computer_vision/01-the-convolutional-classifier.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "dec0e7e0",
   "metadata": {
    "papermill": {
     "duration": 0.020319,
     "end_time": "2022-03-13T16:22:46.240830",
     "exception": false,
     "start_time": "2022-03-13T16:22:46.220511",
     "status": "completed"
    },
    "tags": []
   },
   "source": [
    "**This notebook is an exercise in the [Computer Vision](https://www.kaggle.com/learn/computer-vision) course.  You can reference the tutorial at [this link](https://www.kaggle.com/ryanholbrook/the-convolutional-classifier).**\n",
    "\n",
    "---\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "36d357be",
   "metadata": {
    "papermill": {
     "duration": 0.016769,
     "end_time": "2022-03-13T16:22:46.277681",
     "exception": false,
     "start_time": "2022-03-13T16:22:46.260912",
     "status": "completed"
    },
    "tags": []
   },
   "source": [
    "<blockquote style=\"margin-right:auto; margin-left:auto; background-color: #ebf9ff; padding: 1em; margin:24px;\">\n",
    "    <strong>Accelerate Training with a Kaggle GPU!</strong><br>\n",
    "Did you know Kaggle offers free time with a GPU accelerator? You can speed up training neural networks in this course by switching to <strong>GPU</strong> in the <em>Accelerator</em> option on the right. (It may already be turned on.) Two things to be aware of:\n",
    "<ul>\n",
    "<li>Changing the <em>Accelerator</em> option will cause the notebook session to restart. You'll need to rerun any setup code.\n",
    "<li>You can have only one GPU session at a time, so be sure to shut the notebook down after you've finished the exercise.\n",
    "</ul>\n",
    "</blockquote>"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b01ff44e",
   "metadata": {
    "papermill": {
     "duration": 0.017504,
     "end_time": "2022-03-13T16:22:46.311946",
     "exception": false,
     "start_time": "2022-03-13T16:22:46.294442",
     "status": "completed"
    },
    "tags": []
   },
   "source": [
    "# Introduction #\n",
    "\n",
    "In the tutorial, we saw how to build an image classifier by attaching a head of dense layers to a pretrained base. The base we used was from a model called **VGG16**. We saw that the VGG16 architecture was prone to overfitting this dataset. Over this course, you'll learn a number of ways you can improve upon this initial attempt.\n",
    "\n",
    "The first way you'll see is to use a base more appropriate to the dataset. The base this model comes from is called **InceptionV1** (also known as GoogLeNet). InceptionV1 was one of the early winners of the ImageNet competition. One of its successors, InceptionV4, is among the state of the art today.\n",
    "\n",
    "To get started, run the code cell below to set everything up."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "990fd096",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-03-13T16:22:46.364292Z",
     "iopub.status.busy": "2022-03-13T16:22:46.363263Z",
     "iopub.status.idle": "2022-03-13T16:22:57.433081Z",
     "shell.execute_reply": "2022-03-13T16:22:57.432475Z",
     "shell.execute_reply.started": "2022-03-13T16:11:45.840478Z"
    },
    "lines_to_next_cell": 2,
    "papermill": {
     "duration": 11.103575,
     "end_time": "2022-03-13T16:22:57.433232",
     "exception": false,
     "start_time": "2022-03-13T16:22:46.329657",
     "status": "completed"
    },
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Found 5117 files belonging to 2 classes.\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2022-03-13 16:22:53.411799: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
      "2022-03-13 16:22:53.506148: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
      "2022-03-13 16:22:53.506951: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
      "2022-03-13 16:22:53.509301: I tensorflow/core/platform/cpu_feature_guard.cc:142] This TensorFlow binary is optimized with oneAPI Deep Neural Network Library (oneDNN) to use the following CPU instructions in performance-critical operations:  AVX2 AVX512F FMA\n",
      "To enable them in other operations, rebuild TensorFlow with the appropriate compiler flags.\n",
      "2022-03-13 16:22:53.510784: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
      "2022-03-13 16:22:53.511726: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
      "2022-03-13 16:22:53.512567: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
      "2022-03-13 16:22:55.368351: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
      "2022-03-13 16:22:55.369119: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
      "2022-03-13 16:22:55.369760: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
      "2022-03-13 16:22:55.370332: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1510] Created device /job:localhost/replica:0/task:0/device:GPU:0 with 15403 MB memory:  -> device: 0, name: Tesla P100-PCIE-16GB, pci bus id: 0000:00:04.0, compute capability: 6.0\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Found 5051 files belonging to 2 classes.\n"
     ]
    }
   ],
   "source": [
    "# Setup feedback system\n",
    "from learntools.core import binder\n",
    "binder.bind(globals())\n",
    "from learntools.computer_vision.ex1 import *\n",
    "\n",
    "# Imports\n",
    "import os, warnings\n",
    "import matplotlib.pyplot as plt\n",
    "from matplotlib import gridspec\n",
    "\n",
    "import numpy as np\n",
    "import tensorflow as tf\n",
    "from tensorflow.keras.preprocessing import image_dataset_from_directory\n",
    "\n",
    "# Reproducability\n",
    "def set_seed(seed=31415):\n",
    "    np.random.seed(seed)\n",
    "    tf.random.set_seed(seed)\n",
    "    os.environ['PYTHONHASHSEED'] = str(seed)\n",
    "    os.environ['TF_DETERMINISTIC_OPS'] = '1'\n",
    "set_seed()\n",
    "\n",
    "# Set Matplotlib defaults\n",
    "plt.rc('figure', autolayout=True)\n",
    "plt.rc('axes', labelweight='bold', labelsize='large',\n",
    "       titleweight='bold', titlesize=18, titlepad=10)\n",
    "plt.rc('image', cmap='magma')\n",
    "warnings.filterwarnings(\"ignore\") # to clean up output cells\n",
    "\n",
    "\n",
    "# Load training and validation sets\n",
    "ds_train_ = image_dataset_from_directory(\n",
    "    '../input/car-or-truck/train',\n",
    "    labels='inferred',\n",
    "    label_mode='binary',\n",
    "    image_size=[128, 128],\n",
    "    interpolation='nearest',\n",
    "    batch_size=64,\n",
    "    shuffle=True,\n",
    ")\n",
    "ds_valid_ = image_dataset_from_directory(\n",
    "    '../input/car-or-truck/valid',\n",
    "    labels='inferred',\n",
    "    label_mode='binary',\n",
    "    image_size=[128, 128],\n",
    "    interpolation='nearest',\n",
    "    batch_size=64,\n",
    "    shuffle=False,\n",
    ")\n",
    "\n",
    "# Data Pipeline\n",
    "def convert_to_float(image, label):\n",
    "    image = tf.image.convert_image_dtype(image, dtype=tf.float32)\n",
    "    return image, label\n",
    "\n",
    "AUTOTUNE = tf.data.experimental.AUTOTUNE\n",
    "ds_train = (\n",
    "    ds_train_\n",
    "    .map(convert_to_float)\n",
    "    .cache()\n",
    "    .prefetch(buffer_size=AUTOTUNE)\n",
    ")\n",
    "ds_valid = (\n",
    "    ds_valid_\n",
    "    .map(convert_to_float)\n",
    "    .cache()\n",
    "    .prefetch(buffer_size=AUTOTUNE)\n",
    ")\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "2c1cc383",
   "metadata": {
    "papermill": {
     "duration": 0.019068,
     "end_time": "2022-03-13T16:22:57.472457",
     "exception": false,
     "start_time": "2022-03-13T16:22:57.453389",
     "status": "completed"
    },
    "tags": []
   },
   "source": [
    "The **InceptionV1** model pretrained on ImageNet is available in the [TensorFlow Hub](https://www.tensorflow.org/hub/) repository, but we'll load it from a local copy. Run this cell to load InceptionV1 for your base."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "e643e497",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-03-13T16:22:57.515631Z",
     "iopub.status.busy": "2022-03-13T16:22:57.514959Z",
     "iopub.status.idle": "2022-03-13T16:23:01.515116Z",
     "shell.execute_reply": "2022-03-13T16:23:01.516118Z",
     "shell.execute_reply.started": "2022-03-13T16:12:02.622876Z"
    },
    "lines_to_next_cell": 0,
    "papermill": {
     "duration": 4.024801,
     "end_time": "2022-03-13T16:23:01.516375",
     "exception": false,
     "start_time": "2022-03-13T16:22:57.491574",
     "status": "completed"
    },
    "tags": []
   },
   "outputs": [],
   "source": [
    "import tensorflow_hub as hub\n",
    "\n",
    "pretrained_base = tf.keras.models.load_model(\n",
    "    '../input/cv-course-models/cv-course-models/inceptionv1'\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "f830b8cb",
   "metadata": {
    "papermill": {
     "duration": 0.017743,
     "end_time": "2022-03-13T16:23:01.556904",
     "exception": false,
     "start_time": "2022-03-13T16:23:01.539161",
     "status": "completed"
    },
    "tags": []
   },
   "source": [
    "# 1) Define Pretrained Base #\n",
    "\n",
    "Now that you have a pretrained base to do our feature extraction, decide whether this base should be trainable or not."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "3e73218a",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-03-13T16:23:01.601707Z",
     "iopub.status.busy": "2022-03-13T16:23:01.599595Z",
     "iopub.status.idle": "2022-03-13T16:23:01.605470Z",
     "shell.execute_reply": "2022-03-13T16:23:01.605894Z",
     "shell.execute_reply.started": "2022-03-13T16:12:08.995892Z"
    },
    "lines_to_next_cell": 0,
    "papermill": {
     "duration": 0.031202,
     "end_time": "2022-03-13T16:23:01.606035",
     "exception": false,
     "start_time": "2022-03-13T16:23:01.574833",
     "status": "completed"
    },
    "tags": []
   },
   "outputs": [
    {
     "data": {
      "application/javascript": [
       "parent.postMessage({\"jupyterEvent\": \"custom.exercise_interaction\", \"data\": {\"outcomeType\": 1, \"valueTowardsCompletion\": 0.25, \"interactionType\": 1, \"questionType\": 2, \"questionId\": \"1_Q1\", \"learnToolsVersion\": \"0.3.4\", \"failureMessage\": \"\", \"exceptionClass\": \"\", \"trace\": \"\"}}, \"*\")"
      ],
      "text/plain": [
       "<IPython.core.display.Javascript object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/markdown": [
       "<span style=\"color:#33cc33\">Correct:</span> When doing transfer learning, it's generally not a good idea to retrain the entire base -- at least not without some care. The reason is that the random weights in the head will initially create large gradient updates, which propogate back into the base layers and destroy much of the pretraining. Using techniques known as **fine tuning** it's possible to further train the base on new data, but this requires some care to do well."
