0MuCMM7+7zkSpRIpBOZna8mX3KzMaG/nHExfxHwiSXA78iPgvl5PCaCtQDs7our4JqiKs1NgCF0Ij8Vx0jzeytZvYXoT1hK3GdfAQ8BrSY2WfNrN7MsmY21bqcHptYTibUoefjXqtLNuoehIHANmC7mR0PfKQPlgmAu+8gbo/4H+KqtfmJdRaI91nOzD5P3L7Sa2H7O1/E+3Mn8Bkzy5vZDOBtwGwzqzGz95jZoFAi2Ub8GfT0+UiVKBFIUgvxP/xHzWwHcQJYCHwq/PDfDfy3u69LvJYTN1Imq4d+YaXn0d/Zl0G6ewtwNfE/zM3EDaJzE5NMAn4NbAceBr7h7veHtoa3Eiew5cArwHeIq0jKOQvYRVzaGB+67+uDTfh0iLmFuI3ix32wzKRbiavBbksMmwfcS3xCwEqglS5VVvswhnj7k69xxAf+WcT78hvA+0JJEuKqqRWh+uvDwHvC8LKfz/5tovQlUxuNiEi6qUQgIpJySgQiIilX0URgZjPN7DkzW1ruFDGLbyy2wcyeCq+/qWQ8IiKyt4qd8mdmWeBG4M3EVxE+bmZzw4VIST/ex71oRESkgip57vfpwNKOuyia2WziS+27JoL9MnPmTL/33nv7IDwRkVTp9pqRSlYNjaH09LTV7LniMOldZrbA4htzjSu3IDO70szmm9n8Z599thKxioikVrUbi38BTHD3k4gvVLq13ETufpO7T3f36cOHD+/XAEVEjnSVTARriC846TA2DOsUboDVca/77xDfdExERPpRJRPB48AkM5sYLsu/hNKrPzGzUYne8wHV+4iI9LOKNRa7e8Hix+PNA7LALe6+yMyuA+a7+1zgaosfZFEANhHfj15EpE+1t7ezevVqWltbqx1KxdXV1TF27Fjy+Xyv5znsbjExffp0nz9//r4nFBEJli9fzsCBAxk6dCh2yDwJte+5Oxs3bqSlpYWJEyd2HV2Vs4ZERA4Jra2tR3wSADAzhg4dut8lHyUCEUmFIz0JdDiQ7UxNInh8xSa+et9ztBV023MRkaTUJIInVm7mht8upRApEYhI/9qyZQvf+MY39nu+c889ly1btvR9QF2kJhF0FJYOs7ZxETkCdJcICoVCj/PdfffdDB48uEJR7dFvz5mttpRUD4rIIejaa6/lhRde4OSTTyafz1NXV0dzczOLFy/m+eef5+1vfzurVq2itbWVa665hiuvvBKACRMmMH/+fLZv386sWbN43etexx//+EfGjBnDz3/+c+rr6/skvtQkgg4qEIik25d+sYhn1m7r02VOGd3EF952Qrfjr7/+ehYuXMhTTz3FAw88wHnnncfChQs7T/G85ZZbGDJkCLt27eK0007jXe96F0OHDi1ZxpIlS7j99tu5+eabefe7381PfvITLrvssj6JPzWJwELl0OF23YSIHHlOP/30kvP8b7jhBu68M36096pVq1iyZMleiWDixImcfPLJAJx66qmsWLGiz+JJTyIIVUNKAyLp1tM/9/4yYMCAzu4HHniAX//61zz88MM0NDQwY8aMstcB1NbWdnZns1l27drVZ/GkprG4gwoEItLfBg4cSEtLS9lxW7dupbm5mYaGBhYvXswjjzzSz9GlqkSgIoGIVMfQoUN57Wtfy9SpU6mvr2fEiBGd42bOnMm3vvUtJk+ezHHHHccZZ5zR7/GlJxFUOwARSbUf/ehHZYfX1tZyzz33lB3X0Q4wbNgwFi5c2Dn805/+dJ/Glr6qIRUJRERKpCYRdNYMKQ+IiJRITyII78oDIiKl0pMITNcRiIiUk6JEEL8rDYiIlEpPIqh2ACIih6jUJIIOqhkSkUNdY2MjAGvXruXCCy8sO82MGTPoq8f2picRdLQRqHJIRA4To0ePZs6cORVfT2oSQWfVkPKAiPSza6+9lhtvvLGz/4tf/CL/9E//xNlnn820adM48cQT+fnPf77XfCtWrGDq1KkA7Nq1i0suuYTJkyfzjne8o0/vNZSeK4vVWCwiAPdcC+ue7ttljjwRZl3f7eiLL76Yj3/843zsYx8D4I477mDevHlcffXVNDU18corr3DGGWdw/vnnd/vM4W9+85s0NDTw7LPPsmDBAqZNm9Zn4acnEai5WESq5JRTTmH9+vWsXbuWDRs20NzczMiRI/nEJz7Bgw8+SCaTYc2aNbz88suMHDmy7DIefPBBrr76agBOOukkTjrppD6LLzWJoIMai0VSrod/7pV00UUXMWfOHNatW8fFF1/MD3/4QzZs2MATTzxBPp9nwoQJZW8/3R/S00bQWTWkTCAi/e/iiy9m9uzZzJkzh4suuoitW7dy1FFHkc/nuf/++1m5cmWP85911lmdN65buHAhCxYs6LPYUlMi0MPrRaSaTjjhBFpaWhgzZgyjRo3iPe95D29729s48cQTmT59Oscff3yP83/kIx/hiiuuYPLkyUyePJlTTz21z2JLTyJQY7GIVNnTT+9ppB42bBgPP/xw2em2b98OxA+v77j9dH19PbNnz65IXOmpGtIzi0VEykpNItBJQyIi5aUnEQQqEIikU1pqAw5kO1OTCFQgEEmvuro6Nm7ceMQnA3dn48aN1NXV7dd8KWos7mgjqHIgItLvxo4dy+rVq9mwYUO1Q6m4uro6xo4du1/zVDQRmNlM4GtAFviOu5e9ksPM3gXMAU5z9765nV7XdYR3XUcgkj75fJ6JEydWO4xDVsWqhswsC9wIzAKmAJea2ZQy0w0ErgEerVQs8Xrid5UIRERKVbKN4HRgqbsvc/c2YDZwQZnp/h/wb0BFr63u5j5OIiKpV8lEMAZYlehfHYZ1MrNpwDh3v6unBZnZlWY238zmH2wdnwoEIiKlqnbWkJllgK8Cn9rXtO5+k7tPd/fpw4cPP7D16YIyEZGyKpkI1gDjEv1jw7AOA4GpwANmtgI4A5hrZtMrEYxuMSEiUl4lE8HjwCQzm2hmNcAlwNyOke6+1d2HufsEd58APAKcX6mzhvast5JLFxE5/FQsEbh7AbgKmAc8C9zh7ovM7DozO79S6+3Onqf+KBOIiCRV9DoCd78buLvLsM93M+2MSsaik4ZERMpLzS0mOqhqSESkVGoSgRqLRUTKS08iQPcaEhEpJz2JQM8sFhEpKz2JILyrRCAiUio9iUCnDYmIlJWaRNBBJQIRkVIpSgShsVhtBCIiJVKTCPQ8AhGR8tKTCKodgIjIISo9iUDPLBYRKSs9iaDaAYiIHKJSkwg6qLFYRKRUahKBGotFRMpLXyKobhgiIoec9CQCPbNYRKSs1CQCtRaLiJSXnkQQqDwgIlIqNYlAdx8VESkvPYlAD68XESkrPYkgvKtEICJSKj2JQKePioiUlZ5EoNOGRETKSk0i6KCqIRGRUqlJBHtuMaFMICKSlJ5EEN6VBkRESqUmEaCbzomIlJWaRGB6ZrGISFnpSQQ6aUhEpKzUJIJOKhCIiJRITSJQY7GISHnpSQR6eL2ISFkVTQRmNtPMnjOzpWZ2bZnxHzazp83sKTP7vZlNqVws8bsai0VESlUsEZhZFrgRmAVMAS4tc6D/kbuf6O4nA18GvlqxeMK7SgQiIqUqWSI4HVjq7svcvQ2YDVyQnMDdtyV6B1DBKnydNSQiUl6ugsseA6xK9K8GXt11IjP7GPBJoAZ4UwXjAdRYLCLSVdUbi939Rnc/Fvgs8H/KTWNmV5rZfDObv2HDhgNckx5eLyJSTiUTwRpgXKJ/bBjWndnA28uNcPeb3H26u08fPnz4AQWj5xGIiJRXyUTwODDJzCaaWQ1wCTA3OYGZTUr0ngcsqVQwnU0EygQiIiUq1kbg7gUzuwqYB2SBW9x9kZldB8x397nAVWZ2DtAObAYur1Q8ndcRKBOIiJTYZyIws9cCT7n7DjO7DJgGfM3dV+5rXne/G7i7y7DPJ7qv2f+QD4xOGhIRKa83VUPfBHaa2V8CnwJeAG6raFQVpLZiEZFSvUkEBY9PtbkA+Lq73wgMrGxYfc/0PAIRkbJ600bQYmafAy4DzjKzDJCvbFh9b8/zCEREJKk3JYKLgd3AB919HfFpoF+paFQVoGcWi4iU16sSAXHjcNHMXgUcD9xe2bAqR2lARKRUb0oEDwK1ZjYGuA94L/C9SgZVCbrXkIhIeb1JBObuO4F3At9w94uAqZUNq3JUMyQiUqpXicDMzgTeA9y1H/MdUkzPKBMRKas3B/SPA58D7gxXBh8D3F/RqCpAp4+KiJS3z8Zid/8d8DszazSzRndfBlxd+dD6ViZkgkiJQESkxD5LBGZ2opn9CVgEPGNmT5jZCZUPrW9lw5ZGKhKIiJToTdXQt4FPuvvR7j6e+DYTN1c2rL5nnSUCJQIRkaTeJIIB7t7ZJuDuDxA/VvKwklEiEBEpqzcXlC0zs/8LfD/0XwYsq1xIlZHtSARRlQMRETnE9KZE8AFgOPBT4CfAMOCKSgZVCR1nDalEICJSqjdnDW2my1lCZvZj4nsQHTYyGVUNiYiUc6AXhp3Zp1H0g6xOHxURKeuwu0L4QIUCAUVlAhGREt1WDZnZtO5GcTg+j6DjmcWqGhIRKdFTG8F/9DBucV8HUmnZjKqGRETK6TYRuPsb+zOQSlPVkIhIealpI9CVxSIi5aUmEXRUDSkPiIiUSk0i6KwaUiYQESnRbSIws8sS3a/tMu6qSgZVCbrXkIhIeT2VCD6Z6P7vLuM+UIFYKipjqhoSESmnp0Rg3XSX6z/k6awhEZHyekoE3k13uf5DXlb3GhIRKaunC8qON7MFxP/+jw3dhP5jKh5ZHzPda0hEpKyeEsHkfouin2QMImUCEZESPV1ZvDLZb2ZDgbOAF939iUoHVgnZjKlqSESki55OH/2lmU0N3aOAhcRnC33fzD7eP+H1LTNT1ZCISBc9NRZPdPeFofsK4Ffu/jbg1RyGp49CqBpSiUBEpERPiaA90X02cDeAu7cAvXryr5nNNLPnzGypmV1bZvwnzewZM1tgZr8xs6P3J/j9lTVTG4GISBc9JYJVZvZ3ZvYOYBpwL4CZ1dOL5xGYWRa4EZgFTAEuNbMpXSb7EzDd3U8C5gBf3v9N6L2MqoZERPbSUyL4IHAC8H7gYnffEoafAfxPL5Z9OrDU3Ze5exswG7ggOYG73+/uO0PvI8DY3oe+/0xVQyIie+nprKH1wIfLDL8fuL8Xyx4DrEr0ryZuX+jOB4F7yo0wsyuBKwHGjx/fi1WXp7OGRET21tOjKuf2NKO7n99XQYQb3E0H3tDNum4CbgKYPn36AR/J46ohJQIRkaSeLig7k/gf/e3Ao+z//YXWAOMS/WPDsBJmdg7wj8Ab3H33fq5jv5gZxV41c4uIpEdPiWAk8GbgUuCvgbuA2919US+X/TgwycwmEieAS8JyOpnZKcC3gZmhKqqishk9vF5EpKtuG4vdveju97r75cQNxEuBB3r7LAJ3LwBXAfOAZ4E73H2RmV1nZh3VSl8BGoH/NbOn9lUddbBUNSQisreeSgSYWS1wHnGpYAJwA3Bnbxfu7ncTrj9IDPt8ovuc/Yj1oGVUNSQispeeGotvA6YSH8i/lLjK+LCVUdWQiMheeioRXAbsAK4Bru64jTNxo7G7e1OFY+tzGTM9s1hEpIueriM44h5sn8sYBV1aLCJS4og72Pckn83QXlAjgYhIUqoSQS6rEoGISFfpSgSZDO06bUhEpESqEkFNVolARKSrVCWCXNYoFFU1JCKSlLJEkKFdbQQiIiVSlQhqskZBVUMiIiVSlQjUWCwisrd0JQK1EYiI7CVViSCfzdAeqUQgIpKUskRgtBdUIhARSUpVIshlMxRUIhARKZGqRJDPGO1qIxARKZGuRKAri0VE9pKqRJDLZnTWkIhIF6lKBPms6awhEZEuUpUIanMZ3KFNzyQQEemUqkRQl88C0FooVjkSEZFDR6oSQX1NSARtSgQiIh3SlQhCiWBXuxKBiEgHJQIRkZRLVSKoC1VDu1Q1JCLSKVWJQCUCEZG9pTIRtCoRiIh0SlciCFVDO1U1JCLSKVWJYGBdDoCW1kKVIxEROXSkKhE0N9QAsGlHW5UjERE5dKQqEdTlszTUZNmsRCAi0ilViQDiUsGmnUoEIiIdKpoIzGymmT1nZkvN7Noy488ysyfNrGBmF1Yylg7NA/IqEYiIJFQsEZhZFrgRmAVMAS41syldJnsReD/wo0rF0VVzQw2bd7b31+pERA55lSwRnA4sdfdl7t4GzAYuSE7g7ivcfQHQb/eFjhOBSgQiIh0qmQjGAKsS/avDsKoaMqBGZw2JiCQcFo3FZnalmc03s/kbNmw4qGU1N9TQ0lrQw2lERIJKJoI1wLhE/9gwbL+5+03uPt3dpw8fPvyggjqqqRaADdt3H9RyRESOFJVMBI8Dk8xsopnVAJcAcyu4vl4Z19wAwIsbd1Y5EhGRQ0PFEoG7F4CrgHnAs8Ad7r7IzK4zs/MBzOw0M1sNXAR828wWVSqeDkcPDYlg045Kr0pE5LCQq+TC3f1u4O4uwz6f6H6cuMqo34waVAfAXU+v4+LTxvfnqkVEDkmHRWNxX8pl403eolNIRUSAFCYCgHMmH8Xudp01JCICKU0Ek0YMZNkr29ld0HMJRERSmQiaG/K0F50/vrCx2qGIiFRdKhPBjOOOAmD9ttYqRyIiUn2pTATHDm+koSbL4nUt1Q5FRKTqUpkIshnj+JEDeWbttmqHIiJSdalMBAAThzXy6PJNRJFXOxQRkapKbSIYN6QeQNVDIpJ6qU0EF02P74f36HKdOSQi6ZbaRDBmcD1jBtfz0JJXqh2KiEhVpTYRABSiiN8uXk97UVcZi0h6pToR/O3rjwHQhWUikmqpTgRvPWk0AL/889oqRyIiUj2pTgQjB9Vxwugmlr+iZxOISHqlOhEAnDN5BE++uJm1W3ZVOxQRkapIfSJ457QxRA53LXip2qGIiFRF6hPB0UMHMHVME3f+aQ3uuspYRNIn9YkA4JLTxvPMS9uYv3JztUMREel3SgTAu6aNZWBtjtsfe7HaoYiI9DslAqC+Jstbpo7kp0+uUaOxiKSOEkHw0RnHAnD9PYurHImISP9SIgiOGd7IR2Ycy9w/r+XPq7ZUOxwRkX6jRJDwN6+bCMAFN/5BZxCJSGooESQMbazlQ2+I7z80498fqG4wIiL9RImgi8++5XgAVm7cya+febnK0YiIVJ4SQReZjPHQZ94IwN/cNp/fPb+hyhGJiFSWEkEZ44Y08Pm3TgHg8lseY31La5UjEhGpHCWCbnzgdRP5yoUnAXD6P/+G51/Ws41F5MikRNCDi6aP43Oz4jaDv/rPB5mtK49F5Ahkh9tpktOnT/f58+f36zqfWrWFt9/4h87+U8YPZvaVZ1Cby/ZrHCIiB8G6HaFE0Dstre2c+MX79hp+8rjB3PGhM6nJqXAlIoc0JYK+0laIeNX/uafb8fmscd6Jozhx7GDee8bR5LOGO+xqLzKgNtePkYqIlKhOIjCzmcDXgCzwHXe/vsv4WuA24FRgI3Cxu6/oaZl9ngiiIrTtgLqm/Z51Q8tu/u3excxbuI6W3YVez3fK+MH86cUtADTW5shnjc0720umGT2ojuNHNfHbxesBaKrLsa21wOhBdazdGp/FNG38YGpyGZ5YuZlzJo9gweqtZDNG5M7aLbuIwkf7/tdM4A9LX2HJ+u3MmjqSyaOaeGbtNu5dtI4Pv+FYhjXWsLsQ0VSXY922Vu5a8BIrNu5kWGMN44c08OSLWxg9qI5JIwbymmOHMrSxli/8fCE72ooAnH38UZgZ0yc0c8NvlvC5cycz6ahGdrYV+M5DyxnZVMe4IQ0sWruN+posQwfUsHrzLqaOaWJccwNjmut59qVtjB5cz662Ig88t57F61oYNaiOGccdxQ8eWcm6ra1c+urx1OezLFq7ldp8vJwxg+tpbqjBcdoKEfU1Ob77++WMa65nR1uBNx0/gi0721jy8nZe2LCdJeu3c+GpY5k8qolhjTWMba4nm8mQNWP+yk3MX7mZcc0NnDVpGEs3bOfFjTs5e/IInlu3jVe2tzGoPs/EYQNoK0Y8tGQDxcjZ0VZkQ8tuRg+qoxA544Y08OxL2zhmWCOO09oeUZ/P8tzL23jDq4YzqD5PU12ezTvbyWWM9S2tLFm/ndccO5RfPfMyI5rqOG3CELa1ttNUl+eFDdsx4IHnN/Chs45l3bZWtu1qJ581Rg+u5+k1W3nyxS1cfubRDKqPp9/VFlGMIoYMqGFIYy2rNu1k+tHNNNXnidyZv2IzX5i7iA+/4Vg27djNkAG1nDahGXcY3JBn7dZW5i1axwvrtzN+SAOjB9czfUIz67ftZmxzPbms0VZwlq5vYeSgehprczyybCNvnjKCyJ2nXtxCa6HIUQPrGDO4no07dvPK9jZ2txeZNGIgi9dtY2dbkeNHDuTJlVsYNrCGKaMGMaKplp89tYbhjXXU5TNMHTOIJ1duZtaJo7j9sRepzWXYuL2NAbU5zp58FIPq87y4aSe/XbyeYY21jGyq5dHlm3jPq49mW2s7f3FUI5t3tJHLGo218b659Y8ruPKsY8hnMwyozfGPdz5NfU2W1xw7lELRufDUseSzGdZs2cWO3QXai85xIxtZun47YKzYuINp45t5aesuxjY38ML67RzVVMurRgxk1aadtLQWOGnsILbvLnDzQ8s4ccxgClHEy9taOeOYobS0FtjVVuSV7bvJZTOMH9LArvYiWTOGD6xlxcYdjGyqY9Xmnax4ZQevnzScxroctbkM44Y00FSX3+9jVdD/icDMssDzwJuB1cDjwKXu/kximo8CJ7n7h83sEuAd7n5xT8vt80SwaTnccDLUDISm0dA0CprGhO7RUDcIMjnI5CGbj7uzecjWJN5r8WyOLW0Z/vM3y5n37Hp2tUeA4RgOOEaEEZEhsiztnqGHz6WMcp/T/swvIoe7f33niVx6+vgDnb3bA0Yl6ypOB5a6+zIAM5sNXAA8k5jmAuCLoXsO8HUzM+/P+qragfDm62DbS7BtDWxbCy/cD9vXgUe9XowBzcB1wHVZ4jJQb+bKZHHLxunCo3CPIwfvSCE9c8vgZMAyYBa/2PMeEb+bxa9i5ETu8XgPIYQpiYpkvIh5BESYR7jF8XmIMyJDJpsDy1J0KEQQWYZ8Nksmk6EQReAR2bA9hUI7GSJymXhrdhecTCYbvpIZstkMGSCKioBjHqdMjyIyGQPL0F70zu1zMrQV4+VlcLIGGXPcnWIUkctkcCByiDCyGaO14DgZGmrzFMLwjGXJ5bLsaIuoyzrmRTwq0tbWRs4iajNO5NAeQT6bibfV423IZ7O0R87uglNfkyNyyGWMXe0RbZGTy2SpzWexTIa2oneWttraC2TNcY8gishloFAsxn8SHHLZLJFDbT5LIYJi53KL1OeztOwuMLg+TyFyipFjQDYT/7bbChHZjJGxuEQYAe0FJ5fLUJ/PsqOtSF0+Ry5si5NhdzEiiiJqMk4mKpLPODmLiIoF2otFIo/jqsnnaI+MbDb+UheKRfLmRFFEzuLPOyoWiTBqa2soeJY2N9xyZHM5zLKdMe8uOrmMsaOtSC6ToT6fYXtrGxmcuhxkwl+nyOM/TzX5HK0FpyafZ9vuIlmDPEWI2mjIOjlvJxO1k7Ni/D3PZGgvQj6XpeBGW5HO71E+m6Utir+z+VyWjBm72gpEDvksZMwoFCPyWSOfzdAWQcYytEdOe+TkslncIWsQefzLymWMYhQfJ3LZLO1FpxA5+Vz8+e9qj2gvRNTms2QNCuFzq80Z7cWItkKRupyRy8TrzhpEUUQ2A1EUf1fq81nw+He7o/ULwAd7c3DZL5VMBGOAVYn+1cCru5vG3QtmthUYCrySnMjMrgSuBBg//oCzYXkDhsFrr9l7eLEA21+G3S0QtUOxPa5G6ugutsfdhd2hvw2KodsjSBzQO989Ai9CFEFUCN1FzIvxgS4csOMDeWbPQX3PjtjTHZZrHh+wKXl55zSZZAyE84XL5dlMNl5nJguWjd8xzItYVAzxhrijInhEziNq8T3r9IgaiOe3OMnVWib0ZwCnJjEtHfOGbY+3OZHUQqz55LTuDOiYJ5n0OpbRdZ/j1CXWV5tcv0dxfybXuc0Nmeyefijdvr269+xX3BmAh17fe/qu25bsLtkfHfPt+WiaSt699DtR1p6YOvoHdPd9tEwo8WZLuus692Xp/sIdMh2xJ78zmXhcVKA2KjCg43vS8T3vEk9nv2U4qmTfJD/70u/1sI4/ZplkiTx0Z3Kd8+TDdHmPqA9/MDq2oY7SbRmQ/I11/r7iba8v+VxCPMlpjNJ5fe/v3mCPSj+v5G848Yetc1ld/sjRJb4BR0/ex2d/YA6L1kt3vwm4CeKqoX5ZaTYHg8b0y6pERKqpkuc8rgHGJfrHhmFlpzGzHDCIuNFYRET6SSUTwePAJDObaGY1wCXA3C7TzAUuD90XAr/t1/YBERGpXNVQqPO/CphH3HR6i7svMrPrgPnuPhf4LvB9M1sKbCJOFiIi0o90QZmISDp0e5aB7osgIpJySgQiIimnRCAiknJKBCIiKXfYNRab2QZg5QHOPowuVy2nlPaD9gFoH3RIy354xd1nlhtx2CWCg2Fm8919erXjqDbtB+0D0D7ooP2gqiERkdRTIhARSbm0JYKbqh3AIUL7QfsAtA86pH4/pKqNQERE9pa2EoGIiHShRCAiknKpSQRmNtPMnjOzpWZ2bbXjOVhmdouZrTezhYlhQ8zsV2a2JLw3h+FmZjeEbV9gZtMS81wepl9iZpcnhp9qZk+HeW4ws309Fqvfmdk4M7vfzJ4xs0Vmdk0Ynpr9YGZ1ZvaYmf057IMvheETzezREPePw63gMbPa0L80jJ+QWNbnwvDnzOwtieGHxW/HzLJm9icz+2XoT90+OGDufsS/iG+D/QJwDFAD/BmYUu24DnKbzgKmAQsTw74MXBu6rwX+LXSfC9xDfPfBM4BHw/AhwLLw3hy6m8O4x8K0FuadVe1tLrMPRgHTQvdA4HlgSpr2Q4irMXTngUdDvHcAl4Th3wI+Ero/CnwrdF8C/Dh0Twm/i1pgYvi9dDx9+7D47QCfBH4E/DL0p24fHOgrLSWC04Gl7r7M3duA2cAFVY7poLj7g8TPcEi6ALg1dN8KvD0x/DaPPQIMNrNRwFuAX7n7JnffDPwKmBnGNbn7Ix7/Qm5LLOuQ4e4vufuTobsFeJb4Odip2Q9hW7aH3nx4OfAmYE4Y3nUfdOybOcDZoZRzATDb3Xe7+3JgKfHv5rD47ZjZWOA84Duh30jZPjgYaUkEY4BVif7VYdiRZoS7vxS61wEjQnd329/T8NVlhh+yQvH+FOJ/xKnaD6FK5ClgPXESewHY4u6FMEky7s5tDeO3AkPZ/31zqPkv4DNAeMI9Q0nfPjhgaUkEqRP+wabi3GAzawR+Anzc3bclx6VhP7h70d1PJn4u+OnA8dWNqH+Z2VuB9e7+RLVjOVylJRGsAcYl+seGYUeal0N1BuF9fRje3fb3NHxsmeGHHDPLEyeBH7r7T8Pg1O0HAHffAtwPnElc7dXxKNpk3J3bGsYPAjay//vmUPJa4HwzW0FcbfMm4Gukax8cnGo3UvTHi/jZzMuIG4A6GntOqHZcfbBdEyhtLP4KpY2kXw7d51HaSPpYGD4EWE7cQNocuoeEcV0bSc+t9vaW2X4jrrf/ry7DU7MfgOHA4NBdDzwEvBX4X0obSj8auj9GaUPpHaH7BEobSpcRN5IeVr8dYAZ7GotTuQ8OaL9VO4B+/IKcS3xWyQvAP1Y7nj7YntuBl4B24jrLDxLXc/4GWAL8OnEwM+DGsO1PA9MTy/kAcaPYUuCKxPDpwMIwz9cJV6EfSi/gdcTVPguAp8Lr3DTtB+Ak4E9hHywEPh+GH0OcxJaGA2JtGF4X+peG8ccklvWPYTufI3F21OH02+mSCFK5Dw7kpVtMiIikXFraCEREpBtKBCIiKadEICKSckoEIiIpp0QgIpJySgSSamZWNLOnwt07nzSz1+xj+sFm9tFeLPcBM0v1A9Hl8KFEIGm3y91Pdve/BD4H/Os+ph9MfPdKkSOGEoHIHk3AZojvX2RmvwmlhKfNrONuk9cDx4ZSxFfCtJ8N0/zZzK5PLO+i8KyA583s9WHarJl9xcweD89E+FAYPsrMHgzLXdgxvUh/yO17EpEjWn24c2cd8fMN3hSGtwLvcPdtZjYMeMTM5hLfsmKqxzd5w8xmEd+S+NXuvtPMhiSWnXP3083sXOALwDnEV4BvdffTzKwW+IOZ3Qe8E5jn7v9sZlmgocLbLdJJiUDSblfioH4mcJuZTSW+HcW/mNlZxLc2HsOe21knnQP8j7vvBHD35DMiOm6C9wTxfaEA/go4ycwuDP2DgEnA48At4SZ6P3P3p/pk60R6QYlAJHD3h8O//+HE95YZDpzq7u3hzpZ1+7nI3eG9yJ7fmgF/5+7zuk4cks55wPfM7KvuftsBbIbIflMbgUhgZscT321yI/E/9fUhCbwRODpM1kL8WMwOvwKuMLOGsIxk1VA584CPhH/+mNmrzGyAmR0NvOzuNxM/ZWtaTwsR6UsqEUjadbQRQPxv/XJ3L5rZD4FfmNnTwHxgMYC7bzSzP5jZQuAed/97MzsZmG9mbcDdwD/0sL7vEFcTPRkej7iB+BGKM4C/N7N2YDvwvj7dSpEe6O6jIiIpp6ohEZGUUyIQEUk5JQIRkZRTIhARSTklAhGRlFMiEBFJOSUCEZGU+//E3apcUnnntgAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: left;\">\n",
       "      <th>epoch</th>\n",
       "      <th>train_loss</th>\n",
       "      <th>valid_loss</th>\n",
       "      <th>time</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>0</td>\n",
       "      <td>0.003628</td>\n",
       "      <td>0.000465</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>0.003583</td>\n",
       "      <td>0.000374</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>2</td>\n",
       "      <td>0.003492</td>\n",
       "      <td>0.000605</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>3</td>\n",
       "      <td>0.003671</td>\n",
       "      <td>0.000393</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>4</td>\n",
       "      <td>0.003391</td>\n",
       "      <td>0.000351</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>5</td>\n",
       "      <td>0.003548</td>\n",
       "      <td>0.000410</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>6</td>\n",
       "      <td>0.003525</td>\n",
       "      <td>0.000366</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>7</td>\n",
       "      <td>0.003588</td>\n",
       "      <td>0.000457</td>\n",
       "      <td>00:06</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>8</td>\n",
       "      <td>0.003521</td>\n",
       "      <td>0.000379</td>\n",
       "      <td>00:06</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>9</td>\n",
       "      <td>0.003433</td>\n",
       "      <td>0.000362</td>\n",
       "      <td>00:06</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>10</td>\n",
       "      <td>0.003620</td>\n",
       "      <td>0.000369</td>\n",
       "      <td>00:06</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>11</td>\n",
       "      <td>0.003368</td>\n",
       "      <td>0.000423</td>\n",
       "      <td>00:06</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12</td>\n",
       "      <td>0.003660</td>\n",
       "      <td>0.000414</td>\n",
       "      <td>00:06</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>13</td>\n",
       "      <td>0.003670</td>\n",
       "      <td>0.000392</td>\n",
       "      <td>00:06</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>14</td>\n",
       "      <td>0.003318</td>\n",
       "      <td>0.000356</td>\n",
       "      <td>00:06</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "No improvement since epoch 4: early stopping\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "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": [
    "# Load and process data\n",
    "\n",
    "if not os.path.exists(\"ProtBERT_encoded_110_swissprot_DenseSAE.pt\"):\n",
    "\n",
    "    prot_dense_embedded = torch.load(\"../Data/EncodedData/ProtBERT_encoded_110_swissprot.pt\")\n",
    "\n",
    "    train, val = train_test_split(np.array(prot_dense_embedded), random_state = 255)\n",
