full_script.ipynb

{"nbformat":4,"nbformat_minor":0,"metadata":{"kernelspec":{"display_name":"Python 3","language":"python","name":"python3"},"language_info":{"codemirror_mode":{"name":"ipython","version":3},"file_extension":".py","mimetype":"text/x-python","name":"python","nbconvert_exporter":"python","pygments_lexer":"ipython3","version":"3.7.4"},"colab":{"name":"notebook-fmlcv-project.ipynb","provenance":[],"collapsed_sections":[]}},"cells":[{"cell_type":"code","metadata":{"_cell_guid":"b1076dfc-b9ad-4769-8c92-a6c4dae69d19","_uuid":"8f2839f25d086af736a60e9eeb907d3b93b6e0e5","id":"5dnQRDmLqduW"},"source":["import os\n","import shutil\n","import numpy as np\n","import pandas as pd\n","import seaborn as sns\n","import matplotlib.pyplot as plt\n","\n","from sklearn.preprocessing import LabelEncoder\n","from sklearn.preprocessing import OneHotEncoder\n","from sklearn.model_selection import train_test_split\n","from sklearn.metrics import classification_report\n","#from sklearn.metrics import confusion_matrix, plot_confusion_matrix\n","\n","from skimage import io\n","from skimage.transform import resize\n"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"jmRc5Hv3qdua"},"source":["from tensorflow.keras.applications.vgg19 import VGG19\n","from tensorflow.keras.applications.vgg16 import VGG16\n","from tensorflow.keras.callbacks import ModelCheckpoint, EarlyStopping\n"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"r9EkBa3Xqdub"},"source":["from tensorflow.keras.preprocessing.image import ImageDataGenerator\n","from tensorflow.keras.models import Sequential\n","from tensorflow.keras.layers import Conv2D, MaxPooling2D\n","from tensorflow.keras.layers import Activation, Dropout, Flatten, Dense\n","from tensorflow.keras import backend as K\n"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"MFzGbEegqdub"},"source":["sns.set(font_scale=1.6)"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"nlogtsvnqdub","executionInfo":{"status":"ok","timestamp":1633776508463,"user_tz":-120,"elapsed":3,"user":{"displayName":"Pietro Obbiso","photoUrl":"https://lh3.googleusercontent.com/a/default-user=s64","userId":"12598544054389310202"}}},"source":["#### Data visualization ####\n","\n","# Code to convert the xls file to csv\n","data_xls = pd.read_excel('../input/ocular-disease-recognition-odir5k/ODIR-5K/ODIR-5K/data.xlsx', \n","                         dtype=str, index_col=None)\n","data_xls.to_csv('odir5k_data.csv', encoding='utf-8', index=False)"],"execution_count":1,"outputs":[]},{"cell_type":"code","metadata":{"_cell_guid":"79c7e3d0-c299-4dcb-8224-4455121ee9b0","_uuid":"d629ff2d2480ee46fbb7e2d37f6b5fab8052498a","id":"pFcwTgs2qduc","outputId":"7a61f5a9-9438-4436-ce2e-e4bd2d3e31c3"},"source":["# Read the ODIR5K train dataset\n","train_df = pd.read_csv('odir5k_data.csv')\n","\n","# Listing the images\n","images_paths = '../input/ocular-disease-recognition-odir5k/ODIR-5K/ODIR-5K/Training Images'\n","images_list = os.listdir(images_paths) \n","print(f'Total number of images in train set {len(images_list)}')\n","print(f'Total number of rows in train dataset  {len(train_df)}')"],"execution_count":null,"outputs":[{"name":"stdout","output_type":"stream","text":["Total number of images in train set 7000\n","Total number of rows in train dataset  3500\n"]}]},{"cell_type":"code","metadata":{"id":"q7ZosYfDqdud","outputId":"ac2b62c1-02db-49ea-979c-7e6bdd93e11a"},"source":["train_df.head()"],"execution_count":null,"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>ID</th>\n","      <th>Patient Age</th>\n","      <th>Patient Sex</th>\n","      <th>Left-Fundus</th>\n","      <th>Right-Fundus</th>\n","      <th>Left-Diagnostic Keywords</th>\n","      <th>Right-Diagnostic Keywords</th>\n","      <th>N</th>\n","      <th>D</th>\n","      <th>G</th>\n","      <th>C</th>\n","      <th>A</th>\n","      <th>H</th>\n","      <th>M</th>\n","      <th>O</th>\n","    </tr>\n","  </thead>\n","  <tbody>\n","    <tr>\n","      <th>0</th>\n","      <td>0</td>\n","      <td>69</td>\n","      <td>Female</td>\n","      <td>0_left.jpg</td>\n","      <td>0_right.jpg</td>\n","      <td>cataract</td>\n","      <td>normal fundus</td>\n","      <td>0</td>\n","      <td>0</td>\n","      <td>0</td>\n","      <td>1</td>\n","      <td>0</td>\n","      <td>0</td>\n","      <td>0</td>\n","      <td>0</td>\n","    </tr>\n","    <tr>\n","      <th>1</th>\n","      <td>1</td>\n","      <td>57</td>\n","      <td>Male</td>\n","      <td>1_left.jpg</td>\n","      <td>1_right.jpg</td>\n","      <td>normal fundus</td>\n","      <td>normal fundus</td>\n","      <td>1</td>\n","      <td>0</td>\n","      <td>0</td>\n","      <td>0</td>\n","      <td>0</td>\n","      <td>0</td>\n","      <td>0</td>\n","      <td>0</td>\n","    </tr>\n","    <tr>\n","      <th>2</th>\n","      <td>2</td>\n","      <td>42</td>\n","      <td>Male</td>\n","      <td>2_left.jpg</td>\n","      <td>2_right.jpg</td>\n","      <td>laser spot，moderate non proliferative retinopathy</td>\n","      <td>moderate non proliferative retinopathy</td>\n","      <td>0</td>\n","      <td>1</td>\n","      <td>0</td>\n","      <td>0</td>\n","      <td>0</td>\n","      <td>0</td>\n","      <td>0</td>\n","      <td>1</td>\n","    </tr>\n","    <tr>\n","      <th>3</th>\n","      <td>3</td>\n","      <td>66</td>\n","      <td>Male</td>\n","      <td>3_left.jpg</td>\n","      <td>3_right.jpg</td>\n","      <td>normal fundus</td>\n","      <td>branch retinal artery occlusion</td>\n","      <td>0</td>\n","      <td>0</td>\n","      <td>0</td>\n","      <td>0</td>\n","      <td>0</td>\n","      <td>0</td>\n","      <td>0</td>\n","      <td>1</td>\n","    </tr>\n","    <tr>\n","      <th>4</th>\n","      <td>4</td>\n","      <td>53</td>\n","      <td>Male</td>\n","      <td>4_left.jpg</td>\n","      <td>4_right.jpg</td>\n","      <td>macular epiretinal membrane</td>\n","      <td>mild nonproliferative retinopathy</td>\n","      <td>0</td>\n","      <td>1</td>\n","      <td>0</td>\n","      <td>0</td>\n","      <td>0</td>\n","      <td>0</td>\n","      <td>0</td>\n","      <td>1</td>\n","    </tr>\n","  </tbody>\n","</table>\n","</div>"],"text/plain":["   ID  Patient Age Patient Sex Left-Fundus Right-Fundus  \\\n","0   0           69      Female  0_left.jpg  0_right.jpg   \n","1   1           57        Male  1_left.jpg  1_right.jpg   \n","2   2           42        Male  2_left.jpg  2_right.jpg   \n","3   3           66        Male  3_left.jpg  3_right.jpg   \n","4   4           53        Male  4_left.jpg  4_right.jpg   \n","\n","                            Left-Diagnostic Keywords  \\\n","0                                           cataract   \n","1                                      normal fundus   \n","2  laser spot，moderate non proliferative retinopathy   \n","3                                      normal fundus   \n","4                        macular epiretinal membrane   \n","\n","                Right-Diagnostic Keywords  N  D  G  C  A  H  M  O  \n","0                           normal fundus  0  0  0  1  0  0  0  0  \n","1                           normal fundus  1  0  0  0  0  0  0  0  \n","2  moderate non proliferative retinopathy  0  1  0  0  0  0  0  1  \n","3         branch retinal artery occlusion  0  0  0  0  0  0  0  1  \n","4       mild nonproliferative retinopathy  0  1  0  0  0  0  0  1  "]},"execution_count":37,"metadata":{},"output_type":"execute_result"}]},{"cell_type":"code","metadata":{"id":"sRsSrK5nqdud","outputId":"1f99f28a-46e0-4a7f-e7d3-017764257a30"},"source":["# Names of columns in the new dataframe\n","col_names = ['Image', 'Diagnosis', 'Age', 'Sex']\n","\n","# DataFrame for left-eye images\n","eyes_L = train_df[['Left-Fundus', 'Left-Diagnostic Keywords', 'Patient Age', 'Patient Sex']]\n","eyes_L.columns = col_names\n","\n","# DataFrame for right-eye images\n","eyes_R = train_df[['Right-Fundus', 'Right-Diagnostic Keywords', 'Patient Age', 'Patient Sex']]\n","eyes_R.columns = col_names\n","\n","# DataFrame for left-eye and right-eye images combined\n","eyes_df = pd.concat([eyes_L, eyes_R], axis=0)\n","\n","eyes_df.head()\n"],"execution_count":null,"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>Image</th>\n","      <th>Diagnosis</th>\n","      <th>Age</th>\n","      <th>Sex</th>\n","    </tr>\n","  </thead>\n","  <tbody>\n","    <tr>\n","      <th>0</th>\n","      <td>0_left.jpg</td>\n","      <td>cataract</td>\n","      <td>69</td>\n","      <td>Female</td>\n","    </tr>\n","    <tr>\n","      <th>1</th>\n","      <td>1_left.jpg</td>\n","      <td>normal fundus</td>\n","      <td>57</td>\n","      <td>Male</td>\n","    </tr>\n","    <tr>\n","      <th>2</th>\n","      <td>2_left.jpg</td>\n","      <td>laser spot，moderate non proliferative retinopathy</td>\n","      <td>42</td>\n","      <td>Male</td>\n","    </tr>\n","    <tr>\n","      <th>3</th>\n","      <td>3_left.jpg</td>\n","      <td>normal fundus</td>\n","      <td>66</td>\n","      <td>Male</td>\n","    </tr>\n","    <tr>\n","      <th>4</th>\n","      <td>4_left.jpg</td>\n","      <td>macular epiretinal membrane</td>\n","      <td>53</td>\n","      <td>Male</td>\n","    </tr>\n","  </tbody>\n","</table>\n","</div>"],"text/plain":["        Image                                          Diagnosis  Age     Sex\n","0  0_left.jpg                                           cataract   69  Female\n","1  1_left.jpg                                      normal fundus   57    Male\n","2  2_left.jpg  laser spot，moderate non proliferative retinopathy   42    Male\n","3  3_left.jpg                                      normal fundus   66    Male\n","4  4_left.jpg                        macular epiretinal membrane   53    Male"]},"execution_count":38,"metadata":{},"output_type":"execute_result"}]},{"cell_type":"code","metadata":{"id":"XFuWaWsuqdue"},"source":["def has_disease(disease, text):\n","    if disease in text:\n","        return 1\n","    else:\n","        return 0"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"jEqSmefOqdue"},"source":["#Using Cataract data\n","keywords = ['normal', 'cataract']\n","class_cols = ['Normal','Cataract']\n","\n","for keyword, class_col in zip(keywords, class_cols):\n","    eyes_df[class_col] = eyes_df['Diagnosis'].apply(lambda x: has_disease(keyword, x))  \n"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"lm8hkbWsqduf","outputId":"2cee6074-167c-4f49-f736-51158de29683"},"source":["fig, axis = plt.subplots(2, 1, figsize=(7,7))\n","\n","for ax, class_col in zip(axis.reshape(-1), class_cols):\n","    sns.countplot(ax=ax, x=class_col, hue='Sex', data=eyes_df)\n","    ax.set_ylabel('')\n","plt.subplots_adjust(hspace=0.4, wspace=0.25)\n","fig.text(0.01, 0.5, 'Counts', va='center', rotation='vertical', fontsize='large')\n","plt.show()"],"execution_count":null,"outputs":[{"data":{"image/png":"iVBORw0KGgoAAAANSUhEUgAAAc4AAAG9CAYAAACcUJbfAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4yLjEsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+j8jraAAAgAElEQVR4nOzdd3RU1d7G8e+kTSCUBNIgcAMYeiAqHURAJAREpYoUQUQRGyLtqvd60YvSi5UiCAiEFsEWAygdqQqIREByKUpPMZWQNjPvH7yMDAkkExKSkOezlmuRffY553cGJw+n7W2wWCwWREREJE8ciroAERGRkkTBKSIiYgcFp4iIiB0UnCIiInZQcIqIiNhBwSkiImIHBaeIiIgdnIq6gOIiPv4yZrNeaRUREXBwMODh4ZbjMgXn/zObLQpOERHJlS7VioiI2EHBKSIiYgcFp4iIiB0UnCIiInaw++Gg3377jW+//ZY9e/Zw9uxZMjMzueeee+jTpw9PPvkkBoPBpv/GjRuZM2cOUVFRuLm50aFDB8aMGUOlSpWybfv48eNMnz6d/fv3Y7FYuO+++xg7diz16tXL1jchIYGZM2eyadMmkpOTCQgI4LnnnqNLly72HpIUEx4VXXByMRZ1GSVOVkY68YkZRV2GSKlhsHdasddee41du3bRqVMnGjZsSEZGBuvWrePgwYM88cQTTJgwwdo3IiKC1157jcDAQHr37k1cXBwLFy7Ez8+PsLAwXF1drX1Pnz5N7969cXV1ZfDgwRiNRkJDQ4mJiWH16tUEBARY+2ZkZNC3b1+ioqIYNGgQNWrUYN26dezatYspU6bQvXt3uz+IuLgUPVVbxLy8yrN/6rNFXUaJ02TcAmJikou6DJG7ioODgcqVy+W4zO7g3L9/P4GBgRiNf58ZmM1mBg8ezL59+wgPD6d27dpkZmbSoUMHjEYj4eHhlClTBoBt27YxbNgwxo0bx9ChQ63bGDFiBBs3buSrr76iTp06AMTExBASEkLTpk2ZN2+ete+yZcuYMGECb7/9Nv369bPW0L9/f06fPs3WrVttQjkvFJxFT8GZPwrO25OZmUFycgJZWRmYzaaiLkcKkYODI05OLpQv746zs0sufW8enHZfqm3SpEkOO3AgODiYffv2cfz4cWrXrs2+ffuIiYnhtddes4YmQLt27fD39yc8PNwanJcvX2bz5s20bNnSGpoAXl5edOnShbVr15KQkIC7uzsA4eHhlC1bll69etnU0L9/f8aOHcvOnTvp2LGjvYcmIqXMlSuXSU6Op1y5ihiNlXBwcMx2u0nuDhaLBbPZRHr6FeLjoylf3oMyZXIe4CA3BfZwUHR0NID13mVkZCQAQUFB2fo2btyY33//nYyMq/dljh8/TmZmJo0bN87WNygoCJPJxNGjR4GrZ5ZHjx6lfv36uLi4ZOsLcPjw4QI6KhG5m6WkJOLu7knZsuVxdHRSaN7FDAYDjo5OlC1bHnd3Ty5fTsz3tgpk5KDY2FhWrVqFn5+f9Yz0WpB6e3tn6+/t7Y3JZCI2NpaqVavm2hfg0qVLACQmJpKWlpanvva42Sm5SEng5VW+qEsokaKjTZQpU0aBWco4OpYhIcGU7+/NbQdnRkYGI0aMIDk5mffff996FnjlyhWAbGeFgPX+6LU+t+p74/bS0tJy3e61PvbQPc6ip1/++ad7nPljNpsxmSyAvvuljdlsvuX3pkDvcV4vKyuLV199lQMHDvDf//6X1q1bW5ddC7Frl2Ovl56eDmB9gOdWfa+15aXvjdsVEREpaPm+x2kymRg9ejSbN2/mX//6F0888YTNcl9fX+DvS7bXi46OxtHRES8vrzz1BfDx8QHA3d0do9F4y745XcYVEREpCPk64zSbzYwbN47169fzz3/+k6eeeipbn4YNGwJw6NAhWrVqZbPs8OHD1K5d23q5tU6dOjg7O/Prr79m286hQ4dwdHSkfv36wNWnZxs0aMDRo0fJyMiwuWR76NAhABo1apSfwxIRKbEOHfqFZcsWERV1nMTEBCpUqIi/fw3atm1Pnz5PFnV5dxW7zzjNZjNvvPEG4eHhjBo1imeeeSbHfs2bN8fLy4uwsDCbe47btm3j9OnTdOvWzdrm5uZG+/bt2bNnD1FRUdb2mJgY1q9fT5s2bfDw8LC2d+3aldTUVNasWWNT1/Lly3F3d6dNmzb2HpaISIn144/bePnl57BYLIwYMZoZMz7ihRdewc+vOtu3bynq8u46dp9xTp06la+++opGjRrh6+vL119/bbP8/vvvp3r16ri4uPD6668zevRoBgwYYB05aNGiRQQEBDBw4ECb9UaNGsXu3bsZMmQITz/9NEajkWXLlmEymRg7dqxN3759+7JmzRree+89zp49i7+/v3X0okmTJtm8NyoicrdbsWIZ/v41mDr1fRwc/j4fCgl5BLPZXISV3Z3yNVYtXL3cOm7cuGzLJ02aRPXq1QHo1q0bLi4uzJ07l4kTJ+Lm5kZwcDBjxozJFm61atUiNDSU6dOnM3v2bOtYtbNmzbIZFAGuPiC0ePFiZsyYwdq1a0lJSSEgIIBZs2bRtWtXew9JRKRES0pKpFKlyjahec31benp6SxZspCNGzcQHX0Jd3cPHn64M8899wIuLi6YzWZGjXqZU6dOsHjxCjw8rr6XHxsbw9NP96NBg0CmTJlV6l/fsXvIvbuVXkcpehpyL3805F7+Xbz4B76+/kVdxm2bPHkC3333DQMHPs3DD3emZs1a2ULUZDIxatTLHDt2hEGDnqFu3fqcPPk/FiyYR5MmzZg0aToAcXGxDBkygICAOsyY8SEWi4VXX32BM2f+ZPHiFdYR3Eq63P7uC+11FBERKXrPP/8y58+fY+nSRSxduogyZcoSFHQvHTsGExLyCAaDgU2bfmD//p+YPv1DWra8+upg06bNqVChIu++O57ffoukYcNAKlf25N//fofRo18hNPRz0tPTOXToIO+/P/uuCc3bpeAUESnhPDw8+PDDuRw/foy9e3dz5EgkBw/uZ8+eXWzZsokpU2ayZ89OKlWqTNOmzcnKyrKu26LF1bceDh06QMOGgQA0b96SAQMGM3/+HAAGDx7K/fc3vfMHVkwpOEVE7hJ16tSjTp2r8xenp6cxadIENm7cwO7dO/nrrzj++iuO9u1b5rhuQkKCzc9dunQjNPRznJyc6NWrb6HXXpIoOEVE7kJGoysDBgxm48YNnDp1gooV3fH09GLy5Bk59q9UqbL1z5mZmfz3v2/h51eNhIQEpk59j4kTp92p0os9BWchKF/BFVejc1GXISKlRGxsLJ6entna//jjFACVK3vSsmVltmzZiKOjE7Vr18nW93qzZ3/IqVMnmDt3IRcunOdf/xrH2rVh9OzZp1DqL2kUnIXA1ehM/3GhRV1GibN86oCiLkGkRBo9+hV8fHxo0+ZBqlf/B1lZWRw5EsmKFUupVq067do9hLOzM+vXRzBq1Mv07dufunXrYbHAhQvn2bNnF8OGvUjNmrX48cfthIWt4NVXx1gv/Xbv3ouPP36fxo3vJSCgdlEfbpFTcIqIlHCDBj3D9u2bWb58CXFxsZjNZry9fena9TEGDRpifW9++vQPWLFiKevXR7Bw4ae4uLjg61uV5s1bULmyJ9HRl5g48R0eeOBBm2H6XnllFIcP/8r48W/w2WfLSv1EGgpOEZESrmPHTnTs2CnXfs7Ozgwa9AyDBuU8VGqFChWIiNiUrd1oNPL55ytuu867Rb5nRxERESmNFJwiIiJ2UHCKiIjYQcEpIiJiBwWniIiIHfRUrYhIMePoaCj1U3flh8ViwWQq/FmuFJwiIsWMwWDg8sU/irqMEsfN1x8o/ODUpVoRERE7KDhFRETsoOAUERGxg4JTRETEDgpOEREptj5fsZKO3Xtw8VJ0UZdipadqRURyUNTz6np4NLD+Oe1KOrGnT+R7W78cjmT0W2/ddHn4iuXWGVQkdwpOEZEcFKd5dQtqrtpOHdrTJCgoW7uLi0uBbL+0UHCKiJQS9QJq06l9+6Iuo8RTcIqICCaTibXh4WzYtJmzFy7gajQSFNiQZ596iup+ftZ+n69YyZJVq5g9fRqbtm9n8/YdXElLo3GDBox68QW8PD35KiKCteHhRMfE4l+tGiOeH0bDevWs20i9coUVa9bw88FfuHDpEmnp6VSrUoVHQ0J4vGuXPNV7KSaGJStXsu/AQZKSk/GqXJmQro8yaNBQnJ0L9xK7glNEpJS4kp5GYlKSTZvRaMTVaOTdGTPYuXcfndq34/GuXUhITOLrdet4edw/mTNjOlV9fW3WmzV7DuXLl2fgE324cPESa8PDGT95Cm1bt2Lj1q10C+5MVlYWK9eu5a2Jk1g+/1NcjUYAYuPi2LBpMw+2bk3njg9hyjKxY88ePvz0U5JTUhj4RJ9bHsf5ixd55Z+v4+TkRLfgYCpXqsTR47+zePFnREUdZ/LkmQX7wd1AwSkiUkosWLKUBUuW2rQN6tuXGv+ozvZdu3lrzBjaP9DGuqxjuwd5dsSrLFm5itdHvmqznoe7OxPf+rd1TN2MzAy+WbeexKQkFnz4AWVcXQGoXMmDqR9+xI5du+nUoT0AVXx8WLFgPo6Ojtbt9Xy0G2P/M57VX33Fkz174OR083j6eP4CnJ2cmDdrJhUrVACgW+dgatZpwJw5H7N//080adIs/x9ULhScIiKlRPeuXWndorlNWxUfH+YtXkzFChW4r3EjmzNSt7JlCahVi/2HDmXb1qMhnW0Gom9Qty7frFvPw+3bWUMTILB+fQDOXbxgbbv+UmpmZiZX0tKwWCzcHxTEwcOH+fPsOWrV8M/xGFJSLvPTwYM8FhICYFNvy5atmTPnY376aa+CU0REbl91P78cn6r98+w5EpOS6DlocI7rOThkf+Xfx8vL5udybm7/3+6dY3tycoq1zWKxsPbbcL7dsIGz589jsdgOzJ5yOYWbOXv+HGazma8iIvgqIiLHPvHxf910/YKg4BQRKeUsFguelSszbsQreV4npzC9VbvlullLVn/1FZ9+voQWTZvwZM+eeLhXxMnJib3797Pmm28xm28+w4n5/0M2pGNHHnqwrc0yVw8fTCYznp5eOa1aYBScIiKlnF+VKuw/dIhGDRrgUshPpAJs3r6DKj4+vPevf9lc7j346+Fc163q64vBYMBisWQ7e3bz9Scry1zg9d5IQ+6JiJRyHR98kMzMTJauWpXj8viEhALd37WzUrP575BLTEpi/aZNua7rXrEiTYKC2LxjB6f+yD5naXp6Gqmplwuu2BzojFNEpJTr0PYBdu7bx/Iv1nAs6n80u+9eypQpw6XoaPbuP0DdgADGvPxSge3vgZYtWBi6nH+99x5tWrQgISGR777/Hs9KlfIU0q8Of55X33iTl8aOI+ThjtT09+fKlTQuxieyefMmJk+eSVDQvQVW740UnCIiOUhLzyywoe5uV9qV9ELdvsFg4N+jR3Fvo0DWbdzI5ytWgsGAZ6VKNGrQgG7BnQp0f0/27ElmVhbfb97CL4cjqeLjzZM9e+Lq6sq0jz7Kdf2qvr7MnTGd0C++YO/P+/nu+x9wK1sWv2r/oG/f/tSsWatA672RwXLj40ylVFxcyi1vSNvDy6t8sRnjsiRZPnUA+6c+W9RllDhNxi0gJia5qMsokS5e/ANf35xfeyhKTk4OXL6Y/TKk3Jo99zhz+7t3cDBQuXK5nJflqzoREZFSSpdqRaTQFPXUXLmJjnbAyUnnD2IfBaeIFJriNDVXTp5/rA5p5sJ9WT4/alWrVNQlyC3on1oiIiJ2UHCKiIjYQcEpIiJiBwWniIiIHRScIiIidlBwioiI2EHBKSIiYgcFp4iIiB0UnCIiInZQcIqIyG1bv2kzHbv34JfDkUVdSqHTkHsiIjkIrO2F0dW1yPbv4dHA+ufMtCtcOnUq39v65XAko996C4CBffowZED/W/YZ1Lcvg/s9me/93e0UnCIiOTC6uhabae6ajFtQINtxcXHh+61bebp/PwwGg82yDZs34+LiQkZGRoHs626mS7UiIqVE6+bNiY6J4ZfDh23ar6SlsWP3btq0aF5ElZUsOuMUESklGtWvT9SJE2zYsoX7Gje2tm/buYu09HQ6tWvPlh0/WtvNZjPLv1jDvgMHOHv+PKmpqXh7e/Fwu3YM6N0bR0fHXPeZmJTEstVh/Lh3L3/Fx+NesSIPtGjBkAH9KefmVijHWdgUnCIipUin9u1Z+eWXvDpsGGXKlAHg+82bua9RI7y9PG36ZmVl8cU33/Bg69a0a9MaRwdHfv7lFz5fsZLomFjGvPzSLfeVnJLCK6+/TlJSMo8Ed6KKjy+nz/xJ+PoNHD1+nA8mTcTZufjO13ozCk4RkVIkuEN7Pl+5km27dhPS8SEuXorm1yNH+OeIEdn6Ojs7s+qzBRiNRmtb90e6MvXDj9iweTNP9++