      ],
      "text/plain": [
       "Correct: When doing transfer learning, it's generally not a good idea to retrain the entire base -- at least not without some care. The reason is that the random weights in the head will initially create large gradient updates, which propogate back into the base layers and destroy much of the pretraining. Using techniques known as **fine tuning** it's possible to further train the base on new data, but this requires some care to do well."
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# YOUR_CODE_HERE\n",
    "pretrained_base.trainable = False\n",
    "\n",
    "# Check your answer\n",
    "q_1.check()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "96ff70e5",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-03-13T16:23:01.649784Z",
     "iopub.status.busy": "2022-03-13T16:23:01.648933Z",
     "iopub.status.idle": "2022-03-13T16:23:01.651430Z",
     "shell.execute_reply": "2022-03-13T16:23:01.651017Z",
     "shell.execute_reply.started": "2022-03-13T16:12:47.864086Z"
    },
    "papermill": {
     "duration": 0.024816,
     "end_time": "2022-03-13T16:23:01.651587",
     "exception": false,
     "start_time": "2022-03-13T16:23:01.626771",
     "status": "completed"
    },
    "tags": []
   },
   "outputs": [],
   "source": [
    "# Lines below will give you a hint or solution code\n",
    "#q_1.hint()\n",
    "#q_1.solution()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "6d9cc53c",
   "metadata": {
    "papermill": {
     "duration": 0.018552,
     "end_time": "2022-03-13T16:23:01.688906",
     "exception": false,
     "start_time": "2022-03-13T16:23:01.670354",
     "status": "completed"
    },
    "tags": []
   },
   "source": [
    "# 2) Attach Head #\n",
    "\n",
    "Now that the base is defined to do the feature extraction, create a head of `Dense` layers to perform the classification, following this diagram:\n",
    "\n",
    "<figure>\n",
    "<img src=\"https://i.imgur.com/i5VU7Ry.png\" alt=\"Diagram of the dense head.\">\n",
    "</figure>\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "05b51d15",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-03-13T16:23:01.738164Z",
     "iopub.status.busy": "2022-03-13T16:23:01.737321Z",
     "iopub.status.idle": "2022-03-13T16:23:01.747624Z",
     "shell.execute_reply": "2022-03-13T16:23:01.747146Z",
     "shell.execute_reply.started": "2022-03-13T16:14:38.383786Z"
    },
    "lines_to_next_cell": 0,
    "papermill": {
     "duration": 0.039877,
     "end_time": "2022-03-13T16:23:01.747758",
     "exception": false,
     "start_time": "2022-03-13T16:23:01.707881",
     "status": "completed"
    },
    "tags": []
   },
   "outputs": [
    {
     "data": {
      "application/javascript": [
       "parent.postMessage({\"jupyterEvent\": \"custom.exercise_interaction\", \"data\": {\"outcomeType\": 1, \"valueTowardsCompletion\": 0.25, \"interactionType\": 1, \"questionType\": 2, \"questionId\": \"2_Q2\", \"learnToolsVersion\": \"0.3.4\", \"failureMessage\": \"\", \"exceptionClass\": \"\", \"trace\": \"\"}}, \"*\")"
      ],
      "text/plain": [
       "<IPython.core.display.Javascript object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/markdown": [
       "<span style=\"color:#33cc33\">Correct</span>"
      ],
      "text/plain": [
       "Correct"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "from tensorflow import keras\n",
    "from tensorflow.keras import layers\n",
    "\n",
    "model = keras.Sequential([\n",
    "    pretrained_base,\n",
    "    layers.Flatten(),\n",
    "    # YOUR CODE HERE. Attach a head of dense layers.\n",
    "    layers.Dense(units=6, activation='relu'),\n",
    "    layers.Dense(units=1, activation='sigmoid')\n",
    "])\n",
    "\n",
    "# Check your answer\n",
    "q_2.check()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "fc8e56b8",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-03-13T16:23:01.792408Z",
     "iopub.status.busy": "2022-03-13T16:23:01.791637Z",
     "iopub.status.idle": "2022-03-13T16:23:01.794254Z",
     "shell.execute_reply": "2022-03-13T16:23:01.793805Z",
     "shell.execute_reply.started": "2022-03-13T16:14:41.173352Z"
    },
    "papermill": {
     "duration": 0.026709,
     "end_time": "2022-03-13T16:23:01.794378",
     "exception": false,
     "start_time": "2022-03-13T16:23:01.767669",
     "status": "completed"
    },
    "tags": []
   },
   "outputs": [],
   "source": [
    "# Lines below will give you a hint or solution code\n",
    "#q_2.hint()\n",
    "#q_2.solution()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "dc661a73",
   "metadata": {
    "papermill": {
     "duration": 0.021159,
     "end_time": "2022-03-13T16:23:01.835640",
     "exception": false,
     "start_time": "2022-03-13T16:23:01.814481",
     "status": "completed"
    },
    "tags": []
   },
   "source": [
    "# 3) Train #\n",