    "    train = SAEDataset(torch.tensor(train))\n",
    "    val   = SAEDataset(torch.tensor(val))\n",
    "\n",
    "    # Initiate learner\n",
    "    prot_loaders = DataLoaders.from_dsets(train, val,\n",
    "                                          bs=BATCHSIZE,\n",
    "                                          device=\"cuda:0\")\n",
    "\n",
    "    ###################################################\n",
    "    # Train SAE & find and record min loss\n",
    "    ###################################################\n",
    "    prot_dense_learner = trainer.train_sae(loaders, prot_sae, dense_sae_splitter,\n",
    "                                           learner_cbs = [EarlyStoppingCallback(patience = 10)])\n",
    "    \n",
    "    torch.save(prot_dense_learner.model, \"ProtBERT_encoded_110_swissprot_DenseSAE.pt\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Embed and concat sequences"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "from Utils.data_reduction import DataReducer\n",
    "\n",
    "full_positive = pd.read_csv(\"../Data/TrainingData/PositiveSamples/RNAInter_small_interactions_current.csv\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "count    12161.000000\n",
       "mean        37.411233\n",
       "std         26.056531\n",
       "min         16.000000\n",
       "25%         22.000000\n",
       "50%         22.000000\n",
       "75%         62.000000\n",
       "max        100.000000\n",
       "Name: rnas, dtype: float64"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "full_positive['rnas'].apply(len).describe()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "full_positive['rnas'] = full_positive['rnas'].str.replace(\"T\",\"U\").str.replace(\"t\",\"u\")\n",
    "full_positive['proteins'] = full_positive['proteins'].apply(\" \".join)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "PROT_LEN_LIMIT = 110\n",
    "# prot_encoder   = ProtBERTEmbedder(PROT_LEN_LIMIT+2, cuda = True)\n",
    "rna_encoder    = SkipGramEmbedder(\"../SeqEmbedders/GensimWord2Vec/RNA2Vec_1024_hidden.model\", reduce = True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "../Utils/encoders.py:241: UserWarning: The given NumPy array is not writeable, and PyTorch does not support non-writeable tensors. This means you can write to the underlying (supposedly non-writeable) NumPy array using the tensor. You may want to copy the array to protect its data or make it writeable before converting it to a tensor. This type of warning will be suppressed for the rest of this program. (Triggered internally at  /pytorch/torch/csrc/utils/tensor_numpy.cpp:141.)\n",
      "  kmer_matrix = torch.stack([torch.from_numpy(self.bio2vec.wv[kmer.lower()]) for kmer in kmers]).T\n"
     ]
    }
   ],
   "source": [
    "encoded_rnas = rna_encoder(full_positive['rnas'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "tensor([[ 0.0594, -0.4377,  0.0526,  ...,  0.2972,  0.2020,  0.4808],\n",
       "        [ 0.1034, -0.4655,  0.0420,  ..., -0.1717,  0.1686,  0.0449],\n",
       "        [ 0.2107, -0.7328,  0.2686,  ...,  0.2232, -0.0424,  0.7081],\n",
       "        ...,\n",
       "        [-0.2783, -0.4611, -0.4148,  ...,  0.2140,  0.2425,  0.1936],\n",
       "        [ 0.2071, -0.2519,  0.5972,  ..., -0.2751, -0.0802, -0.1256],\n",
       "        [-0.0286, -0.3779,  0.4139,  ..., -0.0795,  0.3838,  0.0880]])"
      ]
     },
     "execution_count": 16,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "encoded_rnas"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [],
   "source": [
    "from Utils.negative_sampling import IterableProteinEmbedding"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "d315e869f68b49f68cfd4aaaea6000b9",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(FloatProgress(value=0.0, max=6081.0), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    }
   ],
   "source": [
    "chunksize = 2\n",
    "\n",
    "tokenizer  = AutoTokenizer.from_pretrained(\"Rostlab/prot_bert_bfd\", do_lower_case = False)\n",
    "model      = AutoModel.from_pretrained(\"Rostlab/prot_bert_bfd\")\n",
    "\n",
    "protein_iterator = IterableProteinEmbedding(full_positive['proteins'], tokenizer, model, chunksize, max_len = 110, cuda = True)\n",
    "encoded_proteins = torch.cat([encoded_proteins for encoded_proteins in protein_iterator]).cpu()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "tensor([[ 0.3535, -0.5568, -0.4282,  ..., -0.1077, -0.5357, -0.5210],\n",
       "        [ 0.3535, -0.5568, -0.4282,  ..., -0.1077, -0.5357, -0.5210],\n",
       "        [ 0.3535, -0.5568, -0.4282,  ..., -0.1077, -0.5357, -0.5210],\n",
       "        ...,\n",
       "        [ 0.3841, -0.3273, -0.6873,  ..., -0.2132,  0.6570, -0.9688],\n",
       "        [ 0.2240, -0.3008, -0.3164,  ..., -0.5238, -0.5207, -0.7772],\n",
       "        [ 0.2240, -0.3008, -0.3164,  ..., -0.5238, -0.5207, -0.7772]],\n",
       "       device='cuda:0')"
      ]
     },
     "execution_count": 19,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "prot_dense_learner.model(encoded_proteins.cuda())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [],
   "source": [
    "latent_proteins = prot_dense_learner.model(encoded_proteins.cuda()).cpu()\n",
    "latent_rnas     = rna_dense_learner.model(encoded_rnas.cuda()).cpu()\n",
    "\n",
    "latent_interactions = torch.cat([latent_proteins, latent_rnas], axis = 1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "torch.Size([12161, 1000])"
      ]
     },
     "execution_count": 21,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "latent_interactions.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [],
   "source": [
    "class DataReducer:\n",
    "    \"\"\"\n",
    "    Hyperparameter optimization and negative distribtuion sampling experiments\n",
    "    require a reduced dataset to optimize the time needed to complete each\n",
    "    expertiment. This is the inspiration behind DataReducer, which converts a\n",
    "    list of sequences into their corresponding vectors, optionally reduces them \n",
    "    using a reducer function (e.g., SAE, t-SNE, UMAP transformations) and then \n",
    "    clusters them using a user defined cluster function. \n",
    "    \"\"\"\n",
    "    \n",
    "    # Default fit -> transform clusterer arg\n",
    "    clusterer = hdbscan.HDBSCAN(min_cluster_size=5)\n",
    "    \n",
    "    # Default reducer does nothing but return its input\n",
    "    default_reducer = lambda x: x\n",
    "    \n",
    "    def __init__(self, embedder_fnc, cluster_fnc = clusterer.fit_predict, reducer_fnc = default_reducer, seed = 255):\n",
    "        self.seed         = seed\n",
    "        self.embedder_fnc = embedder_fnc\n",
    "        self.cluster_fnc  = cluster_fnc\n",
    "        self.reducer_fnc  = reducer_fnc\n",
    "        \n",
    "    def _transform_sequence_to_vec(self, sequences):\n",
    "        return self.embedder_fnc(sequences)\n",
    "    \n",
    "    def _reduce_data(self, vectors):\n",
    "        return self.reducer_fnc(vectors)\n",
    "    \n",
    "    def _fit_and_transform_vector_data(self, vectors):\n",
    "        return self.cluster_fnc(vectors)\n",
    "    \n",
    "    def _isolate_data_by_cluster(self, data : np.array, clusters : np.array, cluster_id : int) -> np.array:\n",
    "        return data[clusters == cluster_id,:]\n",
    "    \n",
    "    def _isolate_data_by_clusters(self, data : np.array, clusters : np.array) -> list:\n",
    "        return [self._isolate_data_by_cluster(data, clusters, cluster) for cluster in set(clusters)]\n",
    "    \n",
    "    def _shuffle_array(self, array):\n",
    "        np.random.seed(self.seed)\n",
    "        np.random.shuffle(array)\n",
    "        \n",