HZ6Wbzx26MDSU+PgE5s6cgV+VKtb2BnXq8t7MmWzcto0uDz9csAd4B+gep4hIKeLj7U1QYEO+37wZuPpQkKvRSNvWrbL1NRgM1tA0mUwkp6SQmJTEfY0bYTabOf6/Ezfdj8ViYeuPO7m3USPKubmRmJRk/e/eRoE4Ojqy/9ChwjnIQqYzThGRUia4QwemffQx5y9e5IetW2nXpjWu151VXu+HrVv54utvOPnHH5jNZptlKZdTbrqPhMREkpKT2bVvHz0H7cu5T0Ji/g+iCCk4RURKmQdbt+ajT+cza/YcLly6xLgRr+TYb9vOnUx+/wMa1qvHyOHP41mpMk7OTkSdOMn8JUswmy033YfFcnVZiyZN6PXYozn2Ke9W7vYPpggoOEVESpkyrq60bdWK77dsoYqPD40aNMix36btO3BxcWHaO2/b3Oe8cPFSrvuoWKECbmXLkpaeTpOgoAKrvThQcIqIlEJPdH8cX29v6tYOyDYYwjUODg44GAyYLX+fWWZkZPDVd9/lun1HR0faP/AA333/PT8f/IWm991rs9xkMpF65Qrly5W8s04Fp4hIKVTT35+a/v637PNAyxbs2L2bsePH06lde1LTrvD95i0YXVzytI9nnxrI4SO/8caECXRq3546AfdgMpk4f+EiO3bv5rnBg+jUvn0BHM2dla/gvHz5MosWLSIyMpLIyEhiYmLo3LkzH374YY79N27cyJw5c4iKisLNzY0OHTowZswYKuXwGPPx48eZPn06+/fvx2KxcN999zF27Fjq1auXrW9CQgIzZ85k06ZNJCcnExAQwHPPPUeXLl3yc1giIlbpaWkFNtTd7cpMu1Ik+324XTuSk1P4MuI75ixaRGUPDzp3fIiG9eoxbvzbua5foXx5PpoyhZVrv2TH7t1s3r4dV1dXfLy8CH6oA/cGBhb+QRQCg8Viufnd3Zs4e/YsHTt2xMvLi8DAQLZs2XLT4IyIiOC1114jMDCQ3r17ExcXx8KFC/Hz8yMsLAzX6wZRPn36NL1798bV1ZXBgwdjNBoJDQ0lJiaG1atXExAQYO2bkZFB3759iYqKYtCgQdSoUYN169axa9cupkyZQvfu3e06pri4lFve6LaHl1d5+o8LLZBtlSbLpw4oNmODliRNxi0gJia5qMvIUXH/Ljz/WB3KVfQt6jKyqVWtEpcv/lHUZZQ4br7+ZGWZc+8IXLz4B76+Nz/jdnAwULlyzpeR83XG6e3tzfbt2/Hx8QGgbt26OfbLzMxk4sSJVKtWjWXLlllHqWjUqBHDhg0jNDSUoUOHWvvPnDmT1NRUli9fTp06dQDo0qULISEhTJs2jXnz5ln7rl69miNHjvD222/Tr18/AHr37k3//v2ZPHkyISEhNqEsIiJSEPI1AIKLi4s1NG9l3759xMTE0KdPH2toArRr1w5/f3/Cw8OtbZcvX2bz5s20bNnSGpoAXl5edOnShR07dpCQkGBtDw8Pp2zZsvTq1evvg3FwoH///sTHx7Nz5878HJqIiMgtFerIQZGRVyc0DcrhUeTGjRvz+++/W6ewOX78OJmZmTS+buDha4KCgjCZTBw9ehS4OvDw0aNHqV+/Pi433KS+tq/DN4z+LyIiUhAK9ana6Oho4Oql3Rt5e3tjMpmIjY2latWqufYFuHTp6rtDiYmJpKWl5alvXt3sWrZISeDlVb6oSxApFpyc8nY+6ODgkO/vTaEG55UrV58Eu/GsELC+THutz636Xmu71ictLS3X7V7rk1cF/XCQyJ1UnB8OErmT8vpwkNlsvuX35lYPBxXqpdprIZbTjOLp6ekA1gd4btX3Wlte+t64XRERkYJUqMHp63v1Me9rl2GvFx0djaOjI15eXnnqC1gfSHJ3d8doNN6yb06XcUVErmfBAva/kSclXD7ewrRRqMHZsGFDAA7lMHXM4cOHqV27tvVya506dXB2dubXX3/N1vfQoUM4OjpSv3594Oq16QYNGnD06NFsZ53X9tWoUaMCPRYRufskXc4ES1ZRlyF3WGZmOk5O+Z9Au1CDs3nz5nh5eREWFmZzz3Hbtm2cPn2abt26Wdvc3Nxo3749e/bsISoqytoeExPD+vXradOmDR4eHtb2rl27kpqaypo1a6xtZrOZ5cuX4+7uTps2bQrz0ETkLrD3txiSkuOwmK6Axayzz7uYxWLBZMri8uVkEhJicXOrmO9t5fvhoGXLlpGUlGT9+eTJk8yePRuAZs2a0axZM1xcXHj99dcZPXo0AwYMsI4ctGjRIgICAhg4cKDNNkeNGsXu3bsZMmQITz/9NEajkWXLlmEymRg7dqxN3759+7JmzRree+89zp49i7+/P+vWrePgwYNMmjTJ5r1REZGcnLp4mbDNp2jV0AtvjzIYXRwwkPOA53fSeYdkMpLiirqMEifRTLY5Q6/n4OCIs7MLHh7eODvnbbzdnOQ7OBcuXMi5c+esP0dFRfHBBx8A8PLLL9OsWTMAunXrhouLC3PnzmXixIm4ubkRHBzMmDFjsoVbrVq1CA0NZfr06cyePds6Vu2sWbNsBkWAqw8ILV68mBkzZrB27VpSUlIICAhg1qxZdO3aNb+HJSKlTEJKJuv2ni/qMmxo+Mn8aXiHhp/Md3Bu3rw5z32Dg4MJDg7OU9969eqxYEHeBlb28PDg3Xff5d13381zLSIiIrejUO9xioiI3G0UnCIiInZQcIqIiNhBwSkiImIHBaeIiIgdFJwiIiJ2UHCKiIjYQcEpIiJiBwWniIiIHRScIiIidlBwioiI2EHBKSIiYgcFp4iIiB0UnCIiInZQcIqIiNhBwSkiImIHBaeIiIgdFJwiIiJ2UHCKiIjYQcEpIiJiBwWniIiIHRScIiIidlBwioiI2EHBKSIiYgcFp4iIiB0UnCIiInZQcIqIiNhBwSkiImIHBaeIiIgdFJwiIiJ2UHCKiIjYQcEpIiJiBwWniIiIHRScIiIidlBwioiI2EHBKSIiYgcFp4iIiB0UnCIiInZQcIqIiNhBwSkiImIHBaeIiIgdFJwiIiJ2UHCKiIjYQcEpIiJiBwWniIiIHRScIiIidlBwioiI2EHBKSIiYgcFp4iIiB0UnCIiInZQcIqIiNhBwSkiImIHBaeIiIgdFJwiIiJ2UHCKiIjYQcEpIiJiBwWniIiIHRScIiIidlBwioiI2KFEB6fJZOLTTz+lU6dOBAYG8vDDDzN79myysrKKujQREblLORV1AbdjwoQJrFixgi5dujBs2DAOHz7Mhx9+yNmzZ5k4cWJRlyciInehAg3O9PR09uzZQ3R0NIGBgdSvX78gN2/j999/Z+XKlXTt2pVZs2YB0KdPH8qXL8+CBQt48sknady4caHtX0RESie7L9Vu2bKFZ555hq+++sqm/eLFi/To0YPhw4fzn//8h549ezJlypQCK/RGERERWCwWnnrqKZv2QYMGAfDdd98V2r5FRKT0svuMMzw8nN27dzNmzBib9kmTJnHy5EnKlStHtWrViIqKYvHixbRo0YL27dsXVL1WkZGRODg4EBgYaNPu4+ODj48Phw8ftmt7Dg6GgiwPTw+3At1eaeFSoXJRl1AiFfT/vwVJ34X80Xchfwrqu3Cr7RgsFovFno117tyZ+Ph49u3bZ21LSEjggQcewM3NjW+//RZvb2/Cw8MZM2YMDz30ELNnz85/9Tfx6KOPEhcXx65du7It6927N/Hx8WzatKnA9ysiIqWb3Zdq4+Pj8fHxsWnbs2cPWVlZhISE4O3tDUC3bt2oXLkyv/76a8FUeoMrV67g4uKS4zKj0UhaWlqh7FdEREo3u4MzNTUVZ2dnm7ZffvkFg8FAq1atbNqrVKlCQkLC7VV4E0ajkYyMjByXpaenYzQaC2W/IiJSutkdnO7u7pw/f57rr/Du3LkTgKZNm9r0zcrKws2tcO5v+Pr6Eh8fn2N4RkdHZzsrFhERKQh2B2ejRo1ITExk5cqVAGzbto2oqCgCAgLw9PS09rNYLPz5559Urlw4N7gbNmyI2WwmMjLSpv3SpUtcunQp20NDIiIiBcHu4Bw0aBAWi4X//ve/tGjRghdeeAGDwUD//v1t+v3yyy+kpqbSoEGDAiv2el27dsVgMLB06VKb9iVLlgBX77GKiIgUNLtfR2nVqhUTJkxg+vTpJCYmYjQaGTx4MP369bPpt2bNGmv/wlCvXj2eeOIJVq1ahcVioU2bNkRGRrJq1Sp69OhBUFBQoexXRERKN7tfR7nGbDbz119/UalSJRwcsp+4njhxgszMTGrUqIGrq+ttF5qTrKwsPvvsM8LCwrh48SLe3t707NmT559/PtsDTCIiIgXB7uA8f/48RqMxT/cu4+LiSE9Pp2rVqvkuUEREpDix+x7nQw89xKuvvpqnviNHjuThhx+2uygREZHiKl/TitlzkprPK8EiIiLFUqFOK5aamoqTU4meuUwKkclkst6jvnDhAr6+vvTs2ZNhw4bp/xspVS5fvsyiRYuIjIwkMjKSmJgYOnfuzIcffljUpUkOCm0i6xMnThAVFWUdgk/kRhMmTGDGjBk0bNiQ8ePH07p1az788EP+85//FHVpIndUfHw8H330EZGRkXoHvQTI9Z/1n3/+ufXdyGsiIyPp2LHjTddJT08nLi4OgLZt295miXI30uOOFkEAACAASURBVHyqIn/z9vZm+/bt1hHP6tatW8QVya3kGpzJycmcO3fO+rPBYCA9Pd2m7WYeeOABRo4ceXsVyl3pVvOpLliwgO+++07BKaWGi4uLhgktQXINzh49etC8eXPg6oM+gwcPpk6dOvz73//Osb/BYMBoNFK9enU8PDwKtlq5axT0fKoiIndKrsHp5+eHn5+f9edmzZpRt25da5iK5Ed0dDQeHh45Tg3n7e3NpUuXiqAqEZHc2f3o4o1jw4rkh+ZTFZGSqtCeqhW5Fc2nKiIlVb5fljtz5gxbt27lzz//JDU1FbPZnGM/g8HAxIkT812g3J18fX05efIkGRkZ2c48o6OjbW4PiIgUJ/kKzkmTJrF06VLrqEC3Gh1IwSk5adiwIT/++CORkZHcf//91vZr86l27ty5CKsTEbk5u4Nz8eLFfP7558DVd42CgoLw9PTMcYYUkZvp2rUrn376KUuXLrUJTs2nKiLFnd3B+cUXX2AwGBgzZgxDhw4tjJqkFNB8qiK2li1bRlJSkvXnkydPMnv2bODq2wzNmjUrqtLkBnZPK9a4cWMqVKjAjz/+WFg1SSmh+VRF/vbQQw/ddGCZl19+mVdeeeUOVyQ3Y3dwtm7dmqpVq/LFF18UVk0iIiLFlt03Jps0acKff/5JVlZWYdQjIiJSrNkdnC+99BJpaWnMnTu3MOoREREp1uy+VHv+/Hl+/PFHJkyYQOvWrXniiSeoUaMGZcqUuek6VatWve1CRUREigO7g7N+/fr27cBg4MiRI3atIyIiUlzZ/TqKnTlrd38REZHizO4zThERkdJMw/2IiIjYQcEpIsXC3r17qVu3Lh999FFRlyJySwpOkWLk7Nmz1K1bl7p16zJixIgc+6xdu5a6deuyePHiO1uciAD5eDho0KBBdvU3GAzWQeFFJO++//57Dh8+TKNGjYq6FBG5jt3BuW/fvlz7GAwG4OoTtdf+LCJ5V61aNc6dO8fMmTNZtGhRUZcjItexOzgnTZp002WpqamcPn2aiIgIkpKSeOmll/Dx8bmtAkVKo7p169KwYUM2bNjA7t27adWqVa7r/PTTT8ydO5dff/2V9PR0/P396d69O4MHD8bJ6e+v+t69exk0aBAvv/wyLVu25KOPPiIyMhJ3d3c2b97M66+/zpdffsnGjRtZt24dq1evJjo6mpo1azJ69GgefPBBUlJSmDFjBj/88ANJSUnce++9vP3229SqVcumph9++IHvvvuOw4cPExMTg9FopGHDhgwbNozWrVsX+OcmcifYHZw9evTItc+rr77KmDFjWLVqFWvWrMlXYSKl3ciRI9m4cSMzZszIdVKFiIgIxowZg6urK127dqVChQps376dqVOnsn//fj755JNsV38OHDjAvHnzaNOmDf369ePy5cs2yydOnMhvv/1Ghw4dMJlMfPvtt7z44ossX76cd955B5PJRNeuXTl37hwbN25k+PDhrFu3DkdHR+s2ZsyYgYuLC82bN8fT05OYmBg2btzI0KFD+eCDDwgODi64D0zkDrE7OPOiXLlyTJw4kQcffJCPPvqI8ePHF8ZuRO5qtWrVomfPnoSFhbFhwwY6d+6cY7+UlBT+85//YDQaCQsL45577gFg1KhRPP/882zatImvv/6a7t2726y3a9cupk+fzqOPPprjdk+fPs3XX3+Nh4cHAG3btmXEiBEMHTqUtm3bMm3aNGtITpgwgWXLlrFp0yabMJw/fz7Vq1e32W5sbCw9e/Zk2rRpCk4pkQrtqdpKlSpRu3ZttmzZUli7ELnrvfLKKxiNRt5//31MJlOOfTZu3EhycjJPPPGENTQBnJycGDNmDABffvlltvUCAwNvGpoAzz//vDU0ATp16oSzszNJSUmMHTvW5syyS5cuABw7dsxmGzeGJoCnpyfBwcH8+eefnD179qb7FymuCuWM85rLly8TFxdXmLsQuav5+PgwYMAAFi5cyJdffknv3r2z9Tl69CgAzZs3z7asfv36lC9fPlugwdXgvJV69erZ/Ozg4EClSpVIS0ujSpUqNsu8vLwAiI6OtmmPiYlh3rx57NixgwsXLpCenp5tebVq1W5Zh0hxU2jB+fPPP3PmzJlsXzARsc/zzz9PWFgYH3/8MY899li25SkpKcDVM7mceHp6cubMmWztlStXvuV+y5Url63Nyckpx/ZrZ5/Xz9ObkJBAnz59uHjxIk2aNKFt27aUL18eBwcH9u3bx759+8jIyLhlDSLFkd3B+dNPP910mcViITY2ll9//ZWwsDAAHn744fxXJyK4u7vzzDPP8MEHHxAaGkrFihVtll8LstjY2BzXj4uLyzHsCvtVsS+++IILFy5Y77Veb/z48Xl6tU2kOLI7OJ966qk8feEsFguBgYE3Hf1ERPLu6aefJjQ0lHnz5vHiiy/aLLs21d/PP/9Mx44dbZYdO3aMpKQkWrZsecdqvebPP/8EoEOHDjbtFouFX3755Y7XI1JQ7A7OW01KbTAYKFu2LP7+/rRv357u3bvbvD8mIvlTtmxZXnjhBSZMmMDSpUttlnXs2JFy5cqxatUqnnzySfz9/QEwmUxMnz4dINsTtXfCtd8VBw4coE6dOtb2zz//PMd7riIlhd2ptnnz5sKoQ0Ry0bdvXxYtWmQ9k7umfPnyvPPOO4wdO5aePXvyyCOPUL58ebZv387x48fp0KFDkQTn448/zvz585kwYQJ79+7F19eXyMhIfvnlF9q3b8/WrVvveE0iBUGDvIuUEM7Ozje99dGtWzcWL17MvffeS0REBEuWLMFsNjN69Gg++uijIhn6skqVKixdupTmzZvz448/snr1alxcXAgNDc31iV6R4kwTWYuIiNjhtm5Abtu2jW3btnHy5EkuX76Mm5sbtWrVol27drRr166gahQRESk28nXGef78eUaOHMnhw4eBq0/JWTf4/5eEAgMDmTVrll5uFhGRu4rdwZmSkkKPHj04c+YMjo6OdOzYkTp16uDp6UlsbCzHjx9n06ZNmEwmqlevzpdffpnjO2QiIiIlkd2XahctWsSZM2do0KABH3zwQY5jUZ45c4ZXX32Vo0ePsmjRIl555ZUCKVZERKSo2X3G+eijj3Ly5Em+//57/Pz8btrv3LlzBAcHU6tWLb799tvbLlRERKQ4sPt1lLNnz1K7du1bhiaAn58ftWvX1uwHIiJyV8nXe5w3m97oRmazOT+bFxERKbbsDk5/f39OnDiRbfSSG/3xxx/873//sw7/JSIicjewOziDg4Mxm8289NJLHD9+PMc+//vf/3jllVewWCya4V1ERO4qdj8clJqaSq9evTh16hQODg60atWK2rVr4+npSVxcHMePH2f37t2YzWZq1qzJmjVrKFu2bGHVLyIickflawCE6OhoRo8ebZ2b8/pxMK9trlmzZkyfPh0fH58CKlVERKTo3dZYtT///DPbtm3j1KlT1iH3atasSbt27WjatGlB1ikiIlIsaJB3ERERO+Tp4aD33nuPHj16sHjx4jxtdPHixfTo0YMpU6bcTm0iIiLFTq5nnCdOnKBbt25Uq1aN7777DhcXl1w3mp6eTrdu3Th37hzffvst99xzT4EVLCIiUpRyPeP8+uuvARg+fHieQhPAaDTywgsvYDabreuLiIjcDXINzn379uHo6Ejnzp3t2nBwcDBOTk7s3bs338WJiIgUN7nOjnLq1CmqVatm99Rg5cqVo3r16pw+fTq/td1R8fGXMZv1nJSIiICDgwEPD7ccl+UanJcvX6ZGjRr52nH58uU5c+ZMvta908xmi4JTRERyleul2nLlypGQkJCvjScmJmoSaxERuavkGpxVq1blzJkzdodnfHw8Z86coWrVqvkuTkREpLjJNThbtGiBxWJh1apVdm141apVmM1mWrRoke/iREREiptcg7N3794YDAbmzJnDwYMH87TRAwcOMHfuXBwdHenTp89tFykiIlJc5Ppw0D333