    "\n",
    "Before training a model in Keras, you need to specify an *optimizer* to perform the gradient descent, a *loss function* to be minimized, and (optionally) any *performance metrics*. The optimization algorithm we'll use for this course is called [\"Adam\"](https://keras.io/api/optimizers/adam/), which generally performs well regardless of what kind of problem you're trying to solve.\n",
    "\n",
    "The loss and the metrics, however, need to match the kind of problem you're trying to solve. Our problem is a **binary classification** problem: `Car` coded as 0, and `Truck` coded as 1. Choose an appropriate loss and an appropriate accuracy metric for binary classification."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "7cba6826",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-03-13T16:23:01.886891Z",
     "iopub.status.busy": "2022-03-13T16:23:01.886298Z",
     "iopub.status.idle": "2022-03-13T16:23:01.897655Z",
     "shell.execute_reply": "2022-03-13T16:23:01.898072Z",
     "shell.execute_reply.started": "2022-03-13T16:16:11.741052Z"
    },
    "lines_to_next_cell": 0,
    "papermill": {
     "duration": 0.042342,
     "end_time": "2022-03-13T16:23:01.898216",
     "exception": false,
     "start_time": "2022-03-13T16:23:01.855874",
     "status": "completed"
    },
    "tags": []
   },
   "outputs": [
    {
     "data": {
      "application/javascript": [
       "parent.postMessage({\"jupyterEvent\": \"custom.exercise_interaction\", \"data\": {\"outcomeType\": 1, \"valueTowardsCompletion\": 0.25, \"interactionType\": 1, \"questionType\": 2, \"questionId\": \"3_Q3\", \"learnToolsVersion\": \"0.3.4\", \"failureMessage\": \"\", \"exceptionClass\": \"\", \"trace\": \"\"}}, \"*\")"
      ],
      "text/plain": [
       "<IPython.core.display.Javascript object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/markdown": [
       "<span style=\"color:#33cc33\">Correct</span>"
      ],
      "text/plain": [
       "Correct"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# YOUR CODE HERE: what loss function should you use for a binary\n",
    "# classification problem? (Your answer for each should be a string.)\n",
    "optimizer = tf.keras.optimizers.Adam(epsilon=0.01)\n",
    "model.compile(\n",
    "    optimizer=optimizer,\n",
    "    loss = 'binary_crossentropy',\n",
    "    metrics=['binary_accuracy'],\n",
    ")\n",
    "\n",
    "# Check your answer\n",
    "q_3.check()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "21c58ac3",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-03-13T16:23:01.944402Z",
     "iopub.status.busy": "2022-03-13T16:23:01.943650Z",
     "iopub.status.idle": "2022-03-13T16:23:01.946614Z",
     "shell.execute_reply": "2022-03-13T16:23:01.947014Z",
     "shell.execute_reply.started": "2022-03-13T16:16:15.516775Z"
    },
    "lines_to_next_cell": 0,
    "papermill": {
     "duration": 0.027785,
     "end_time": "2022-03-13T16:23:01.947162",
     "exception": false,
     "start_time": "2022-03-13T16:23:01.919377",
     "status": "completed"
    },
    "tags": []
   },
   "outputs": [],
   "source": [
    "# Lines below will give you a hint or solution code\n",
    "#q_3.hint()\n",
    "#q_3.solution()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "528bcb64",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-03-13T16:23:01.996085Z",
     "iopub.status.busy": "2022-03-13T16:23:01.995533Z",
     "iopub.status.idle": "2022-03-13T16:25:51.150625Z",
     "shell.execute_reply": "2022-03-13T16:25:51.151421Z",
     "shell.execute_reply.started": "2022-03-13T16:16:17.862982Z"
    },
    "papermill": {
     "duration": 169.182545,
     "end_time": "2022-03-13T16:25:51.151695",
     "exception": false,
     "start_time": "2022-03-13T16:23:01.969150",
     "status": "completed"
    },
    "tags": []
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2022-03-13 16:23:02.029146: I tensorflow/compiler/mlir/mlir_graph_optimization_pass.cc:185] None of the MLIR Optimization Passes are enabled (registered 2)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/30\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2022-03-13 16:23:08.733350: I tensorflow/stream_executor/cuda/cuda_dnn.cc:369] Loaded cuDNN version 8005\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "80/80 [==============================] - 36s 300ms/step - loss: 0.7006 - binary_accuracy: 0.5419 - val_loss: 0.6893 - val_binary_accuracy: 0.6003\n",
      "Epoch 2/30\n",
      "80/80 [==============================] - 3s 43ms/step - loss: 0.6814 - binary_accuracy: 0.6490 - val_loss: 0.6632 - val_binary_accuracy: 0.7468\n",
      "Epoch 3/30\n",
      "80/80 [==============================] - 4s 44ms/step - loss: 0.6384 - binary_accuracy: 0.7340 - val_loss: 0.6062 - val_binary_accuracy: 0.7854\n",
      "Epoch 4/30\n",