    "    def _shuffle_clusterd_data_rows(self, isolated_data):\n",
    "        \n",
    "        for cluster in isolated_data:\n",
    "            self._shuffle_array(cluster)\n",
    "    \n",
    "    def _merge_clustered_data(self, clusters):\n",
    "        return np.concatenate(clusters)\n",
    "    \n",
    "    def _sample_reduced_data(self, n, vectors, cluster_idxs):\n",
    "        CLUSTER_COUNT      = len(set(cluster_idxs))\n",
    "        SAMPLE_PER_CLUSTER = ceil(n / CLUSTER_COUNT)\n",
    "        \n",
    "        print(f\"TOTAL SAMPLES     : {n}\")\n",
    "        print(f\"TOTAL CLUSTERS    : {CLUSTER_COUNT}\")\n",
    "        print(f\"SAMPLE_PER_CLUSTER: {SAMPLE_PER_CLUSTER}\")\n",
    "        \n",
    "        clusters     = self._isolate_data_by_clusters(vectors, cluster_idxs)\n",
    "        samples      = [cluster[:SAMPLE_PER_CLUSTER] for cluster in clusters]\n",
    "        sample_array = self._merge_clustered_data(samples)\n",
    "        self._shuffle_array(sample_array)\n",
    "        \n",
    "        return sample_array\n",
    "    \n",
    "    #################################################################################\n",
    "    def sample_reduced_data_from_encoded_seqs(self, n, vectors):\n",
    "        # converts tensors, nested lists, and nested tuples to arrays\n",
    "        vectors = np.array(vectors)\n",
    "        \n",
    "        reduced_vectors = self._reduce_data(vectors)\n",
    "        clusters        = self._fit_and_transform_vector_data(reduced_vectors)\n",
    "        samples         = self._sample_reduced_data(n, reduced_vectors, clusters)\n",
    "        return samples\n",
    "    \n",
    "    def sample_reduced_data_from_sequences(self, n, sequences):\n",
    "        print(f\"Processing {len(sequences)} sequences...\")\n",
    "        vectors = self._transform_sequence_to_vec(sequences)\n",
    "        samples = self.sample_reduced_data_from_encoded_seqs(n,vectors)\n",
    "        return samples\n",
    "    \n",
    "    def load_and_filter_data_from_fasta(self, infile, alphabet, min_seq_len, max_seq_len, seq_limit):\n",
    "        def comprised_of_proper_alpabet(sequence):\n",
    "            \"returns 'True' if sequence only contains letters of an alphabet\"\n",
    "            return not set(sequence).difference(alphabet)\n",
    "        \n",
    "        def is_eq_or_greater_than_min_len(sequence, min_seq_len):\n",
    "            return len(sequence) >= min_seq_len\n",
    "        \n",
    "        def is_eq_or_less_than_max_len(sequence, min_seq_len):\n",
    "            return len(sequence) <= max_seq_len\n",
    "        \n",
    "        def is_valid_sequence(sequence, min_seq_len):\n",
    "            return is_eq_or_greater_than_min_len(sequence, min_seq_len) and \\\n",
    "                   is_eq_or_less_than_max_len(sequence, max_seq_len) and \\\n",
    "                   comprised_of_proper_alpabet(sequence)\n",
    "        \n",
    "        def shuffle_seq_list(sequences):\n",
    "            random.seed(self.seed)\n",
    "            random.shuffle(sequences)\n",
    "        \n",
    "        with open(infile) as handle:\n",
    "            sequences = [str(record.seq) for record in SeqIO.parse(handle, 'fasta') if is_valid_sequence(str(record.seq), min_seq_len)]\n",
    "            shuffle_seq_list(sequences)\n",
    "            sequences = sequences[:seq_limit]\n",
    "            return sequences\n",
    "    \n",
    "    def encode_data_from_fasta(self, infile, alphabet, min_seq_len, max_seq_len, seq_limit):\n",
    "        sequences = self.load_and_filter_data_from_fasta(infile, alphabet, min_seq_len, max_seq_len, seq_limit)\n",
    "        encoded_seqs = self._transform_sequence_to_vec(sequences)\n",
    "        return encoded_seqs\n",
    "    \n",
    "    def sample_reduced_data_from_fasta(self, n, infile, alphabet, min_seq_len, max_seq_len, seq_limit):\n",