EO/fv0IDQ1l8ODBDB8+nH79+uHh4ZGtb3x8PMuXL2fevHlkZmYyYMAAatWqVSiFy93Ho6ILTi7Goi6jxMnKSCc+MaOoyxApNfI0Vm1WVhYvvfQS27Ztw2Aw4OjoSEBAANWrV6ds2bKkpqZy9uxZoqKiMJlMWCwW2rdvzyeffIKjo+OdOI7bFheXoqdqi5iXV3n2T322qMsocZqMW0BMTHJRlyFyV3FwMFC5cs4Pt+Z6xgng5OTE3LlzmT9/Pp999hmJiYkcO3aMY8eOYTAYuD57K1asyLPPPsuzzz5rM92YiEhxlJmZQXJyAllZGZjNpqIuRwqRg4MjTk4ulC/vjrNz3kbCy0meghOuzrk5bNgwBg4cyPbt29m/fz+XLl0iJSWFcuXK4ePjQ5MmTXjwwQc1cbWIlAhXrlwmOTmecuUqYjRWwsHBUf/gv0tZLBbMZhPp6VeIj4+mfHkPypTJeYCD3OQ5OK8pW7YsISEhhISE5GuHIiLFRUpKIu7unri4uBZ1KVLIrt5mdKJs2fI4OTmTlPRXvoMzT9OKiYjcjUymTJyd9UBaaePsbCQrKzPf6ys4RaRU06XZ0ud2/84VnCIiInZQcIqIiNjB7oeDRESk+Dl06BeWLVtEVNRxEhMTqFChIv7+NWjbtj19+jxZ1OXdVRScIiIl3I8/buONN8bQokUrRowYjbu7O9HRlzh06Be2b9+i4CxgCk4RkRJuxYpl+PvXYOrU93Fw+PsOXEjII5jN5iKs7O6ke5wiIiVcUlIilSpVtgnNa65vS09PZ/78OfTt250OHVrRo0dXPvnkAzIyro51bDabGTnyRR5/vDPx8X9Z14uNjaFbt4cZN24keRil9a6n4BQRKeEaNmzEwYP7mTfvE06c+F+OZ5kmk4lx40byxRcreeyxHkyb9gH9+g3k66/XMn78m8DVkH3rrf9iscCECeP/f7QdM++882+cnJx588239foOulQrIlLiPf/8y5w/f46lSxexdOkiypQpS1DQvXTsGExIyCMYDAY2bfqB/ft/Yvr0D2nZsjUATZs2p0KFirz77nh++y2Shg0DqVzZk3//+x1Gj36F0NDPSU9P59Chg7z//mzc3d2L+EiLBwWniEgJ5+HhwYcfzuX48WPs3bubI0ciOXhwP3v27GLLlk1MmTKTPXt2UqlSZZo2bU5WVpZ13RYtWgFw6NABGjYMBKB585YMGDCY+fPnADB48FDuv7/pnT+wYkrBKSJyl6hTpx516tQDID09jUmTJrBx4wZ2797JX3/F8ddfcbRv3zLHdRMSEmx+7tKlG6Ghn+Pk5ESvXn0LvfaSRMEpInIXMhpdGTBgMBs3buDUqRNUrOiOp6cXkyfPyLF/pUqVrX/OzMzkv/99Cz+/aiQkJDB16ntMnDjtTpVe7Ck4RURKuNjYWDw9PbO1//HHKQAqV/akZcvKbNmyEUdHJ2rXrnPL7c2e/SGnTp1g7tyFXLhwnn/9axxr14bRs2efQqm/pFFwioiUcKNHv4KPjw9t2jxI9er/ICsriyNHIlmxYinVqlWnXbuHcHZ2Zv36CEaNepm+fftTt249LBa4cOE8e/bsYtiwF6lZsxY//ridsLAVvPrqGOul3+7de/Hxx+/TuPG9BATULurDLXIGi17KASAuLgWzWR9FUfLyKs/+qc8WdRklTpNxC4iJSS7qMkqkixf/wNfXv6jLuG2bNv3A9u2bOXbsKHFxsZjNZry9fWnVqg2DBg3Bw6MScPUS7IoVS/n++/WcP38WFxcXfH2r0rx5CwYOHEJa2hWefro/jRsHMXnyTOv209PTGTbsabKyMvnss2W4upb8+Utz+7t3cDBQuXK5HJfZHZwnT57k448/5rfffiMmJgaLxUK1atXo0qULgwYNolw52x1t3LiROXPmEBUVhZubGx06dGDMmDFUqlQp27aPHz/O9OnT2b9/PxaLhfvuu4+xY8dSr169bH0TEhKYOXMmmzZtIjk5mYCAAJ577jm6dOliz+FYKTiLnoIzfxSc+Xe3BKfY73aC0+5LtZcuXSIuLo6QkBB8fHwwGAxERkYye/ZsNm7cyMqVK3FxcQEgIiKC1157jcDAQN544w3i4uJYuHAhhw8fJiwszOZfLadPn6Z///64uroyfPhwjEYjoaGh9O/fn9WrVxMQEGDtm5GRwZAhQ4iKimLQoEHUqFGDdevWMXLkSNLT0+nevbu9hyUiIpIndgdnq1ataNWqVbb2mjVrMm3aNHbs2EHHjh3JzMxk4sSJVKtWjWXLllGmTBkAGjVqxLBhwwgNDWXo0KHW9WfOnElqairLly+nTp2rN667dOlCSEgI06ZNY968eda+q1ev5siRI7z99tv069cPgN69e9O/f38mT55MSEjIXXEpQUREip8CG3LPz88PgKSkJAD27dtHTEwMffr0sYYmQLt27fD39yc8PNzadvnyZTZv3kzLli2toQng5eVFly5d2LFjh807RuHh4ZQtW5ZevXr9fSAODvTv35/4+Hh27txZUIclIiJiI99P1aalpZGamkp6ejpHjhxh+vTpuLi40Lx5cwAiIyMBCAoKyrZu48aNiYiIICMjAxcXF44fP05mZiaNGzfO1jcoKIiwsDCOHj1Kq1atMJvNHD16lIYNG1ovCV/fF+Dw4cN07Ngxv4d228pXcMXV6Fxk+xcRkcKT7+BcsmQJM2b8/SLtPffcwyeffGI984yOjgbA29s727re3t6YTCZiY2OpWrVqrn3h6r1VgMTERNLS0vLUt6i4Gp3pPy60SGsoiZZPHVDUJYiI5CrfwfnII48QGBhIUlISBw4cYN++faSmplqXX7lyBSDbWSGA0Wi06XOrvtfarvVJS0vLdbvX+tjjZk9PiZQEXl7li7qEEik62gEnJ00SVRo5ODjk+3uT7+D08/Oznl2GhIQQHh7Oq6++yqJFi2jdurU1xK7N83a99PR0AOsDPLfqe60tL31v3K49CvJ1FP0SkztNr6Pkj9lsJitLEz2XRmaz+Zbfm1u9jlJg/9Tq3Lkzzs7OrFmzBgBfX1/g70u214uOjsbR0REvL6889QXw8fEBwN3dHaPReMu+OV3GFRERKQgFFpxZWVmYTCbrU7UNGzYE4NChQ9n6Hj58mNq1a1svt9apUwdnZ2d+/fXXbH0PHTqEo6Mj9evXv1qwgwMNGjTg6NGj2c46r+2rUaNGBXVYIiIiNuwOztjY2BzbV65cidlstj7Z2rx5c7y8vAgLC7O557ht2zZOnz5Nt27drG1ubm60b9+ePXv2EBUVZW2PiYlh/fr1tGnTBg8PD2t7165dSU1NtZ7dwtXT7uXLl+Pu7k6bNm3sPSwREZE8sfse5/jx44mLi6Nly5ZUrVqVlJQU9u3bx5YtW7jnnnsYPHgwcPXhnddff53Ro0czYMAAevfuTVxcHIsWLSIgIICBAwfabHfUqFHs3r2bIUOG8PTTT2M0Glm2bBkmk4mxY8fa9O3bty9r1qzhvffe4+zZs/j7+7Nu3ToOHjzIpEmTbN4bFRERKUh2B+cjjzzCl19+yZo1a4iPj8fJyQl/f39efvllhgwZYjNWbbdu3XBxcWHu3LlMnDgRNzc3goODGTNmTLZwq1WrFqGhoUyfPp3Zs2dbx6qdNWuWzaAIcPUBocWLFzNjxgzWrl1LSkoKAQEBzJo1i65du+bzoxARkeLms8/msWjRfMLCvqFKlapFXQ6Qj+Ds2rWrXeEUHBxMcHBwnvrWq1ePBQsW5Kmvh4cH7777Lu+++26eaxERyaviNJBJWnomyUn2v2Z3zYEDPzNixPCbLv/+++2ULVs239svbTQfp4hIDorTQCbLpw4gmfwH5zUhIY/QrFmLbO05vRcvN6fgFBEpJerXb0DnzrqddbsUnCIigslkIixsBRER33L27BmMRlfuu68Jw4e/xD/+UcPa79o9xwULlvDDD+v54YcNXLmSSlDQfYwb9y+8vX1Ys2Y1YWEriY6+iL9/DUaN+ieNGv09bnlq6mWWLl3Mvn17OH/+HOnpaVSrVp3u3XvTs2efPNV78eJFFi6cx969u0hMTMTLy4fg4BCefvpZnJ0L9xK7glNEpJS4cuWKzUxTcHWkNVdXV8aPf5MdO7bSuXNXevbsQ0JCAmvXhjFs2BA++2wpfn7VbNabOnUiFSpUYPDgoZw/f44vvljJm2+OpX37h9iwIYLHH+9JVlYmoaGf88Ybo/nii3DrqG4xMTFERHxLhw4d6dr1UUymLLZt28LMmVNISkrk6advPaH9uXNnGT78GZycnHjssZ54enrx22+HWbJkISdORDF58syC/eBuoOAUESkl5s79mLlzP7ZpGzLkOWrWvIetWzfxzjuT6Nixk3VZp04hDBrUl4ULP+Wtt/5rs16lSpWYNu0DDAYDcHUY1C+/DCMxMYElS1ZZ35zw9PTivffeZuvWTYSEPAJA1ap+rFkTjpPT3xHUp08/Ro58kRUrljJw4NM2y270/vvTcHZ2ZtGiUCpWdAfg8cd7Ur36P5g37xP27/+JJk2a3cYndWsKThGRUqJXryd44IF2Nm1Vq/rxyScf4O7uTpMmzWzOSN3cylGnTl1+/nlvtm11797LGpoAgYGN+PLLMDp37mrzuuG1S7Tnzp21tl1/KTUzM5MrV1Ixmy00bdqc/ft/4o8/TnPPPQE5HkNycjJ79+6mR4/eWCzY1NuiRSvmzfuEn37aq+AUEZHb949/+Of4VO2ff54mISGBbt0eznE9B4fsg8z5+FSx+blcuauTW1wbe/zG9qSkRGubxWIhLGwFX321hjNn/sRisZ1gIzk56abHcObMH5jNZtasWc2aNatz7BMf/9dN1y8ICk4RkVLObDbj5eXNm2+Oz/M6jo45j9jq4OCYY/v14bhixVJmz/6Q1q0fYODAp/HwqISTkxN79uxk1arlmM03n7Hm2ixWjzzyGA8/3DnHPp6eXnk9jHxRcIqIlHLVqv2Dn3/eS1DQfXfknc4fflhPlSp+TJkyy+Zy74EDP+e6rp9fNQwGAxaLJcez5ztBM7iKiJRynTp1JiMjg0WL5ue4vKAvfTo4OGIwYHNmmZiYwHfffZPruh4eHjRr1oKNGzdw8uT/si1PT08jNfVygdZ7I51xioiUcg8/3Jkff9zG0qWLOHr0N5o3b0XZsmW5ePECu3fvpH79Brz++lsFtr8HH2zP/PlzGDfuNdq2bUdCQjzffPMlnp5e/PVXXK7rjx79Oi++OJTnnhvMI488Rq1aAVy5coU///yDrVs3MXnyTIKC7i2wem+k4BQRyUFaeibLpw4o6jKAq7UUJoPBwNtvT+S++5ry3Xdfs3DhPAwGA56eXgQF3cdjj/Uo0P0NGDCYrKws1q0L5+DBn6lSpSoDBgyiTJmyTJz4Tq7r+/lV47PPQlmy5DN2797JN998Sbly5ahSxY++fftTs2atAq33RgbLjY8zlVJxcSnWm863y8urfLEZ47IkWT51APun3vrFZ8muybgFxMQkF3UZJdLFi3/g6+tf1GVIEcjt797BwUDlyuVyXlZYRYmIiNyNFJwiIiJ2UHCKiIjYQcEpIiJiBwWniIiIHRScIiIidlBwioiI2EHBKSIiYgcFp4iIiB0UnCIiInZQcIqIiNhBwSkiImIHBaeIiNy2iIhveeCBpnmajLqk07RiIiI58KjogpOLsajLACArI534xIx8r3/gwM+MGDEcgMGDh/Lccy/css+QIc8xdOjz+d7f3U7BKSKSAycXY7GZ5q7JuAVA/oPzGhcXI+vXf8ezzw7HYDDYLFu3LhwXFyMZGem3vZ+7nd2Xan/77TcmT55M9+7dadq0KUFBQfTs2ZMVK1aQ09SeGzdupFevXjRu3JhWrVrx5ptv8tdff+W47ePHjzNs2DCaNGnC/fffz9ChQzl27FiOfRMSEvjPf/5DmzZtaNy4MT179mTdunX2Ho6ISKnRtu2DXLp0Mdvl1CtXrrB162batm1XRJWVLHafcS5YsIBdu3bRqVMn+vbtS0ZGBuvWrePtt9/myJEjTJgwwdo3IiKC1157jcDAQN544w3i4uJYuHAhhw8fJiwsDFdXV2vf06dP079/f1xdXRk+fDhGo5HQ0FD69+/P6tWrCQgIsPbNyMhgyJAhREVFMWjQIGrUqMG6desYOXIk6enpdO/e/TY/FhGRu0/jxvfy++/HWLcunCZNmlnbt2zZSFraFUJCurJp0/fWdrPZzJIlC9m7dxdnzvzJ5cuX8fGpQufOXXjqqSE4OeUeIYmJCSxe/Bnbt28hLi4WD49KPPhge5577kXKlct5oujizu7gHDhwIJMnT8Zo/Pva/1NPPcXgwYNZvXo1gwYNonbt2mRmZjJx4kSqVavGsmXLKFOmDACNGjVi2LBhhIaGMnToUOs2Zs6cSWpqKsuXL6dOnToAdOnShZCQEKZNm8a8efOsfVevXs2RI0d4++236devHwC9e/emf//+TJ48mZCQEJtQFhGRq0JCHiE09HNGjfonZcuWBa4+2HP//c3w8fG16ZuZmcmqVcvp0KEjHTp0wtHRgX379vDZZ/O4dOkir7/+1i33lZSUxPPPP0NiYgKPPdaDqlX9OHXqJF9/vYbffotkzpzPcHZ2LrRjLSx2X6pt0qSJTWgCODg4EBwcDFy93Aqwb98+YmJi6NOnjzU0Adq1a4e/vz/h4eHWtsuXL7N582ZatmxpDU0ALy8vunTpwo4dO0hISLC2h4eHU7ZsWXr16mVTQ//+/YmPj2fnzp32HpaISKkQEvIIaWlpbN26CYALF85z6NBBunR5JFtfFxcXvvoqgnHj/sUTT/SjV6++TJkyi65dHyUi4ltiY2Nuua/58+fw119xzJ//OS+88AqPP96TkSPH8Oab4zl27AgbNkQUyjEWtgJ7HSU6OhqASpUqARAZGQlAUFBQtr6NGzfm999/JyPj6s3u48ePk5mZSePGjbP1DQoKwmQyAkA+AwAAFnhJREFUcfToUeDqpYOjR49Sv359XFxcsvUFOHz4cAEdlYjI3cXXtwr33deEiIhvgatnm66uZWjfvmO2vgaDAaPx6tW7rKwskpKSSEhIoEmTZpjNZo4dO3rT/VgsFjZv/p4mTZpSrlx5EhISrP/df39THB0d+emnvYVzkIWsQJ6qjY2NZdWqVfj5+dGkSRPg7yD19vbO1t/b2xuTyURsbCxVq1bNtS/ApUuXAEhMTCQtLS1Pfe1RuXLJvNYuAuDlVb6oSyiRoqMdcHIqGa+z306djo5X13VwMODk5MAjjzzKu+++zaVL59mwIYKOHR+mXLmy2foBrFv3HStWLOPEif9hMplstpuammLt5+BgsO7LycmBv/76i8TERHbs2MaOHdtyrCsxMb7IPn8HB4d8f29uOzgzMjIYMWIEycnJvP/++9azwCtXrgBkOysErJd6r/W5Vd8bt5eWlpbrdq/1sUdcXApmc/angvNDv8TkTouJSS7qEkoks9lMVpa5qMvIk9up02S6uq7ZbCEry8yDDz6Eq+sUJk58l/Pnz/Hmm+PJyjJn67d580beeectGjVqzOjRr+Pl5YWTkzPHjx9jzpyPyMoyWeu69vvTZLr6mWZmZgHQqlWb/2vv7qOiqvM/gL+ZkRkM5SEdhuzBXdJBRMYFD6hQLVBKTHoKNUk0pSdyNysLRa1OeeSk1PqwmcvJslBCEB/I0/GBzYXNWoG1XQiZxFYXdUGDQQKJx4GZ+f3Bj8lxeLrjyHXk/fpv7vd7v/P5YvDu3vu992L+/Lge6xo50k20n7/RaOzz90Yicer1gOqGgrOzsxOvvvoqiouLsW7dOoSGhprbukOs+3Tstdrbu+4T6l7A01ff7m0D6Xv9uEREZG348OEID4/E0aOHMGbM3Zg8ObDHfseO5UImk+PPf041n7IFgMuXL/X7HR4enhgxYgTa2toQHDzVbrXfCmwOToPBgMTEROTn5+Ott97C/PnzLdq9vbtWZ+l0Otx///0WbTqdDlKpFAqFwqrv9bq3KZVKAICHhwfkcnmffXs6jUtERL9asGARvL3vgp+fv9XDELpJpRJIJE4WZ+Pa29tx4EB2v+NLpVJERs7Al19+gZMnixASMs2ivbOzEy0tLXBzc7uxiYjApuA0Go1ISkpCbm4uVq1ahaefftqqj7+/PwCgtLQU06dPt2grKyvD+PHjzadbVSoVnJ2dcerUKatxSktLIZVK4efnB6DrvPTEiRNRXl4OvV5vccq2tLQUQNctL0RE1Dsfn3Hw8RnXZ5+HHorA11/nY/nyPyIqSoOWlmbk5h6GbICPIly6dBlKS0uwcuWriIrSwNfXDwaDAZcuVeLrr/Pxxz++gqgojT2mM6gEB6fRaMSaNWtw6NAhvP7663j22Wd77BcSEgKFQoF9+/YhPj7efPr0+PHjuHDhAlasWGHu6+rqivDwcOTn5+Ps2bMYP348AKC2tha5ubkICwuDp6enub9Go0FJSQkOHDhgvo/TaDQiMzMTHh4eCAsLEzotIiILnfr2/3/Unfg6RXoM3syZ0WhsvIr9+/di27YtGDVqNDSa2Zg0SY3XXnup3/3d3NyxfftO7N69C19/nYdjx3Lh4jIc3t7eiI6ehcDAKYMwC/tzMvX0nLw+pKSkIC0tDQEBAT0eaQYFBeHee+8F0HW/ZWJiIiZNmoR58+ahrq4OaWlp8Pb2xv79+y3u76yoqDDf8xkfHw+5XI6MjAzodDpkZ2db3N/Z3t6O+fPn47///S+WLFmCsWPH4ujRoygoKMCGDRswZ84cwT8Iey8OikvabZexhpLM9xfeMs8GdSRTknZwcZCNqqsvwtt7rNhlkAj6+7e36+KgH374AUDX6dakpCSr9g0bNpiDc9asWZDJZPjoo4+wfv16uLq6YubMmVixYoVFaAKAj48Pdu/ejY0bNyI1NRUmkwmBgYHYsmWLRWgCXQuEdu7ciU2bNiEnJwdNTU0YN24ctmzZAo3G8Q77iYjIcQg+4rxd8YhTfDzitA2POG3HI86h60aOOB3jzl8iIqJbBIOTiIhIAAYnERGRAAxOIiIiARicRDSkcX3k0HOj/+YMTiIasqRSZ3R0iPNwARJPR0c7hg2z/QXaDE4iGrJGjHBHQ8MVNDf/AoOhk0eftzGTyQSDoRPNzb+goeEKXF3dbR7LLu/jJCJyRMOHu2LYMGc0NTWgufkqjEZD/zuRw5JIpHB2lsHT0wvOztavphwoBicRDWndf0iJBoqnaomIiARgcBIREQnA4CQiIhKAwUlERCQAg5OIiEgABicREZEADE4iIiIBGJxEREQCMDiJiIgEYHASEREJwOAkIiISgMFJREQkAIOTiIhIAAYnERGRAAxOIiIiARicREREAjA4iYiIBGBwEhERCWBTcDY3N2Pbtm1YunQpHnjgAfj6+uKVV17ptf/f/vY3zJ07F2q1GtOnT8cbb7yBn3/+uce+//nPf5CQkIApU6YgKCgIzz33HM6cOdNj34aGBrz99tsICwuDWq3GnDlzcPToUVumRERENCA2BWd9fT0+/PBDaLVaTJo0qc++R44cwUsvvQQAWLNmDRYuXIjc3FwsWbIEbW1tFn0vXLiAuLg4nD59GkuXLsXy5ctRVVWFuLg4nDt3zqKvXq/HM888g5ycHDz++ON466234O7ujuXLl+PgwYO2TIuIiKhfw2zZycvLC9988w2USiUAwNfXt8d+HR0dWL9+Pe655x5kZGRg+PDhAICAgAAkJCRg9+7deO6558z9N2/ejJaWFmRmZkKlUgEAoqOj8eijj+JPf/oTtm/fbu67d+9enD59GmvXrsWCBQsAAPPmzUNcXBxSUlLw6KOPwsXFxZbpERER9cqmI06ZTGYOzb6cPHkStbW1ePLJJ82hCQC///3vMXbsWBw6dMi8rbm5Gfn5+Zg2bZo5NAFAoVAgOjoa3377LRoaGszbDx06hDvuuANz5879dTISCeLi4lBfX48TJ07YMjUiIqI+3dTFQVqtFgAwefJkqza1Wo0ff/wRer0eQNe1zY6ODqjVaqu+kydPhsFgQHl5OQDAaDSivLwcfn5+kMlkVn0BoKyszK5zISIiAm5ycOp0OgBdp3av5+XlBYPBgCtXrgyoLwDU1NQAAK5evYq2trYB9SUiIrInm65xDlRraysAWB0VAoBcLrfo01ff7m3dfboXFfU17vULj/ozatQIQf2JbiUKxUixSyAaMm5qcHaHWPfp2Gu1t7cDgHkBT199u7cNpO/14w5UXV0TjEaToH16wz9iNNhqa38RuwSi24pE4tTrAdVNPVXr7e0N4NfTsNfS6XSQSqVQKBQD6gvAvCDJw8MDcrm8z749ncYlIiK6UTc1OP39/QEApaWlVm1lZWUYP368+XSrSqWCs7MzTp06ZdW3tLQUUqkUfn5+ALpWz06cOBHl5eVWR53d3xUQEGDXuRAREQE3OThDQkKgUCiwb98+i2uOx48fx4ULFzBr1izzNldXV4SHh6OoqAhnz541b6+trUVubi7CwsLg6elp3q7RaNDS0oIDBw6YtxmNRmRmZsLDwwNhYWE3c2pERDRE2XyNMyMjA42NjebPFRUVSE1NBQAEBwcjODgYMpkMq1evRmJiIhYuXIh58+ahrq4OaWlpGDduHBYtWmQx5uuvv47CwkI888wziI+Ph1wuR0ZGBgwGA1auXGnRNzY2FgcOHMC7776LqqoqjB07FkePHkVJSQk2bNhgcd8oERGRvTiZTCabVsRERkbi0qVLPbYtW7YML7/8svnzV199hY8++ghnz56Fq6srIiIisGLFCowaNcpq3zNnzmDjxo0oLi6GyWRCYGAgVqxYgYkTJ1r1ra+vx6ZNm5CXl4empiaMGzcOL7zwAjQajeD52HtxUFzSbruMNZRkvr8Q/37/ebHLcDhTknZwcRCRnfW1OMjm4LzdMDjFx+C0DYOTyP5EW1VLRER0u2FwEhERCcDgJCIiEoDBSUREJACDk4iISAAGJxERkQAMTiIiIgEYnERERAIwOImIiARgcBIREQnA4CQiIhKAwUlERCQAg5OIiEgABicREZEADE4iIiIBGJxEREQCMDiJiIgEYHASEREJwOAkIiISgMFJREQkAIOTiIhIAAYnERGRAAxOIiIiARicREREAjA4iYiIBGBwEhERCTBM7AKI6PY10s0FLnJnsctwOG3tHfilsU3sMqgXDE4iumlc5M6IS9otdhkOJ/P9hfgFDM5blUMHp8FgwKeffop9+/bhp59+gre3N+bMmYOEhAQMG+bQUyOiIczY2QGFYqTYZTicTn076q/qb/r3OHS6JCcnIysrC9HR0UhISEBZWRm2bt2KqqoqrF+/XuzyiIhsIhnmjH+//7zYZTicKUk7ADA4e/Xjjz9iz5490Gg02LJlCwDgySefxMiRI7Fjxw489dRTUKvVIldJRES3G4ddVXvkyBGYTCY8/fTTFtsXL14MADh8+LAYZRER0W3OYY84tVotJBIJJk2aZLFdqVRCqVSirKxM0HgSiZM9y8NoT1e7jjdUyNxGiV2CQ7L3f7/2xN8F2/B3wTb2+l3oaxwnk8lkssu3DLLZs2ejrq4OBQUFVm3z5s1DfX098vLyRKiMiIhuZw57qra1tRUymazHNrlcjrY2LuUmIiL7c9jglMvl0Ot7Xj3V3t4OuVw+yBUREdFQ4LDB6e3tjfr6+h7DU6fTQalUilAVERHd7hw2OP39/WE0GqHVai2219TUoKamxmrREBERkT04bHBqNBo4OTnh888/t9ienp4OAJg1a5YYZRER0W3OYW9HmTBhAubPn4/s7GyYTCaEhYVBq9UiOzsbMTExmDx5stglEhHRbchhb0cBgM7OTvOzaqurq+Hl5YU5c+bgxRdfhLMz38hARET259DBSURENNgc9honERGRGBicREREAjjs4iByfHyfKlGX5uZmpKWlQavVQqvVora2FlFRUdi6davYpVEPeMRJoklOTsamTZvg7++Pd955B6Ghodi6dSvefvttsUsjGlT19fX48MMPodVqeQ+6A+D/1pMo+D5Vol95eXnhm2++MT/xzNfXV+SKqC884iRR8H2qRL+SyWR8TKgDYXCSKOz9PlUiosHC4CRR6HQ6eHp69vhqOC8vL9TU1IhQFRFR/xicJAq+T5WIHBWDk0TB96kSkaNicJIo+D5VInJUDE4SBd+nSkSOisFJouD7VInIUfEBCCQKvk+VyFJGRgYaGxvNnysqKpCamgoACA4ORnBwsFil0XX4WjESDd+nSvSryMhIXLp0qce2ZcuW4eWXXx7kiqg3DE4iIiIBeI2TiIhIAAYnERGRAAxOIiIiARicREREAjA4iYiIBGBwEhERCcDgJCIiEoDBSUREJACDk0hExcXFSEpKwsMPPwy1Wo3f/e53iI6Oxtq1a/HDDz/YNGZVVRV8fX2xevVqO1c7uFavXg1fX19UVVWJXQqRBT6rlkgEBoMBycnJyMrKgkwmw/Tp0zFz5kw4OTmhoqICBw8exJ49e/CXv/wFDz/8sNjlEtE1GJxEIti4cSOysrKgVqvxwQcfYMyYMRbtDQ0N2LZtm8VDv4no1sBn1RINsvPnz0Oj0cDDwwOHDx/GnXfe2WtfvV4PmUyGwsJCfPnllyguLkZNTQ2cnJzg6+uLxYsXQ6PRmPvn5ORgzZo1PY6Vl5eHe+65B+fPn8fevXtRUFCAy5cvo62tDffddx8ee+wxvPDCC1YP2I+MjDSPvXnzZuTn56Ourg47d+7E1KlTB1zbtQoLC7Fr1y58//33aGpqgkKhQFBQEBISEuDr69vrA89jYmKQkpLS78+Y6GbiESfRIDt48CCMRiNiY2P7DE0AkMlkAIAdO3agqqoKarUaSqUSV69eRX5+Pl577TXodDrEx8cDAPz8/LB48WKkp6djwoQJeOSRR8xjubm5AQCOHTuGnJwcTJs2DaGhodDr9Th58iQ++OADaLVa86usrqXX67FkyRJ0dHQgKioKBoMBI0aMEFRbt7S0NKSkpGDEiBGYMWMGFAoFfvrpJxQVFSEgIMAcul988QXOnDmDxYsXm2v38/Oz6WdOZFcmIhpUixYtMqlUKlNBQcGA9/nf//5nta25udk0e/ZsU1BQkKmlpcW8vbKy0qRSqUyrVq3qcazq6mpTe3u7xTaj0Wh68803TSqVyvTdd99ZtEVERJhUKpXpxRdftNpPaG2nT582TZgwwRQREWGqrq622Kejo8NUW1tr/rxq1SqTSqUyVVZW9jgPIrFwVS3RILty5QoAQKlUDnife++912rbHXfcgZiYGDQ1NeHUqVMDHkupVJqPZLs5OTlhwYIFALpOo/YkMTHRaj+hte3ZswdGoxGJiYlW8x82bBhGjx494HkQiYWnaokcQFNTEz755BPk5+ejsrISra2tFu21tbUDHstoNGL//v3IycnBuXPn0NTUBNM1Sx16GsvFxQXjx4+/4drKysoAAGFhYQOul+hWw+AkGmSjR49GRUUFampq4OPj029/vV6PRYsWoby8HP7+/oiJiYG7uzukUinKy8uRl5cHvV4/4O9PTk5GZmYm7r77bvM1RmdnZzQ2NiI9Pb3HsXq7Fiu0tqamJshkMnh4eAy4XqJbDYOTaJAFBQXh5MmT+Oc//4np06f32z8vLw/l5eWIjY3FunXrLNo+/vhj5OXlDfi7r1y5gqysLEyYMAHZ2dlwcXExt5WWliI9Pb3H/ZycnOxS28iRI3Hx4kU0NDQwPMlh8Ron0SB74oknIJFIsHfvXtTX1/fZV6/Xo7KyEgAQERFh1V5cXGy1TSqVAuh6yML1qqqqYDKZEBoaahGavY3VH6G1BQQEAABOnDjR79gSSdefJ6PRKLguopuJwUk0yH77298iPj4edXV1+MMf/oDq6mqrPo2NjUhJScGRI0dw1113AbAOoq+++gp///vfrfbtvnWjpqbGqq17rJKSEovrmufPn8f27dsFz0VobbGxsZBIJNi8eTN0Op1FW2dnJ+rq6syf3d3de50HkZh4qpZIBImJiWhtbUVWVhZmzJiBsLAw8/XOixcvoqCgAK2trUhNTUVISAjGjBmDTz75BGfPnoWPjw/OnTuHb7/9FjNmzMCxY8csxnZ1dUVAQAC+++47vPnmm7jvvvvMq2aVSqV5n7lz52Lq1KnQ6XTIz8/Hgw8+iL/+9a+C5hERESGoNj8/P6xcuRLvvfceoqOjMXPmTCgUCuh0OhQUFODZZ5813/c5bdo0fPbZZ3jnnXcQFRWF4cOHQ6VSITw83OafO5E9SNeuXbtW7CKIhhqJRILw8HCEhoaivb0dZWVlKCoqwqlTp9DZ2YnIyEgkJycjJCQEMpkMkZGRuHz5MkpKSvD999/D3d0d69atg1KpRF5eHh555BGLhwNMmTIFFy5cwD/+8Q8cP34chYWFiI2NhZubGx566CG0tbXh9OnTKCoqQkdHB55//nksXLgQ6enp8PPzs3hwwq5duwAAS5YssZqHLbUFBgYiMDAQ1dXVKCwsxL/+9S80NDQgODgYMTExGDVqFADgN7/5DWQyGcrKynD8+HGcOHECzs7OFrURiYGP3CMiIhKA1ziJiIgEYHASEREJwOAkIiISgMFJREQkAIOTiIhIAAYnERGRAAxOIiIiARicREREAjA4iYiIBGBwEhERCfB/ZHkp1L2K6JIAAAAASUVORK5CYII=\n","text/plain":["<Figure size 504x504 with 2 Axes>"]},"metadata":{},"output_type":"display_data"}]},{"cell_type":"code","metadata":{"id":"4gFshmDEqduf"},"source":["#### Creating the dataset ####\n","# Normal sample\n","normal = eyes_df[eyes_df.Normal == 1].copy()\n","normal.loc[:, 'Label'] = 0\n","\n","# Cataract sample\n","cataract = eyes_df[eyes_df.Cataract == 1].copy()\n","cataract.loc[:, 'Label'] = 1\n","\n","dataset = pd.concat([normal, cataract])\n","\n","dataset = dataset.drop(columns=class_cols)\n"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"oKQnQODxqdug","outputId":"7fb02845-de2d-4fd6-9544-78287fbeab8e"},"source":["dataset.head()"],"execution_count":null,"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>Image</th>\n","      <th>Diagnosis</th>\n","      <th>Age</th>\n","      <th>Sex</th>\n","      <th>Label</th>\n","    </tr>\n","  </thead>\n","  <tbody>\n","    <tr>\n","      <th>1</th>\n","      <td>1_left.jpg</td>\n","      <td>normal fundus</td>\n","      <td>57</td>\n","      <td>Male</td>\n","      <td>0</td>\n","    </tr>\n","    <tr>\n","      <th>3</th>\n","      <td>3_left.jpg</td>\n","      <td>normal fundus</td>\n","      <td>66</td>\n","      <td>Male</td>\n","      <td>0</td>\n","    </tr>\n","    <tr>\n","      <th>8</th>\n","      <td>8_left.jpg</td>\n","      <td>normal fundus</td>\n","      <td>59</td>\n","      <td>Male</td>\n","      <td>0</td>\n","    </tr>\n","    <tr>\n","      <th>9</th>\n","      <td>9_left.jpg</td>\n","      <td>normal fundus</td>\n","      <td>54</td>\n","      <td>Male</td>\n","      <td>0</td>\n","    </tr>\n","    <tr>\n","      <th>14</th>\n","      <td>14_left.jpg</td>\n","      <td>normal fundus</td>\n","      <td>55</td>\n","      <td>Male</td>\n","      <td>0</td>\n","    </tr>\n","  </tbody>\n","</table>\n","</div>"],"text/plain":["          Image      Diagnosis  Age   Sex  Label\n","1    1_left.jpg  normal fundus   57  Male      0\n","3    3_left.jpg  normal fundus   66  Male      0\n","8    8_left.jpg  normal fundus   59  Male      0\n","9    9_left.jpg  normal fundus   54  Male      0\n","14  14_left.jpg  normal fundus   55  Male      0"]},"execution_count":43,"metadata":{},"output_type":"execute_result"}]},{"cell_type":"code","metadata":{"id":"XDTjHD4_qdug"},"source":["# Training and testing folders\n","#os.mkdir('my_training_images')\n","#os.mkdir('my_testing_images')"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"YgUzZD4nqdug"},"source":["#import shutil\n","#shutil.rmtree('my_training_images/Normal')\n","#shutil.rmtree('my_testing_images/Normal')"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"rGFPyjl7qduh"},"source":["# Creating