      "80/80 [==============================] - 3s 43ms/step - loss: 0.5872 - binary_accuracy: 0.7799 - val_loss: 0.5632 - val_binary_accuracy: 0.8125\n",
      "Epoch 5/30\n",
      "80/80 [==============================] - 3s 43ms/step - loss: 0.5522 - binary_accuracy: 0.8013 - val_loss: 0.5348 - val_binary_accuracy: 0.8284\n",
      "Epoch 6/30\n",
      "80/80 [==============================] - 3s 43ms/step - loss: 0.5273 - binary_accuracy: 0.8165 - val_loss: 0.5126 - val_binary_accuracy: 0.8313\n",
      "Epoch 7/30\n",
      "80/80 [==============================] - 3s 43ms/step - loss: 0.5072 - binary_accuracy: 0.8247 - val_loss: 0.4958 - val_binary_accuracy: 0.8363\n",
      "Epoch 8/30\n",
      "80/80 [==============================] - 3s 43ms/step - loss: 0.4908 - binary_accuracy: 0.8306 - val_loss: 0.4818 - val_binary_accuracy: 0.8386\n",
      "Epoch 9/30\n",
      "80/80 [==============================] - 3s 43ms/step - loss: 0.4760 - binary_accuracy: 0.8355 - val_loss: 0.4697 - val_binary_accuracy: 0.8416\n",
      "Epoch 10/30\n",
      "80/80 [==============================] - 3s 43ms/step - loss: 0.4629 - binary_accuracy: 0.8409 - val_loss: 0.4592 - val_binary_accuracy: 0.8412\n",
      "Epoch 11/30\n",
      "80/80 [==============================] - 3s 43ms/step - loss: 0.4511 - binary_accuracy: 0.8456 - val_loss: 0.4491 - val_binary_accuracy: 0.8434\n",
      "Epoch 12/30\n",
      "80/80 [==============================] - 3s 43ms/step - loss: 0.4404 - binary_accuracy: 0.8493 - val_loss: 0.4406 - val_binary_accuracy: 0.8450\n",
      "Epoch 13/30\n",
      "80/80 [==============================] - 4s 44ms/step - loss: 0.4307 - binary_accuracy: 0.8521 - val_loss: 0.4323 - val_binary_accuracy: 0.8474\n",
      "Epoch 14/30\n",
      "80/80 [==============================] - 3s 43ms/step - loss: 0.4215 - binary_accuracy: 0.8556 - val_loss: 0.4250 - val_binary_accuracy: 0.8489\n",
      "Epoch 15/30\n",
      "80/80 [==============================] - 3s 43ms/step - loss: 0.4132 - binary_accuracy: 0.8581 - val_loss: 0.4182 - val_binary_accuracy: 0.8487\n",
      "Epoch 16/30\n",
      "80/80 [==============================] - 4s 46ms/step - loss: 0.4054 - binary_accuracy: 0.8593 - val_loss: 0.4123 - val_binary_accuracy: 0.8501\n",
      "Epoch 17/30\n",
      "80/80 [==============================] - 3s 43ms/step - loss: 0.3978 - binary_accuracy: 0.8624 - val_loss: 0.4065 - val_binary_accuracy: 0.8497\n",
      "Epoch 18/30\n",
      "80/80 [==============================] - 3s 44ms/step - loss: 0.3904 - binary_accuracy: 0.8648 - val_loss: 0.4015 - val_binary_accuracy: 0.8509\n",
      "Epoch 19/30\n",
      "80/80 [==============================] - 3s 43ms/step - loss: 0.3838 - binary_accuracy: 0.8663 - val_loss: 0.3968 - val_binary_accuracy: 0.8511\n",
      "Epoch 20/30\n",
      "80/80 [==============================] - 3s 43ms/step - loss: 0.3773 - binary_accuracy: 0.8677 - val_loss: 0.3930 - val_binary_accuracy: 0.8513\n",
      "Epoch 21/30\n",
      "80/80 [==============================] - 4s 45ms/step - loss: 0.3713 - binary_accuracy: 0.8677 - val_loss: 0.3885 - val_binary_accuracy: 0.8505\n",
      "Epoch 22/30\n",
      "80/80 [==============================] - 3s 44ms/step - loss: 0.3654 - binary_accuracy: 0.8687 - val_loss: 0.3846 - val_binary_accuracy: 0.8515\n",
      "Epoch 23/30\n",
      "80/80 [==============================] - 4s 44ms/step - loss: 0.3599 - binary_accuracy: 0.8716 - val_loss: 0.3811 - val_binary_accuracy: 0.8521\n",
      "Epoch 24/30\n",
      "80/80 [==============================] - 4s 45ms/step - loss: 0.3542 - binary_accuracy: 0.8726 - val_loss: 0.3786 - val_binary_accuracy: 0.8527\n",
      "Epoch 25/30\n",
      "80/80 [==============================] - 3s 43ms/step - loss: 0.3488 - binary_accuracy: 0.8741 - val_loss: 0.3758 - val_binary_accuracy: 0.8517\n",
      "Epoch 26/30\n",
      "80/80 [==============================] - 4s 44ms/step - loss: 0.3440 - binary_accuracy: 0.8741 - val_loss: 0.3730 - val_binary_accuracy: 0.8511\n",
      "Epoch 27/30\n",
      "80/80 [==============================] - 3s 43ms/step - loss: 0.3396 - binary_accuracy: 0.8767 - val_loss: 0.3697 - val_binary_accuracy: 0.8523\n",
      "Epoch 28/30\n",
      "80/80 [==============================] - 3s 44ms/step - loss: 0.3350 - binary_accuracy: 0.8782 - val_loss: 0.3672 - val_binary_accuracy: 0.8535\n",
      "Epoch 29/30\n",
      "80/80 [==============================] - 4s 44ms/step - loss: 0.3304 - binary_accuracy: 0.8800 - val_loss: 0.3636 - val_binary_accuracy: 0.8549\n",
      "Epoch 30/30\n",
      "80/80 [==============================] - 3s 43ms/step - loss: 0.3258 - binary_accuracy: 0.8802 - val_loss: 0.3613 - val_binary_accuracy: 0.8547\n"
     ]
    }
   ],
   "source": [
    "history = model.fit(\n",
    "    ds_train,\n",
    "    validation_data=ds_valid,\n",
    "    epochs=30,\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b725704d",