    "        sequences = self.load_and_filter_data_from_fasta(infile, alphabet, min_seq_len, max_seq_len, seq_limit)\n",
    "        samples = self.sample_reduced_data_from_sequences(n, sequences)\n",
    "        return samples\n",
    "                        \n",
    "    ###############################################################\n",
    "    def save_reduced_data_to_file(self, outfile, reduced_samples):\n",
    "        \"End extension should be .npy to save as numpy array or .pt to save as torch tensor\"\n",
    "        \n",
    "        matrix_type = outfile.split('.')[-1]\n",
    "        assert matrix_type in ['npy','pt'], f'outfile needs to have either an \"npy\" or \"pt\" extension for save as numpy array or pytorch tensor respectively. Got: {matrix_type}'\n",
    "        \n",
    "        if matrix_type == 'npy':\n",
    "            with open(outfile, 'wb') as handle:\n",
    "                np.save(outfile, reduced_samples)\n",
    "                \n",
    "        elif matrix_type == 'pt':\n",
    "            torch_reduced_samples = torch.from_numpy(reduced_samples)\n",
    "            torch.save(torch_reduced_samples, outfile)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [],
   "source": [
    "import hdbscan\n",
    "from Utils.data_reduction import DataReducer\n",
    "\n",
    "# embedder_fnc, cluster_fnc = clusterer.fit_predict, reducer_fnc = default_reducer, seed = 255)\n",
    "\n",
    "clusterer = hdbscan.HDBSCAN(min_cluster_size=4, random_state = 255)\n",
    "TOTAL_SAMPLES = 5000\n",
    "\n",
    "# empty_fnc = lambda x: x\n",
    "\n",
    "# reducer = DataReducer(embedder_fnc = empty_fnc, cluster_fnc = clusterer.fit_predict, \n",
    "#                       reducer_fnc = empty_fnc, seed = 255)\n",
    "\n",
    "# # sampled_interactions = reducer.sample_reduced_data_from_encoded_seqs(TOTAL_SAMPLES, latent_interactions)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [],
   "source": [
    "clusterer = hdbscan.HDBSCAN(min_cluster_size=4)\n",
    "clusters  = clusterer.fit_predict(latent_interactions)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "TOTAL SAMPLES     : 5000\n",
      "TOTAL CLUSTERS    : 1037\n",
      "SAMPLE_PER_CLUSTER: 5\n"
     ]
    }
   ],
   "source": [
    "from math import ceil\n",
    "\n",
    "CLUSTER_COUNT      = len(set(clusters))\n",
    "SAMPLE_PER_CLUSTER = ceil(TOTAL_SAMPLES / CLUSTER_COUNT)\n",
    "\n",
    "print(f\"TOTAL SAMPLES     : {TOTAL_SAMPLES}\")\n",
    "print(f\"TOTAL CLUSTERS    : {CLUSTER_COUNT}\")\n",
    "print(f\"SAMPLE_PER_CLUSTER: {SAMPLE_PER_CLUSTER}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [],
   "source": [
    "from Utils.misc import flatten\n",
    "\n",
    "def return_cluster_idxs(clusters, cluster_id, sample_per_cluster):\n",
    "    cluster_series = pd.Series(clusters)\n",
    "    cluster_idxs   = cluster_series[cluster_series == cluster_id].index[:sample_per_cluster]\n",
    "    return list(cluster_idxs)\n",
    "\n",
    "def return_all_cluster_idxs(clusters, cluster_ids, sample_per_cluster):\n",
    "    return flatten([return_cluster_idxs(clusters, cluster_id, sample_per_cluster) for cluster_id in cluster_ids])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "total samples: 4939\n"
     ]
    }
   ],
   "source": [
    "cluster_idxs = return_all_cluster_idxs(clusters, set(clusters), SAMPLE_PER_CLUSTER)\n",
    "print(f\"total samples: {len(cluster_idxs)}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [],
   "source": [
    "reduced_positive_samples = full_positive.iloc[cluster_idxs,:]\n",
    "\n",
    "# reduced_positive_samples.to_csv(\"../Data/TrainingData/PositiveSamples/reduced_RNAInter_small_interactions_current.csv\", index = False)"
   ]
  },
  {
   "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": "3.8.8"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}