folder classes:\n","for label, class_col in enumerate(class_cols):\n","    y = dataset.loc[dataset['Label'] == label, 'Label']\n","    x = dataset.loc[dataset['Label'] == label, 'Image']\n","    img_train, img_test, y_train, y_test = train_test_split(x, y, test_size=0.2, stratify=y, random_state=10)\n","\n","    os.mkdir('my_training_images/' + class_col)\n","    os.mkdir('my_testing_images/' + class_col)\n","    for img in img_train:\n","        shutil.copy('../input/ocular-disease-recognition-odir5k/ODIR-5K/ODIR-5K/Training Images/'+img, './my_training_images/'+class_col+'/'+img)\n","    for img in img_test:\n","        shutil.copy('../input/ocular-disease-recognition-odir5k/ODIR-5K/ODIR-5K/Training Images/'+img, './my_testing_images/'+class_col+'/'+img)\n","        "],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"S28b-rQvqduh","outputId":"a9780807-8fab-48d0-e896-5240e627a4ea"},"source":["print(len(normal), len(cataract))"],"execution_count":null,"outputs":[{"name":"stdout","output_type":"stream","text":["3104 313\n"]}]},{"cell_type":"code","metadata":{"id":"52YCyC0rqduh"},"source":["train_datagen = ImageDataGenerator(rescale=1./255)\n","test_datagen = ImageDataGenerator(rescale=1./255)"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"7eDiUYFdqdui"},"source":["# Dimensions of our images\n","img_width, img_height = 250, 250"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"Nffj37DUqdui"},"source":["batch_size = 32"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"nlFeqMtcqdui","outputId":"a93b771f-35b5-4a55-e514-94568614a2a1"},"source":["train_generator = train_datagen.flow_from_directory(\n","        'my_training_images/', \n","        target_size=(img_width, img_height),\n","        batch_size=batch_size,\n","        class_mode='binary') \n","test_generator = test_datagen.flow_from_directory(\n","        'my_testing_images/',\n","        target_size=(img_width, img_height),\n","        batch_size=batch_size,\n","        class_mode='binary',\n","        shuffle=False)"],"execution_count":null,"outputs":[{"name":"stdout","output_type":"stream","text":["Found 2733 images belonging to 2 classes.\n","Found 684 images belonging to 2 classes.\n"]}]},{"cell_type":"code","metadata":{"id":"fI0g7SlAqdui"},"source":["if K.image_data_format() == 'channels_first':\n","    input_shape = (3, img_width, img_height)\n","else:\n","    input_shape = (img_width, img_height, 3)"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"tTqiJdG2qduj"},"source":["nb_train_samples = 2000\n","nb_validation_samples = 650\n","\n","epochs = 15"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"-_BGFWyjqduj"},"source":["model = Sequential()\n","model.add(Conv2D(32, (3, 3), input_shape=input_shape))\n","model.add(Activation('relu'))\n","model.add(MaxPooling2D(pool_size=(2, 2)))\n","\n","model.add(Conv2D(32, (3, 3)))\n","model.add(Activation('relu'))\n","model.add(MaxPooling2D(pool_size=(2, 2)))\n","\n","model.add(Conv2D(64, (3, 3)))\n","model.add(Activation('relu'))\n","model.add(MaxPooling2D(pool_size=(2, 2)))\n","\n","model.add(Flatten())\n","model.add(Dense(64))\n","model.add(Activation('relu'))\n","model.add(Dropout(0.5))\n","model.add(Dense(1))\n","model.add(Activation('sigmoid'))\n","\n","model.compile(loss='binary_crossentropy',\n","              optimizer='rmsprop',\n","              metrics=['accuracy'])\n"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"yBknR1Toqduj","outputId":"27d2b31d-5052-494f-80ce-bd6286704273"},"source":["model.summary()\n"],"execution_count":null,"outputs":[{"name":"stdout","output_type":"stream","text":["Model: \"sequential_1\"\n","_________________________________________________________________\n","Layer (type)                 Output Shape              Param #   \n","=================================================================\n","conv2d_3 (Conv2D)            (None, 248, 248, 32)      896       \n","_________________________________________________________________\n","activation_5 (Activation)    (None, 248, 248, 32)      0         \n","_________________________________________________________________\n","max_pooling2d_3 (MaxPooling2 (None, 124, 124, 32)      0         \n","_________________________________________________________________\n","conv2d_4 (Conv2D)            (None, 122, 122, 32)      9248      \n","_________________________________________________________________\n","activation_6 (Activation)    (None, 122, 122, 32)      0         \n","_________________________________________________________________\n","max_pooling2d_4 (MaxPooling2 (None, 61, 61, 32)        0         \n","_________________________________________________________________\n","conv2d_5 (Conv2D)            (None, 59, 59, 64)        18496     \n","_________________________________________________________________\n","activation_7 (Activation)    (None, 59, 59, 64)        0         \n","_________________________________________________________________\n","max_pooling2d_5 (MaxPooling2 (None, 29, 29, 64)        0         \n","_________________________________________________________________\n","flatten_1 (Flatten)          (None, 53824)             0         \n","_________________________________________________________________\n","dense_2 (Dense)              (None, 64)                3444800   \n","_________________________________________________________________\n","activation_8 (Activation)    (None, 64)                0         \n","_________________________________________________________________\n","dropout_1 (Dropout)          (None, 64)                0         \n","_________________________________________________________________\n","dense_3 (Dense)              (None, 1)                 65        \n","_________________________________________________________________\n","activation_9 (Activation)    (None, 1)                 0         \n","=================================================================\n","Total params: 3,473,505\n","Trainable params: 3,473,505\n","Non-trainable params: 0\n","_________________________________________________________________\n"]}]},{"cell_type":"code","metadata":{"scrolled":false,"id":"QNk1fdmGqduj","outputId":"5fb248c1-0af3-4a86-a7c3-31fb5c2f40cf"},"source":["model.fit(\n","    train_generator,\n","    steps_per_epoch=nb_train_samples // batch_size,\n","    epochs=epochs,\n","    validation_data=test_generator,\n","    validation_steps=nb_validation_samples // batch_size)\n","\n","model.save_weights('cataract_first_try.h5')\n"],"execution_count":null,"outputs":[{"name":"stdout","output_type":"stream","text":["Epoch 1/15\n","62/62 [==============================] - 124s 2s/step - loss: 0.3562 - accuracy: 0.8987 - val_loss: 0.3479 - val_accuracy: 0.9016\n","Epoch 2/15\n","62/62 [==============================] - 124s 2s/step - loss: 0.3219 - accuracy: 0.9074 - val_loss: 0.2946 - val_accuracy: 0.9016\n","Epoch 3/15\n","62/62 [==============================] - 125s 2s/step - loss: 0.2785 - accuracy: 0.9135 - val_loss: 0.4713 - val_accuracy: 0.9016\n","Epoch 4/15\n","62/62 [==============================] - 122s 2s/step - loss: 0.2730 - accuracy: 0.9115 - val_loss: 0.3125 - val_accuracy: 0.9031\n","Epoch 5/15\n","62/62 [==============================] - 123s 2s/step - loss: 0.2717 - accuracy: 0.9125 - val_loss: 0.2645 - val_accuracy: 0.9047\n","Epoch 6/15\n","62/62 [==============================] - 123s 2s/step - loss: 0.2416 - accuracy: 0.9189 - val_loss: 0.4105 - val_accuracy: 0.9031\n","Epoch 7/15\n","62/62 [==============================] - 124s 2s/step - loss: 0.2349 - accuracy: 0.9158 - val_loss: 0.2875 - val_accuracy: 0.9078\n","Epoch 8/15\n","62/62 [==============================] - 123s 2s/step - loss: 0.2454 - accuracy: 0.9211 - val_loss: 0.2531 - val_accuracy: 0.9141\n","Epoch 9/15\n","62/62 [==============================] - 124s 2s/step - loss: 0.2141 - accuracy: 0.9206 - val_loss: 0.3583 - val_accuracy: 0.9125\n","Epoch 10/15\n","62/62 [==============================] - 125s 2s/step - loss: 0.2227 - accuracy: 0.9204 - val_loss: 0.2792 - val_accuracy: 0.9172\n","Epoch 11/15\n","62/62 [==============================] - 123s 2s/step - loss: 0.1938 - accuracy: 0.9267 - val_loss: 0.2298 - val_accuracy: 0.9250\n","Epoch 12/15\n","62/62 [==============================] - 123s 2s/step - loss: 0.2052 - accuracy: 0.9298 - val_loss: 0.2693 - val_accuracy: 0.9203\n","Epoch 13/15\n","62/62 [==============================] - 123s 2s/step - loss: 0.1816 - accuracy: 0.9333 - val_loss: 0.2238 - val_accuracy: 0.9172\n","Epoch 14/15\n","62/62 [==============================] - 124s 2s/step - loss: 0.1964 - accuracy: 0.9262 - val_loss: 0.2334 - val_accuracy: 0.9187\n","Epoch 15/15\n","62/62 [==============================] - 124s 2s/step - loss: 0.1646 - accuracy: 0.9420 - val_loss: 0.2223 - val_accuracy: 0.9344\n"]}]},{"cell_type":"code","metadata":{"id":"59Vozzucqduk"},"source":["y_pred = [1 if i > 0.5 else 0 for i in model.predict(test_generator)]\n","y_test = test_generator.classes\n","print(classification_report(y_test, y_pred))"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"1vJh-qMeqduk"},"source":["cm = confusion_matrix(y_test, y_pred)\n","df_cm = pd.DataFrame(cm, index = ['Cataract', 'Normal'],\n","                  columns = ['Cataract', 'Normal'])\n","plt.figure(figsize = (10,7))\n","sns.heatmap(df_cm, annot=True)"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"0zou6IVAqduk"},"source":["## Balancing the dataset ##\n","# Training and testing folders\n","os.mkdir('my_training_balanced')\n","os.mkdir('my_testing_balanced')"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"lXXjr9-kqduk","outputId":"7257fa07-1caa-4ade-f067-ae6c2689be90"},"source":["size = len(cataract)\n","print(size)"],"execution_count":null,"outputs":[{"name":"stdout","output_type":"stream","text":["313\n"]}]},{"cell_type":"code","metadata":{"id":"xyCwdOpUqduk"},"source":["# Creating folder classes:\n","for label, class_col in enumerate(class_cols):\n","    y = dataset.loc[dataset['Label'] == label, 'Label'].sample(n=size, random_state=1)\n","    x = dataset.loc[dataset['Label'] == label, 'Image'].sample(n=size, random_state=1)\n","    img_train, img_test, y_train, y_test = train_test_split(x, y, test_size=0.2, stratify=y, random_state=10)\n","\n","    os.mkdir('my_training_balanced/' + class_col)\n","    os.mkdir('my_testing_balanced/' + class_col)\n","    for img in img_train:\n","        shutil.copy('../input/ocular-disease-recognition-odir5k/ODIR-5K/ODIR-5K/Training Images/'+img, './my_training_balanced/'+class_col+'/'+img)\n","    for img in img_test:\n","        shutil.copy('../input/ocular-disease-recognition-odir5k/ODIR-5K/ODIR-5K/Training Images/'+img, './my_testing_balanced/'+class_col+'/'+img)\n","        "],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"5v1gVJqbqdul","outputId":"491e6ebf-6cb9-42f7-893b-d136c8462b0d"},"source":["train_generator = train_datagen.flow_from_directory(\n","        'my_training_balanced/', \n","        target_size=(img_width, img_height),\n","        batch_size=batch_size,\n","        class_mode='binary') \n","test_generator = test_datagen.flow_from_directory(\n","        'my_testing_balanced/',\n","        target_size=(img_width, img_height),\n","        batch_size=batch_size,\n","        class_mode='binary',\n","        shuffle=False)"],"execution_count":null,"outputs":[{"name":"stdout","output_type":"stream","text":["Found 500 images belonging to 2 classes.\n","Found 126 images belonging to 2 classes.\n"]}]},{"cell_type":"code","metadata":{"id":"D4p5lxptqdul"},"source":["nb_train_samples = 2000\n","nb_validation_samples = 120\n","\n","epochs = 15"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"_7P3sjROqdul"},"source":["model_balanced = Sequential()\n","model_balanced.add(Conv2D(32, (3, 3), input_shape=input_shape))\n","model_balanced.add(Activation('relu'))\n","model_balanced.add(MaxPooling2D(pool_size=(2, 2)))\n","\n","model_balanced.add(Conv2D(32, (3, 3)))\n","model_balanced.add(Activation('relu'))\n","model_balanced.add(MaxPooling2D(pool_size=(2, 2)))\n","\n","model_balanced.add(Conv2D(64, (3, 3)))\n","model_balanced.add(Activation('relu'))\n","model_balanced.add(MaxPooling2D(pool_size=(2, 2)))\n","\n","model_balanced.add(Flatten())\n","model_balanced.add(Dense(64))\n","model_balanced.add(Activation('relu'))\n","model_balanced.add(Dropout(0.5))\n","model_balanced.add(Dense(1))\n","model_balanced.add(Activation('softmax'))\n","\n","model_balanced.compile(loss='categorical_crossentropy',\n","              optimizer='adam',\n","              metrics=['accuracy'])\n"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"wX7QgHRjqdul","outputId":"eb15676a-ea6a-4e3b-e5de-0767b0f941bb"},"source":["history_balanced = model.fit(\n","    train_generator,\n","    steps_per_epoch=nb_train_samples // batch_size,\n","    epochs=epochs,\n","    validation_data=test_generator,\n","    validation_steps=nb_validation_samples // batch_size)\n","\n","model_balanced.save_weights('cataract_balanced_try.h5')"],"execution_count":null,"outputs":[{"name":"stdout","output_type":"stream","text":["Epoch 1/15\n","62/62 [==============================] - 99s 2s/step - loss: 0.0975 - accuracy: 0.9684 - val_loss: 3.1885 - val_accuracy: 0.4688\n","Epoch 2/15\n","62/62 [==============================] - 98s 2s/step - loss: 0.0707 - accuracy: 0.9768 - val_loss: 1.2322 - val_accuracy: 0.8438\n","Epoch 3/15\n","62/62 [==============================] - 97s 2s/step - loss: 0.0777 - accuracy: 0.9730 - val_loss: 1.4074 - val_accuracy: 0.8438\n","Epoch 4/15\n","62/62 [==============================] - 99s 2s/step - loss: 0.0828 - accuracy: 0.9723 - val_loss: 1.4209 - val_accuracy: 0.8021\n","Epoch 5/15\n","62/62 [==============================] - 102s 2s/step - loss: 0.1146 - accuracy: 0.9746 - val_loss: 3.0164 - val_accuracy: 0.7708\n","Epoch 6/15\n","62/62 [==============================] - 98s 2s/step - loss: 0.0705 - accuracy: 0.9746 - val_loss: 1.2735 - val_accuracy: 0.7917\n","Epoch 7/15\n","62/62 [==============================] - 98s 2s/step - loss: 0.0814 - accuracy: 0.9730 - val_loss: 1.3536 - val_accuracy: 0.8229\n","Epoch 8/15\n","62/62 [==============================] - 98s 2s/step - loss: 0.0703 - accuracy: 0.9761 - val_loss: 2.3458 - val_accuracy: 0.8646\n","Epoch 9/15\n","62/62 [==============================] - 98s 2s/step - loss: 0.0774 - accuracy: 0.9746 - val_loss: 1.7810 - val_accuracy: 0.7188\n","Epoch 10/15\n","62/62 [==============================] - 98s 2s/step - loss: 0.0918 - accuracy: 0.9700 - val_loss: 1.7967 - val_accuracy: 0.7917\n","Epoch 11/15\n","62/62 [==============================] - 98s 2s/step - loss: 0.0760 - accuracy: 0.9751 - val_loss: 2.6199 - val_accuracy: 0.8021\n","Epoch 12/15\n","62/62 [==============================] - 99s 2s/step - loss: 0.0725 - accuracy: 0.9778 - val_loss: 0.9843 - val_accuracy: 0.8229\n","Epoch 13/15\n","62/62 [==============================] - 98s 2s/step - loss: 0.0658 - accuracy: 0.9751 - val_loss: 2.2287 - val_accuracy: 0.8646\n","Epoch 14/15\n","62/62 [==============================] - 98s 2s/step - loss: 0.1183 - accuracy: 0.9715 - val_loss: 1.8803 - val_accuracy: 0.8021\n","Epoch 15/15\n","62/62 [==============================] - 99s 2s/step - loss: 0.0619 - accuracy: 0.9748 - val_loss: 1.4319 - val_accuracy: 0.8542\n"]}]},{"cell_type":"code","metadata":{"id":"04Xj3pxpqdum","outputId":"9674ff41-c8a1-4569-f233-f438c407dbaa"},"source":["y_pred = [1 if i > 0.5 else 0 for i in model_balanced.predict(test_generator)]\n","y_test = test_generator.classes\n","print(classification_report(y_test, y_pred))"],"execution_count":null,"outputs":[{"name":"stdout","output_type":"stream","text":["              precision    recall  f1-score   support\n","\n","           0       0.00      0.00      0.00        63\n","           1       0.50      1.00      0.67        63\n","\n","    accuracy                           0.50       126\n","   macro avg       0.25      0.50      0.33       126\n","weighted avg       0.25      0.50      0.33       126\n","\n"]},{"name":"stderr","output_type":"stream","text":["/opt/conda/lib/python3.7/site-packages/sklearn/metrics/_classification.py:1221: UndefinedMetricWarning: Precision and F-score are ill-defined and being set to 0.0 in labels with no predicted samples. Use `zero_division` parameter to control this behavior.\n","  _warn_prf(average, modifier, msg_start, len(result))\n"]}]},{"cell_type":"code","metadata":{"id":"ef3yFw1Kqdum"},"source":["cm = confusion_matrix(y_test, y_pred)\n","df_cm = pd.DataFrame(cm, index = ['Normal', 'Cataract'],\n","                  columns = ['Cataract', 'Normal'])\n","plt.figure(figsize = (10,7))\n","sns.heatmap(df_cm, annot=True)"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"IP0U51Gwqdum","executionInfo":{"status":"ok","timestamp":1633776530587,"user_tz":-120,"elapsed":207,"user":{"displayName":"Pietro Obbiso","photoUrl":"https://lh3.googleusercontent.com/a/default-user=s64","userId":"12598544054389310202"}}},"source":["#Using VGG19 model\n","vgg19 = VGG19(weights=\"imagenet\", include_top=False, input_shape=input_shape)"],"execution_count":2,"outputs":[]},{"cell_type":"code","metadata":{"id":"CwyP1uy9qdum"},"source":["for layer in vgg19.layers:\n","    layer.trainable = False\n","    "],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"h_mqQ6Aoqdum","executionInfo":{"status":"ok","timestamp":1633776546517,"user_tz":-120,"elapsed":199,"user":{"displayName":"Pietro Obbiso","photoUrl":"https://lh3.googleusercontent.com/a/default-user=s64","userId":"12598544054389310202"}}},"source":["model_vgg19 = Sequential()\n","model_vgg19.add(vgg19)\n","model_vgg19.add(Flatten())\n","model_vgg19.add(Dense(1, activation=\"sigmoid\"))\n"],"execution_count":3,"outputs":[]},{"cell_type":"code","metadata":{"id":"Lj1Wx-VXqdun"},"source":["model_vgg19.summary()\n"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"_ni4YMu5qdun"},"source":["model_vgg19.compile(optimizer=\"adam\", loss=\"binary_crossentropy\", metrics=[\"accuracy\"])\n"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"Ddt4ZkFrqdun"},"source":["epochs = 15"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"Emn39RHsqdun"},"source":["history_vgg19 = model_vgg19.fit(train_generator, \n","                    steps_per_epoch=nb_train_samples // batch_size, \n","                    epochs=epochs, \n","                    validation_data=test_generator,\n","                    validation_steps=nb_validation_samples // batch_size,\n","                    verbose=1)\n","model_vgg19.save_weights('cataract_balanced_vgg19_try.h5')\n"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"xQIo2BDwqdun","outputId":"2d9418f7-6eab-4a1e-a188-1a4e045b9f81"},"source":["y_pred = [1 if i > 0.5 else 0 for i in model_vgg19.predict(test_generator)]\n","y_test = test_generator.classes\n","print(classification_report(y_test, y_pred))\n"],"execution_count":null,"outputs":[{"name":"stdout","output_type":"stream","text":["              precision    recall  f1-score   support\n","\n","           0       0.00      0.00      0.00        63\n","           1       0.50      1.00      0.67        63\n","\n","    accuracy                           0.50       126\n","   macro avg       0.25      0.50      0.33       126\n","weighted avg       0.25      0.50      0.33       126\n","\n"]}]},{"cell_type":"code","metadata":{"id":"Y5A-i0Idqduo"},"source":["sns.set(font_scale=1.6)\n","cm = confusion_matrix(y_test, y_pred)\n","df_cm = pd.DataFrame(cm, index = ['Normal', 'Cataract'],\n","                  columns = ['Normal', 'Cataract'])\n","plt.figure(figsize = (10,7))\n","sns.heatmap(df_cm, annot=True)"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"w2KXuAduqduo"},"source":["#Using Diabetic data\n","eyes_df['Diabetic'] = eyes_df['Diagnosis'].apply(lambda x: has_disease('proliferative retinopathy', x))  \n"," \n","diabetic = eyes_df[eyes_df.Diabetic == 1].copy()\n","diabetic.loc[:, 'Label'] = 1\n"," \n","dataset2 = pd.concat([normal, diabetic])\n","\n","\n","dataset2 = dataset2.drop(columns=class_cols)"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"deNWcbsuqduo"},"source":["os.mkdir('my_training_diabetic')\n","os.mkdir('my_testing_diabetic')"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"H8hBB8pVqdup"},"source":["class_cols = ['Normal','Diabetic']\n","size = len(diabetic)\n","for label, class_col in enumerate(class_cols):\n","    y = dataset2.loc[dataset2['Label'] == label, 'Label'].sample(n=size, random_state=1)\n","    x = dataset2.loc[dataset2['Label'] == label, 'Image'].sample(n=size, random_state=1)\n","    img_train, img_test, y_train, y_test = train_test_split(x, y, test_size=0.2, stratify=y, random_state=10)\n","\n"," \n","\n","    #os.mkdir('my_training_diabetic/' + class_col)\n","    #os.mkdir('my_testing_diabetic/' + class_col)\n","    for img in img_train:\n","        shutil.copy('../input/ocular-disease-recognition-odir5k/ODIR-5K/ODIR-5K/Training Images/'+img, './my_training_diabetic/'+class_col+'/'+img)\n","    for img in img_test:\n","        shutil.copy('../input/ocular-disease-recognition-odir5k/ODIR-5K/ODIR-5K/Training Images/'+img, './my_testing_diabetic/'+class_col+'/'+img)"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"bgvIThaaqdup","outputId":"46f25359-ebdd-4ec5-d0e9-cd163a4a9e16"},"source":["train_generator = train_datagen.flow_from_directory(\n","        'my_training_diabetic/',\n","        target_size=(img_width, img_height),\n","        batch_size=batch_size,\n","        class_mode='binary')\n","test_generator = test_datagen.flow_from_directory(\n","        'my_testing_diabetic/',\n","        target_size=(img_width, img_height),\n","        batch_size=batch_size,\n","        class_mode='binary',\n","        shuffle=False)"],"execution_count":null,"outputs":[{"name":"stdout","output_type":"stream","text":["Found 2728 images belonging to 2 classes.\n","Found 684 images belonging to 2 classes.\n"]}]},{"cell_type":"code","metadata":{"id":"guo-bNeVqdup"},"source":["model_balanced_diabetic = Sequential()\n","model_balanced_diabetic.add(Conv2D(32, (3, 3), input_shape=input_shape))\n","model_balanced_diabetic.add(Activation('relu'))\n","model_balanced_diabetic.add(MaxPooling2D(pool_size=(2, 2)))\n","\n","model_balanced_diabetic.add(Conv2D(32, (3, 3)))\n","model_balanced_diabetic.add(Activation('relu'))\n","model_balanced_diabetic.add(MaxPooling2D(pool_size=(2, 2)))\n","\n","model_balanced_diabetic.add(Conv2D(64, (3, 3)))\n","model_balanced_diabetic.add(Activation('relu'))\n","model_balanced_diabetic.add(MaxPooling2D(pool_size=(2, 2)))\n","\n","model_balanced_diabetic.add(Flatten())\n","model_balanced_diabetic.add(Dense(64))\n","model_balanced_diabetic.add(Activation('relu'))\n","model_balanced_diabetic.add(Dropout(0.5))\n","model_balanced_diabetic.add(Dense(1))\n","model_balanced_diabetic.add(Activation('sigmoid'))\n","\n","model_balanced_diabetic.compile(loss='binary_crossentropy',\n","              optimizer='adam',\n","              metrics=['accuracy'])"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"OA6UHRQSqdur","outputId":"0b70755e-9385-407f-aadf-c12565c70edc"},"source":["history_balanced_diabetic = model_balanced_diabetic.fit(\n","    train_generator,\n","    steps_per_epoch=nb_train_samples // batch_size,\n","    epochs=epochs,\n","    validation_data=test_generator,\n","    validation_steps=nb_validation_samples // batch_size)\n","\n","model_balanced.save_weights('diabetic_balanced_try.h5')"],"execution_count":null,"outputs":[{"name":"stdout","output_type":"stream","text":["Epoch 1/15\n","62/62 [==============================] - 100s 2s/step - loss: 0.6932 - accuracy: 0.5628 - val_loss: 0.6689 - val_accuracy: 0.6458\n","Epoch 2/15\n","62/62 [==============================] - 98s 2s/step - loss: 0.6798 - accuracy: 0.5730 - val_loss: 0.7072 - val_accuracy: 0.3229\n","Epoch 3/15\n","62/62 [==============================] - 98s 2s/step - loss: 0.6722 - accuracy: 0.5974 - val_loss: 0.6084 - val_accuracy: 0.6979\n","Epoch 4/15\n","62/62 [==============================] - 99s 2s/step - loss: 0.6728 - accuracy: 0.5878 - val_loss: 0.6144 - val_accuracy: 0.6979\n","Epoch 5/15\n","62/62 [==============================] - 98s 2s/step - loss: 0.6666 - accuracy: 0.5724 - val_loss: 0.6746 - val_accuracy: 0.6562\n","Epoch 6/15\n","62/62 [==============================] - 98s 2s/step - loss: 0.6660 - accuracy: 0.5939 - val_loss: 0.6375 - val_accuracy: 0.6771\n","Epoch 7/15\n","62/62 [==============================] - 100s 2s/step - loss: 0.6692 - accuracy: 0.5892 - val_loss: 0.6727 - val_accuracy: 0.5833\n","Epoch 8/15\n","62/62 [==============================] - 99s 2s/step - loss: 0.6611 - accuracy: 0.6099 - val_loss: 0.8340 - val_accuracy: 0.5833\n","Epoch 9/15\n","62/62 [==============================] - 99s 2s/step - loss: 0.6652 - accuracy: 0.6097 - val_loss: 0.6668 - val_accuracy: 0.6354\n","Epoch 10/15\n","62/62 [==============================] - 98s 2s/step - loss: 0.6610 - accuracy: 0.6000 - val_loss: 0.6328 - val_accuracy: 0.6667\n","Epoch 11/15\n","62/62 [==============================] - 99s 2s/step - loss: 0.6559 - accuracy: 0.6010 - val_loss: 0.7389 - val_accuracy: 0.5833\n","Epoch 12/15\n","62/62 [==============================] - 98s 2s/step - loss: 0.6637 - accuracy: 0.6066 - val_loss: 0.7265 - val_accuracy: 0.6250\n","Epoch 13/15\n","62/62 [==============================] - 99s 2s/step - loss: 0.6589 - accuracy: 0.6092 - val_loss: 0.6505 - val_accuracy: 0.6875\n","Epoch 14/15\n","62/62 [==============================] - 98s 2s/step - loss: 0.6553 - accuracy: 0.6026 - val_loss: 0.6943 - val_accuracy: 0.6562\n","Epoch 15/15\n","62/62 [==============================] - 100s 2s/step - loss: 0.6418 - accuracy: 0.6250 - val_loss: 0.6391 - val_accuracy: 0.6458\n"]}]}]}