   "metadata": {
    "papermill": {
     "duration": 0.538806,
     "end_time": "2022-03-13T16:25:52.219320",
     "exception": false,
     "start_time": "2022-03-13T16:25:51.680514",
     "status": "completed"
    },
    "tags": []
   },
   "source": [
    "Run the cell below to plot the loss and metric curves for this training run."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "9421219d",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-03-13T16:25:52.899846Z",
     "iopub.status.busy": "2022-03-13T16:25:52.899190Z",
     "iopub.status.idle": "2022-03-13T16:25:53.501054Z",
     "shell.execute_reply": "2022-03-13T16:25:53.501488Z",
     "shell.execute_reply.started": "2022-03-13T16:20:52.615960Z"
    },
    "papermill": {
     "duration": 0.933714,
     "end_time": "2022-03-13T16:25:53.501650",
     "exception": false,
     "start_time": "2022-03-13T16:25:52.567936",
     "status": "completed"
    },
    "tags": []
   },
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAagAAAEYCAYAAAAJeGK1AAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjUuMSwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy/YYfK9AAAACXBIWXMAAAsTAAALEwEAmpwYAAA660lEQVR4nO3dd3xUVd7H8c8vkwokIZAG6UBCS6ihSBNEEGyoIAiroq7g2ld3fXS7svqsWx5XXV17Y1cFRFEUBVGpSklCDaGHkgQSQkKH9PP8cQdIAxJSZjL5vV+veU3mnjuT33WQL+fcc88VYwxKKaWUs3FzdAFKKaVUdTSglFJKOSUNKKWUUk5JA0oppZRT0oBSSinllNwdXUBlgYGBJjo62tFlKKWUagQpKSmHjTFB1bU5XUBFR0eTnJzs6DKUUko1AhHZd6E2HeJTSinllDSglFJKOSUNKKWUUk7J6c5BKaVUU1FcXExmZiYFBQWOLsXpeXt7Ex4ejoeHR43fowGllFKXKTMzE19fX6KjoxERR5fjtIwx5OXlkZmZSUxMTI3fp0N8Sil1mQoKCmjbtq2G0yWICG3btq11T1MDSiml6kDDqWYu579TjQJKRMaIyHYR2SUiT1XT/k8R2WB/7BCRo+XaporITvtjaq0rVEop1Sxd8hyUiNiAV4FRQCaQJCLzjTFpZ/cxxjxWbv+Hgd72n9sAfwISAQOk2N97pF6PopLsYwWE+Hnpv2yUUi6vVatWnDx50tFlNIia9KD6A7uMMenGmCJgFjDuIvtPBj62/3wNsNgYk28PpcXAmLoUfCkHj53h2pdX8MyXaejNGJVSqumqSUCFARnlXmfat1UhIlFADPBDbd4rItNFJFlEknNzc2tS9wWF+nkzrld73v9pL3/4IpWyMg0ppZTrM8bwxBNPEB8fT0JCArNnzwbg4MGDDBs2jF69ehEfH8+KFSsoLS3lrrvuOrfvP//5TwdXX736nmZ+GzDXGFNamzcZY94E3gRITEysU6KICH+8vhue7m68sSyd4hLDX25JwM1Nh/uUUg3nmS+3kHbgeL1+Zrf2fvzphu412vezzz5jw4YNbNy4kcOHD9OvXz+GDRvGRx99xDXXXMPvfvc7SktLOX36NBs2bCArK4vU1FQAjh49Wq9115eaBFQWEFHudbh9W3VuAx6s9N7hld67tOblXR4R4akxXfC0ufGvH3ZRXFbG3yf0xKYhpZRyUStXrmTy5MnYbDZCQkK48sorSUpKol+/ftxzzz0UFxdz00030atXLzp06EB6ejoPP/ww1113HaNHj3Z0+dWqSUAlAbEiEoMVOLcBUyrvJCJdgABgVbnNi4D/FZEA++vRwG/qVHENiQi/Gt0ZD5sbLyzeQXGp4Z8Te+Ju05n1Sqn6V9OeTmMbNmwYy5cvZ8GCBdx11108/vjj3HnnnWzcuJFFixbx+uuvM2fOHN59911Hl1rFJf+2NsaUAA9hhc1WYI4xZouIzBCRG8vtehswy5SbmWCMyQf+jBVyScAM+7ZG88jIWJ4c04UvNx7g4Y/XU1xa1pi/XimlGsXQoUOZPXs2paWl5Obmsnz5cvr378++ffsICQlh2rRp3Hvvvaxbt47Dhw9TVlbG+PHjefbZZ1m3bp2jy69Wjc5BGWO+Br6utO2PlV4/fYH3vgs0bjRnb4a2ncDDB4D7h3fEwyY8u2ArD3y4jlem9MbL3daoJSmlVEO6+eabWbVqFT179kRE+Nvf/kZoaCgffPABf//73/Hw8KBVq1bMnDmTrKws7r77bsrKrH+w/+Uvf3Fw9dUTZ5uKnZiYaOp0w8LjB+Dl3tDzNrjhpQpNM1ft5Y9fbGFE5yBeu70v3h4aUkqpy7d161a6du3q6DKajOr+e4lIijEmsbr9Xe+EjF97GPgApLwPG2dXaLrzimj+9+YElmzPZdrMZM4U1WqyoVJKqUbkegEFMOJ3EDUYvvol5G6v0DRlQCR/m9CDlbsOc/f7azlVWOKYGpVSSl2UawaUzR3GvwMeLWDOVCg6VaF5YmIE/5zYi7V78rnrvbWc1JBSSimn45oBBeDXDsa/DbnbYMGvoNK5tpt6h/Hy5N6s23+UO95Zw4mCYgcVqpRSqjquG1AAHUfAlU/Cxo9h/X+rNF/foz2vTunDxoyjvPLDLgcUqJRS6kJcO6AArvwfiLkSvv41ZKdWaR4TH8q1Ce34aM1+7UUppZQTcf2AcrNZQ33e/vDJVCg8UWWX+4Z15ERhCR+v3e+AApVSSlXH9QMKoFUwTHgX8tPhy0ernI9KCPdnUMe2vLtyL0UlutKEUsp1tWrV6oJte/fuJT4+vhGrubjmEVAA0UPgqt9D6qeQXHVhi+nDOpB9vID5Gw84oDillFKV1fftNpzb4Mdg3ypY+BSE9YX2vc41XRkXRJdQX95cvpvxfcL0brxKqdr55ilrmbX6FJoAY5+/6C5PPfUUERERPPigdSOJp59+Gnd3d5YsWcKRI0coLi7m2WefZdy4i91ntqqCggLuv/9+kpOTcXd354UXXmDEiBFs2bKFu+++m6KiIsrKyvj0009p3749EydOJDMzk9LSUv7whz8wadKkyz7ss5pPDwrAzQ1ufgNaBlnnowqOnWsSEaYP68COnJMs3V63myYqpVRjmTRpEnPmzDn3es6cOUydOpV58+axbt06lixZwq9+9ata32H81VdfRUTYvHkzH3/8MVOnTqWgoIDXX3+dRx99lA0bNpCcnEx4eDgLFy6kffv2bNy4kdTUVMaMqZ8bpzevHhRAy7Yw4T14/1r44kGY+B+w95Zu6Nmevy/azhvLdzOiS7CDC1VKNSmX6Ok0lN69e3Po0CEOHDhAbm4uAQEBhIaG8thjj7F8+XLc3NzIysoiJyeH0NDQGn/uypUrefjhhwHo0qULUVFR7NixgyuuuILnnnuOzMxMbrnlFmJjY0lISOBXv/oVTz75JNdffz1Dhw6tl2NrXj2osyIHwNVPw9YvYc3r5zZ72Nz4+ZAYVqfnszHjqMPKU0qp2rj11luZO3cus2fPZtKkSXz44Yfk5uaSkpLChg0bCAkJoaCgoF5+15QpU5g/fz4+Pj5ce+21/PDDD8TFxbFu3ToSEhL4/e9/z4wZM+rldzXPgAK44iHofB18+wfIPL96+m39I/H1dufN5ekOLE4ppWpu0qRJzJo1i7lz53Lrrbdy7NgxgoOD8fDwYMmSJezbt6/Wnzl06FA+/PBDAHbs2MH+/fvp3Lkz6enpdOjQgUceeYRx48axadMmDhw4QIsWLbj99tt54okn6u3+Us03oETgpletJZE+uQtOW/dRbOXlzu0Do/gm9SD78k5d/DOUUsoJdO/enRMnThAWFka7du342c9+RnJyMgkJCcycOZMuXbrU+jMfeOABysrKSEhIYNKkSbz//vt4eXkxZ84c4uPj6dWrF6mpqdx5551s3ryZ/v3706tXL5555hl+//vf18txud79oGorKwXeGgnDn7IewKHjBQz56xIm9Yvgzzc5zzUBSinnoveDqh29H1RthfW1rpHaPPfcBbzBft7c3DuMOckZ5J0sdHCBSinVPGlAAcTfAnk7K1zDMG1YDIUlZcxcVfuxW6WUcmabN2+mV69eFR4DBgxwdFlVNL9p5tXpOg6+fgJS50K7HgB0Cvbl6q7BzFy1l19c2REfT709vFKqKmNMk7uwPyEhgQ0bNjTq77yc00nagwLr2qgOIyD1swrr9N13ZUeOnC7mk5QMBxanlHJW3t7e5OXlXdZfvs2JMYa8vDy8vb1r9b4a9aBEZAzwEmAD3jbGVLkiTUQmAk8DBthojJli314KnB0722+MubFWFTaWhAkw7z7IWGtdJwUkRgXQO7I1b6/Yw5T+kbjbNM+VUueFh4eTmZlJbq6uPnMp3t7ehIeH1+o9lwwoEbEBrwKjgEwgSUTmG2PSyu0TC/wGGGyMOSIi5ZdhOGOM6VWrqhyh87Xg7m0tJmsPKBHhvmEd+cV/U1i4JZvre7R3cJFKKWfi4eFBTEyMo8twWTXpEvQHdhlj0o0xRcAsoPKqg9OAV40xRwCMMYfqt8xG4O0HsaNhyzwoLTm3eVS3EGICW/Lm8nTtxiulVCOqSUCFAeVPwmTat5UXB8SJyI8isto+JHiWt4gk27ffVN0vEJHp9n2SHdpVjh8Ppw7BvpXnNtnchHuHxrAp8xir0/MdV5tSSjUz9XVSxR2IBYYDk4G3RKS1vS3KfhHWFOBFEelY+c3GmDeNMYnGmMSgoKB6KukyxF0Dnq2sa6LKGd8nnMBWnryxfLeDClNKqeanJgGVBUSUex1u31ZeJjDfGFNsjNkD7MAKLIwxWfbndGAp0LuONTccDx/och1snQ8lRec2e3vYmHpFNEu357It+7gDC1RKqeajJgGVBMSKSIyIeAK3AfMr7fM5Vu8JEQnEGvJLF5EAEfEqt30wkIYzi59g3Sdq9/cVNt9xRRQ+HjZdRFYppRrJJQPKGFMCPAQsArYCc4wxW0RkhoicnTK+CMgTkTRgCfCEMSYP6Aoki8hG+/bny8/+c0odhoNPgDWbr5zWLTyZ1C+C+RsOcPDYGcfUppRSzYguFludLx+FTZ/AE7vAs8W5zRn5pxn+j6X8fEgMv71WF4hUSqm60sViayt+PBSfgh0LK2yOaNOC6xLa8dGa/RwvKHZQcUop1TxoQFUnajC0Cq0yzAcwfVgHThaW8GlKpgMKU0qp5kMDqjpuNuh+M+z81powUU58mD89wv2ZnZShF+4qpVQD0oC6kIQJUFoEW7+q0nRrYgTbsk+QmqVTzpVSqqFoQF1IWF9oHVXtMN+NPdvj5e7GnGRd5VwppRqKBtSFiFiTJdKXwqnDFZr8fTwYGx/K5xuyKCgudUx9Sinl4jSgLiZhAphSSPu8StPEfhGcKChh0Zbsxq9LKaWaAQ2oiwnuBkFdYHPVYb6BMW2JaOPD7CQd5lNKqYagAXUxItbSR/t/gmMVlx90cxNu7RvBT7vzyMg/7aAClVLKdWlAXUr8Ldbzls+qNE3oG44IfKKTJZRSqt5pQF1K247Qvne1s/nat/ZhaGwQc1MyKS3Ta6KUUqo+aUDVRPx4OLAe8qreD2pSYgQHjhXw467D1bxRKaXU5dKAqonu9mG+1KrDfFd3CyaghQezdZhPKaXqlQZUTfiHQeQgSJ0LlZY38nK3cVPvMBZvyeHIqaILfIBSSqna0oCqqYTxkLsNDlW9ndXExAiKSsv4fEPlGw0rpZS6XBpQNdXtJhAbbJ5bpalrOz9dQFYppeqZBlRNtQy07rab+mmVYT7QBWSVUqq+aUDVRvx4OLoPslKqNOkCskopVb80oGqj6/Vg86x2mE8XkFVKqfqlAVUb3v4QOxq2zIOyqiE0MVEXkFVKqfqiAVVb8ePhZDbsWV6laWAHawFZHeZTSqm6q1FAicgYEdkuIrtE5KkL7DNRRNJEZIuIfFRu+1QR2Wl/TK2vwh2m81hoEQirX6vSdHYB2R936QKySilVV5cMKBGxAa8CY4FuwGQR6VZpn1jgN8BgY0x34Jf27W2APwEDgP7An0QkoD4PoNF5+MCA+2DnIji0tUrzuQVkUzIdUJxSSrmOmvSg+gO7jDHpxpgiYBYwrtI+04BXjTFHAIwxh+zbrwEWG2Py7W2LgTH1U7oD9bsXPFrAT/+q0nRuAdnkDF1AViml6qAmARUGlD+pkmnfVl4cECciP4rIahEZU4v3Nj0t2kDv22HTHDh+oEqzLiCrlFJ1V1+TJNyBWGA4MBl4S0Ra1/TNIjJdRJJFJDk3N7eeSmpgVzxo3Q6+mnNRuoCsUkrVXU0CKguIKPc63L6tvExgvjGm2BizB9iBFVg1eS/GmDeNMYnGmMSgoKDa1O84AdHW8kcp70NBxdUjdAFZpZSqu5oEVBIQKyIxIuIJ3AbMr7TP51i9J0QkEGvILx1YBIwWkQD75IjR9m2uYfAjUHjcCqlKbu2rC8gqpVRdXDKgjDElwENYwbIVmGOM2SIiM0TkRvtui4A8EUkDlgBPGGPyjDH5wJ+xQi4JmGHf5hra94boodYwX0nFnlK39n4khOkCskopdblqdA7KGPO1MSbOGNPRGPOcfdsfjTHz7T8bY8zjxphuxpgEY8yscu991xjTyf54r2EOw4EG/xJOHLDuFVXJxH7WArJbDugCskopVVu6kkRddRoJwd2tKeeVekpnF5CdnaSTJZRSqrY0oOpKBAY9bN3IcNd3FZrOLiD7hS4gq5RStaYBVR/ix4NfGPz4UpWmSf0iOV5Qwkdr9jugMKWUaro0oOqDuycMvB/2roCsdRWaBnZow5BOgbz0/U6OntYp50opVVMaUPWlz1Tw8oOfXq6wWUT43XVdOVFQzMvf73JQcUop1fRoQNUXbz9IvBvSvoD8PRWaurbzY1K/CGau2kt67kkHFaiUUk2LBlR9GnA/iA1WvVql6bFRcXi5u/H8N9scUJhSSjU9GlD1ya8d9JgE6/8Lp/IqNAX7evPAiE58m5bDqt15F/gApZRSZ2lA1bdBD0PJGUh6q0rTz4fE0N7fm2cXpFGmt+JQSqmL0oCqb8FdIG4MrH0TiireVdfbw8aTY7uw5cBxPluva/QppdTFaEA1hEGPwOk82PhRlaYberSnZ0Rr/r5oG6eLShxQnFJKNQ0aUA0hahCE9YWfXoGyiitIuLkJf7iuKznHC3lzebqDClRKKeenAdUQRGDwo3BkD2z9skpzYnQbrktoxxvL0sk+VuCAApVSyvlpQDWULtdDmw7WhbvV3G7jyTFdKC0z/OPb7Q4oTimlnJ8GVENxs1m3hc9KgX0/VWmObNuCuwdH8+m6TFKzjjmgQKWUcm4aUA2p18+gRdtqF5EFePCqTgS08OTZBWl6U0OllKpEA6ohefhA//tg5yI4sL5Ks5+3B4+NimN1ej7fpuU4oECllHJeGlANbcB08G0H834BxVUnREzuF0Gn4Fb85eutFJWUOaBApZRyThpQDc0nAMa9Arnb4PsZVZrdbW787rqu7M07zX9W73NAgUop5Zw0oBpDp6uh3zRY/SqkL6vSPDwuiKGxgbys94xSSqlzNKAay6gZ0LYTfP4AnDlaoan8PaNe+n6nY+pTSiknowHVWDxbwM1vwomD8M2TVZq7hPoxqV8k/1m1T+8ZpZRS1DCgRGSMiGwXkV0i8lQ17XeJSK6IbLA/7i3XVlpu+/z6LL7JCe8Lw56ATbNgy+dVmh+33zPqL3rPKKWUunRAiYgNeBUYC3QDJotIt2p2nW2M6WV/vF1u+5ly22+sn7KbsGG/hvZ94KtfwonsCk1Bvl48MKITi9NymL/xgGPqU0opJ1GTHlR/YJcxJt0YUwTMAsY1bFkuzOYBt7xpTTn/4sEqyyBNG9qBftEBPDl3E9uyjzuoSKWUcryaBFQYkFHudaZ9W2XjRWSTiMwVkYhy271FJFlEVovITdX9AhGZbt8nOTc3t8bFN1mBsTD6z7DrO0h+p0KTp7sbr07pg6+3O/f9J4Vjp4sdVKRSSjlWfU2S+BKINsb0ABYDH5RrizLGJAJTgBdFpGPlNxtj3jTGJBpjEoOCguqpJCfX717oOBK+/QMc3lWhKdjPm9du78OBo2f45ez1evddpVSzVJOAygLK94jC7dvOMcbkGWMK7S/fBvqWa8uyP6cDS4HedajXdYjAuFfB5gnzpkNpxZsX9o1qwx9v6M6S7bm8qFPPlVLNUE0CKgmIFZEYEfEEbgMqzMYTkXblXt4IbLVvDxARL/vPgcBgIK0+CncJfu3g+n9aK56vfKFK8+0DIpnQN5yXv9/JYl2rTynVzFwyoIwxJcBDwCKs4JljjNkiIjNE5OysvEdEZIuIbAQeAe6yb+8KJNu3LwGeN8ZoQJUXfwsk3ApLn4esdRWaRIRnb4onIcyfx2dvYLdeH6WUakbE2W7zkJiYaJKTkx1dRuM6cwReGwyeLWH6Muui3nKyjp7hhn+tpG1LT+Y9OJhWXu4OKlQppeqXiKTY5ylUoStJOAOfALjp33B4B3z3dJXmsNY+vDK5N7tzT/I/czfqvaOUUs2CBpSz6DAcBtwPa9+A3T9UaR7UKZCnxnbh683ZvL4svfHrU0qpRqYB5Uyu/hMEdrYWlD1Z9XqwaUM7cF2Pdvx90TZW7GwG14sppZo1DShn4uED49+yVjv/6FYorDgpQkT42/gexAb78vDH68nIP+2YOpVSqhFoQDmbdj3h1vfg4EaYe3eV66Naernz+h19KS0z/OK/KRQUlzqoUKWUalgaUM6o81i47gXY+a21qGylSRExgS15cVIvthw4zm8/26yTJpRSLkkDylkl3g3D/gfW/8e6RqqSkV1D+OXVsXy2PouZq/RW8Uop16MX1DizEb+F4wdg2fPg1x76Tq3Q/MhVsWzOPMafv0oj1N+ba7qHOqhQpZSqf9qDcmYicMOL0Olq+Oox2LGoQrObm/Dibb1ICPfnoY/W8f1WXQ5JKeU6NKCcnc0Dbv0AQhPgk7sgM6VCs6+3Bx/c059u7fy4/7/rWLL9kGPqVEqpeqYB1RR4tYKffQItg6zp53m7KzT7eXsw854BxIW24r7/pLB8h14jpZRq+jSgmopWwXD7Z9aMvv+Or3Ihr38LD/5zzwA6BrVi2sxkftx12EGFKqVU/dCAakoCO8GUOXAiGz6aCEWnKjQHtPTkw3sHEBPYkp9/kMSq3XkOKlQppepOA6qpiehnv5B3g3VOqtKFvG1aevLfewcQEdCCe95PYu2efIeUqZRSdaUB1RR1HgvX/d8FL+QNbOXFR9MG0r61N3e/t5aUfRpSSqmmRwOqqUq8B4Y9ccELeYN8vfh42kCC/byZ+m4S6/cfcUCRSil1+TSgmrIRv4NeP7Mu5F30OyiruC5fsJ83H08bSNtWntz5zlo2Zhx1TJ1KKXUZNKCaMhG44WXoPx1WvQJz7qwycSLU3wqp1i09uOOdNaRmHXNQsUopVTsaUE2dzR2u/TuM+Sts/xreu9aa5VdO+9Y+fDxtIL7eHtz+zhrSDhx3ULFKKVVzGlCuYuAv4LaP4fBOeGskZKdWaA4PaMHH0wbSwsPGlLdXsyZdp6ArpZybBpQr6TwG7vkGTCm8ew3sXFyhObJtC2ZNv4I2LT25/Z01zE7a76BClVLq0moUUCIyRkS2i8guEXmqmva7RCRXRDbYH/eWa5sqIjvtj6mV36vqWbueMO0HaNPBuph37VsVmiPbtmDeA4MZ2KEtT366mecWpFFapveTUko5n0sGlIjYgFeBsUA3YLKIdKtm19nGmF72x9v297YB/gQMAPoDfxKRgHqrXlXPrz3c/Q3EXgNf/xoW/qbCDD9/Hw/eu6sfU6+I4q0Ve5g2M5kTBcUOLFgppaqqSQ+qP7DLGJNujCkCZgHjavj51wCLjTH5xpgjwGJgzOWVqmrFqxXc9iEMuB9W/xtm3w6FJ881u9vceGZcPH8e151lO3KZ8NoqMvJPO7BgpZSqqCYBFQZklHudad9W2XgR2SQic0UkojbvFZHpIpIsIsm5uboSd71xs8HY5+Haf8COhfDeWOsGiOXccUU0H9zdn4PHznDTqz+StFdXnVBKOYf6miTxJRBtjOmB1Uv6oDZvNsa8aYxJNMYkBgUF1VNJ6pz+06xFZvPTrRl+BzdVaB4SG8i8Bwfj5+PBz95aw9yUTAcVqpRS59UkoLKAiHKvw+3bzjHG5BljCu0v3wb61vS9qpHEjoJ7FloX9757DaS8X2ENv45BrZj3wCASowP49Scbef6bbZTp5AmllAPVJKCSgFgRiRERT+A2YH75HUSkXbmXNwJb7T8vAkaLSIB9csRo+zblCKEJ1gy/iP7w5aMw62dw6vz1UK1bePLBPf2ZMiCS15ft5r7/pnCqsOQiH6iUUg3nkgFljCkBHsIKlq3AHGPMFhGZISI32nd7RES2iMhG4BHgLvt784E/Y4VcEjDDvk05im8o3D4PRj8HuxbDa1fAru/PNXvY3HjupnievqEb32/NYfxrP5F19IwDC1ZKNVdijHMN4yQmJprk5GRHl9E8ZG+GT++F3G0w8AEY+Sfw8D7XvHT7IR7+aD2e7m78c1IvhsXp+UGlVP0SkRRjTGJ1bbqSRHMWmgDTl1qLza7+N7x1FeSknWse3jmYeQ8Opm0rT6a+t5Z/LNpOSWmZ4+pVSjUrGlDNnYePtdjslE/g1CF4czisfv3cBIpOwa344sEh3No3nFeW7GLKW2vIPlbg2JqVUs2CBpSyxI2G+1dBxxGw8En4cAKcyAHAx9PG3yb05J+TepJ64BjXvryCpdsPObhgpZSr04BS57UKgsmzrNvJ7/3RmkCx7etzzTf3Dmf+Q0MI9vXirveS+OvCbTrkp5RqMBpQqiIR6Hcv3LcM/MJg1mSY/wicOQpYQ36fPziYyf0jeG3pbm57czUHdJafUqoBaECp6gV1hnu/h8GPwvr/wCv9YNMnYAzeHjb+cksPXrqtF1sPHue6l1ewZJsO+Sml6pcGlLowd08YNcOa6ecfDp/dCzPHweFdAIzrFcaXDw8h1N+Hu99P4i9fb6VYh/yUUvVEA0pdWruecO931rmpAxusc1NL/gLFBXSwL5H0swGRvLE8nUlvrNILe5VS9UIDStWMm806N/VQEnQbB8uet4Jq9w94e9h47uYE/jW5NztyTjLmxeXMTcnE2S4CV0o1LRpQqnZ8Q2D823DH54DAf26GuT+HEznc0LM9Cx4ZQpdQX379yUZ+/kEyOcf1miml1OXRgFKXp+MIuP8nGP4b2PolvJIIa98iKsCb2dOv4I/Xd+On3YcZ9cIy7U0ppS6LBpS6fB7eMPwpeGAVhPW1bi//9kjcDq7nniExfPPoMDrbe1P3am9KKVVLGlCq7tp2hDvmwfh3rDv2vjUCPrmLGA4ya/oV/OH6bvxo7019tk57U0qpmtHVzFX9KjgGP70Cq16FkgLocwdc+SR7ivx54pONJO87wsguwfzvLQmE+Hlf+vOUUi7tYquZa0CphnHyEKz4P0h6x5oB2H86pYMf4711R/n7ou14ubvx9I3dubl3GCLi6GqVUg6it9tQja9VMIz9KzycAt1vhp/+he3lXtxrPmXh/X2IDfHl8TkbmTYzhUN6bkopVQ3tQanGkZMGPzwL2xdAy2DKhv6a9wuH89fF6Xja3Hh8dBx3DIzC3ab/ZlKqOdEelHK8kG4w+SP4+WIIjMNt4f9wz/pbWTE2l94RvjzzZRrX/2sla/fkO7pSpZST0IBSjSuiP9z1Fdz+KXj7E7z4YT4ofJx5Iw5z4kwxE99YxeOzN3DohA77KdXcaUCpxicCna6G6ctgwntIWQm9Vz3CioBn+L9eB/lq0wFG/mMZ76zco/ebUqoZ03NQyvFKS2DzJ9b6fkf2UhDSh5fMJF7bH06XUD+eubE7Azq0dXSVSqkGUOdzUCIyRkS2i8guEXnqIvuNFxEjIon219EickZENtgfr1/eISiXZnOHXpPhoWS44SW8zxziyUNPsi78JTqc3sSkN1fz2OwNOttPqWbmkj0oEbEBO4BRQCaQBEw2xqRV2s8XWAB4Ag8ZY5JFJBr4yhgTX9OCtAelKCmElA9gxT/gZA57/AfwRN4NbLPF8dioOKZeobP9lHIVde1B9Qd2GWPSjTFFwCxgXDX7/Rn4K6D/zFV14+4FA6bDIxtg9LPEFO9irvvvmdniBT5d8DXXvLicb7dk65JJSrm4mgRUGJBR7nWmfds5ItIHiDDGLKjm/TEisl5ElonI0Op+gYhMF5FkEUnOzc2tae3K1Xm2gEEPw6Mb4arf09ts5Wuv3/LM6f/lvQ9nMvH1n1i3/4ijq1RKNRD3un6AiLgBLwB3VdN8EIg0xuSJSF/gcxHpbow5Xn4nY8ybwJtgDfHVtSblYrx8YdgTSL9psPrfDE56myGla9iRE8Xbb4zmvbhbeOzaHnQIauXoSpVS9agmPagsIKLc63D7trN8gXhgqYjsBQYC80Uk0RhTaIzJAzDGpAC7gbj6KFw1Qz6tYcRvkce2wI2v0CmoJX/zeIsZ6RNZ+PKD/HXOD+SeKHR0lUqpelKTSRLuWJMkRmIFUxIwxRiz5QL7LwV+bZ8kEQTkG2NKRaQDsAJIMMZccLkAnSShaswY2LuCwpWv4rF7EaXGjW8ZwPGe93LjdeNo6VXnAQKlVAO72CSJS/4fbIwpEZGHgEWADXjXGLNFRGYAycaY+Rd5+zBghogUA2XALy4WTkrVigjEDMMrZhjk7+HEsn9z1eYP8dl0D6mbYjnS4x4GXn8PHp56Ww+lmiK9UFe5lsITZCx5B1vSm7QvzeKwBHC4y+3Ejn0Im1+oo6tTSlWi94NSzY4pK2XDkrmU/vQaiaXrKcad7PBraDfqEdwjB1i9L6WUw2lAqWartMywfNVPHF32GiMLv8NPznDEvxu+wx7AvccE8PBxdIlKNWsaUKrZKyszfL8pne3fvs2ok/Pp7JZJoYc/7ol3Yut/LwREO7pEpZolDSil7IwxfJ+Ww5JFnzH4yDxG25KxYSiLvQbbgOnQYQS46TJKSjUWDSilKjHGsGxHLh9+u4qEnM+43X0JbThGWZtOuPW9E3pMAl+dVKFUQ9OAUuoCjDH8tDuPf3+3hcD9C7nH83t6sh0jbkjHq6DnZOhynZ6rUqqBaEApVQOr0/N45YddHNi9ids8f2SS10/4F+WAlz90vwl6TYEInQGoVH3SgFKqFjZnHuP15btZuDmLIbatPBKYQu+Ty3ArOQNtOli9qh6TICDK0aUq1eRpQCl1GfYePsXbK9P5JDkT99LTPBG+nfHuy/E9uMraIXoo9LwNut4I3n6OLVapJkoDSqk6OHyykPd/3MvMVXs5XlDCdRHF/Cp0PTFZXyL5u8HmBbGjIP4WiBsDni0dXbJSTYYGlFL14GRhCbPW7uedlXs4eKyALiGteKrHSYYWLseW9jmczAaPFlZIxY+HTleDh64DqNTFaEApVY+KSsr4cuMB3li+mx05Jwn29eL2/uHcEXaAgPQvIe0LOJ0HXn7Q+VorrDoMB3dPR5eulNPRgFKqAZSVWddSfbBqL0u35+JhE8bGt2PqwDD6lKYiWz6FrV9CwTHwbg1db7DCKnoI2DwcXb5STkEDSqkGtufwKf6zah+fpGRwoqCE+DA/7hwYzY0JgXjvWwZbPoNtC6DoJHj7W8N/cWOh00ho0cbR5SvlMBpQSjWSU4UlfL4hiw9+2suOnJO0buHBpH4R3D4gighfgV3fwfaFsHMRnMoFsUHkQIi7xgqswFi9zko1KxpQSjUyYwyr0/OZuWov36blUGYMI7uEMHVQFIM7BuKGgQPrYPs3sGMR5Gy23hgQA53HWoEVOUjPWymXpwGllAMdOHqGD9fs4+O1GeSfKiKqbQsm9YtgQt9wgn3ts/yOZli9qu0LYc9yKC20Jll0vMqaFRg7CloGOvZAlGoAGlBKOYGC4lK+ST3Ix2szWLsnH3c3YWTXYCb3j2RobBA2N/vQXtEpSF9q9a52fgsncwCBsL5Wzyp2NLTrqUOByiVoQCnlZHbnnmR2UgZzUzLJP1VEWGsfJiZGMLFfOO38yy1MW1YG2RutYcAdi6xhQQDfdlavKvYaawq7VyuHHIdSdaUBpZSTKiwpZXFaDrPWZrBy12HcBIZ3Dua2fhFc1SUYd1ule1OdPAQ7F1vDgbt+gKITYPOEqMHne1dtOmjvSjUZGlBKNQH7804zO3k/nyRncuhEIcG+XtyaGM6kxEgi27ao+oaSIti/yhoG3LEI8nZa2/0jIGaYtVZgzFDwD2/cA1GqFuocUCIyBngJsAFvG2Oev8B+44G5QD9jTLJ922+AnwOlwCPGmEUX+10aUKq5Kykt44dth5iVlMHS7YcoMzCoY1sm9Yvgmu6heHvYqn9jfjrs/A72Loe9K+HMEWt7QIwVWGdDyzek8Q5GqUuoU0CJiA3YAYwCMoEkYLIxJq3Sfr7AAsATeMgYkywi3YCPgf5Ae+A7IM4YU3qh36cBpdR5B4+d4dOUTGYnZ5CRfwY/b3du7h3GxH4RdG/vf+E3lpVBTirsXQF7VsC+H6HwuNUW2NnqWUXbHy3bNs7BKFWNugbUFcDTxphr7K9/A2CM+Uul/V4EFgNPAL+2B1SFfUVkkf2zVl3o92lAKVVVWZlhdXoes5Mz+CY1m6KSMuLD/JiUGMGNvcLw97nE0kllpXBwozWFfe8K2LcKik9ZbUFdIGqQdR4r8grwD2v4A1LK7mIB5V6D94cBGeVeZwIDKv2CPkCEMWaBiDxR6b2rK71X//QrVUtubsKgToEM6hTIM6eL+GLDAWYlZfCHL7bw7IKtXJvQjomJEQzs0AapboKEmw3C+liPIb+E0mI4sN4KrP2rYNMnkPyutW/rKCusogZZD510oRykJgF1USLiBrwA3FWHz5gOTAeIjIysa0lKubTWLTyZOiiaO6+IIjXrOLOT9/PFhgPMW59FVNsWTOgTzs19wggPqGZixVk2D4jobz0ASkus1Sz2rbKGA3cugo0fWW2tQqygihxkLcsU3FUXu1WNos5DfCLiD+wGTtrfEgrkAzdinbfSIT6lGtiZolIWbjnI7KQMVqfnA9bEivF9whmbEEoLz1r+W9QYOLzDCqt9P1mP41lWm80TgrtBux4Q2gPa9YKQ7uB5kUBU6gLqeg7KHWuSxEggC2uSxBRjzJYL7L+U8+egugMfcX6SxPdArE6SUKrhZOSfZt76LOamZLI//zQtPW2MTWjHhL7h9I9ug5vbZQzXGQNH90PGWuvC4YObIHvT+ZmC4gZtY60VLs4FVw/wCajfg1Mupz6mmV8LvIg1zfxdY8xzIjIDSDbGzK+071LsAWV//TvgHqAE+KUx5puL/S4NKKXqhzGGpL1H+DQlkwWbD3KysITwAB/G9wlnfJ/w6q+tqt0vgGMZ58Pq7PPZnhZY57MiBlhDiZEDrZ6X2wWmyatmSS/UVaqZO1NUyqIt2Xy6LpOVuw5jDPSPacOEPuFc0z0U/xb1eE7p1GFrxmD2JshKsXpdJ3OsNs9WEJ54PrTC+1n3x1LNlgaUUuqcA0fPMG99Fp+mZJJ++BQeNmFobBDX92jH1d1C8POu5wkQ54YH15x/5GwBUwaI1auK6G+FVng/a9agm9slP1a5Bg0opVQVxhg2ZR5jweaDLNh0kKyjZ/C0uXFlZyusRnYNoZVXnSf6Vq/wBGQmW72rjDWQmXT+QmIvP+scVvte1gSM9r2gTUcNLRelAaWUuqiyMsOGzKN8tfEgX28+SPbxArzc3RjROZjre7bjqi7BtZ8JWKsCSiF3mzUkeGADHNwA2anWfbHAGhqsHFptO+n5LBegAaWUqrGyMkPK/iMs2HSQBZsPknuiEB8PG1d1Deb6hHYM7xyMj2cjBENpsRVaBzeWC63NUFJgtXu0hNB4a4gwpLv1CO4GPq0bvjZVbzSglFKXpbTMkLQ3n682HeCbzdnknSrCx8PGiC5BjI23elYtG2oYsNqCSuDw9oq9rENboODY+X38ws6H1dngahsL7p6NV6eqMQ0opVSdlZSWsXZPPt+kZvNNajaHTxbi5e7GlXFBXJvQjqu6Btf/BIuaMMaa2p6TZoVVThocSoPc7VBWbO3j5g6BcRAYa019D4iC1tHWs38EeHg3ft0K0IBSStWz0jJDyr4jfL35IAtTs8k+XoCnzY2hsYGMiQ9lVLcQWrdwcI+ltBgO77TCKmeL9Zy3y5pRWFpUcV/fduWCy/4cEG3dqsS3nU7QaEAaUEqpBlNWZlifcZRvNh/km9Rsso6ewd2+uO3Y+FBGdg0m2NeJeihlZXAyG47sg6P7yj3vtX4+ngWU+3vR3dsKqjYdoM3Z5w7QtqM1nKgTNepEA0op1SjOTl3/OvUg32zOZn/+aUSgV0RrRnULYXS3EDoGtap+xXVnUVJkrZBxdB/k77FuBHnuOf38zEKw1iUMiD4fWuUf/hFga8Tzc02UBpRSqtEZY9iWfYLFaTksTsthc5Y1kSEmsCWjuoUwqlsIfSIDsF3O2oCOUlYGJw6cD6uzjzz7c8mZ8/u6eVhDhW06WNdxnet5dQD/SA0vOw0opZTDHTx2hu/Scvg2LYfV6XkUlxratvTkqi7BjOoWwtDYoMaZvt5QjIET2ZC/u1xw7T7f+zp7g0iwJm20jrTubhzcBYK6Ws+BceDh47hjcAANKKWUUzleUMyy7bksTsthyfZDnCgowdvDjSGdAhnZNYSRXYIJ9nOi81Z1ZYy1HmGF4NptzTTM2wVlJdZ+4mad7wruat3p+OxzYCy4ezn2GBqIBpRSymkVlVjT1xenZfPd1kNkHbWGyRLC/BnZNZiru4bQvb2fc5+3qouSIiusDm21Lkw++5y3G87emUhs5ydptI60zm+1jjz/aBnUZO96rAGllGoSjDFszznB91sP8f3WHNZnHMUYCPHz4qouIVzdNZhBHQOb9lBgTZUUWtPky4fW0X1wNAMKjlbc1927XGidfY6y977inPoiZQ0opVSTdPhkIUu35/L91hyW78jlVFEp3h5uDO5oDQWO6BJEO//mdc4GgILj9pmG+63AOrrP+vnsttN55/d184CgzhASby0NFWJ/tApyXP3laEAppZq8wpJS1u7J5/uth/huaw6ZR6yhwLiQVgyLDeLKzkH0i26Dt0cz6F1dStEp65quQ2nW+oU5WyAnFU4cPL9PqxB7WHWH0ATruXUUeLVq1FI1oJRSLsUYw85DJ1m6/RDLdxxm7Z58ikrL8PZwY0BMW4bFBXFlXBAdg1q67rmry3EqD3LsgZWdav2cu73iyho+AeAfbk2F9w+3hgzLv24ZVK8ra2hAKaVc2umiEtak57NsRy7Ld+SSftia0h3W2odhcYEMiw1iUKdA/H0csFagsyu/JNSxDGvI8Fjm+Z+LTlTc3+YF/mFWWF3/orWiRh1cLKD0SjGlVJPXwtOdEV2CGdElGICM/NMs35nLsu25fLnxIB+vzcDmJvSOaM2ILsFcGRfk2jMDa8PmASHdrEd1zhy1B5Y9tMqHmHvDXgqgPSillEsrLi1j/f6jLNtxiGU7cknNsu7cG+zrxZVxQYzoEsyQ2EDHrMSudIhPKaXOOnSigGXbc1m6PZflO3M5UVCCzU3oGxXA8M5BDI8Lpms7X+1dNRINKKWUqkZJaRnrM46yZNshlm7PJe2g1bsK9fPmyjhrZuDgjoH4t9DeVUOpc0CJyBjgJcAGvG2Meb5S+y+AB4FS4CQw3RiTJiLRwFZgu33X1caYX1zsd2lAKaUcJee4vXe14xArdhzmRGEJbvbV2IfFBTE0Noie4f642/T+UPWlTgElIjZgBzAKyASSgMnGmLRy+/gZY47bf74ReMAYM8YeUF8ZY+JrWqwGlFLKGRSXlrEh4ygrduSybOdhNmVaq1r4ebszJDaQobFBDIsLIqx1M7xQuB7VdRZff2CXMSbd/mGzgHHAuYA6G052Lalwty+llGp6PGxu9ItuQ7/oNjw+ujNHThWxctdhVuzMZfmOw3y9ORuAjkEtGRYXxLDYIPrHtKGll06Ori81+S8ZBmSUe50JDKi8k4g8CDwOeAJXlWuKEZH1wHHg98aYFdW8dzowHSAyMrLGxSulVGMJaOnJDT3bc0PP9ucuFF6+I5flOw/z0Zr9vPfjXmxuQnyYPwNi2tDfHm56/ury1WSIbwIwxhhzr/31HcAAY8xDF9h/CnCNMWaqiHgBrYwxeSLSF/gc6F6px1WBDvEppZqaguJSkvbmsyY9nzV78tiYcYyi0jJEoHOILwM7tKV/jBVYQb6ueduMy1XXIb4sIKLc63D7tguZBbwGYIwpBArtP6eIyG4gDtAEUkq5DG8PG0NjrUkUYAXWhoyjrN2Tz9o9+cxOyuD9n/YC0CGopdXDimlDYlQbwgN8dEr7BdQkoJKAWBGJwQqm24Ap5XcQkVhjzE77y+uAnfbtQUC+MaZURDoAsUB6fRWvlFLOyNvDxsAObRnYoS1gTbhIzTp2LrC+2mStbgEQ5OtF38gA+kYF0CcqgPgwP7zcdcFbqEFAGWNKROQhYBHWNPN3jTFbRGQGkGyMmQ88JCJXA8XAEWCq/e3DgBkiUgyUAb8wxuQ3xIEopZSz8rC50TsygN6RAdx3ZUdKywzbs0+Qsv8I6/YdIWXfERZusSZdeNrcSAj3p09kayu0IgNc6+7CtaAX6iqllBM4dKKAdfuOsm6/FVibM63zWAARbXzoExlAYlQAfaPa0DnUF5ubawwL6koSSinVxBSWlJKadZz19sBK3neE3BOFAPh6udMrsjWJUW1IjA6gV0TrJju9XQNKKaWaOGMMGflnSN6XT/K+I6TsPcKOQycwBmxuQtd2viRGtaFvVACJ0QFN5k7DGlBKKeWCjp0ptoYE9x4heV8+GzKOUlBsDQuGtfahX3QAidHWjMFOQa1wc8JhQb0flFJKuSB/Hw9GdA5mRGfrPljFpWVsPXicZHtg/bg7j883HACgdQsPEqOswOoX3YaEMH883Z17TUHtQSmllIsyxrA//zRr9+STvPcISXvzz91t2MvdjV4RrekXbZ3H6hsVgK8D7omlQ3xKKaUAOHyykOS9+STZA2vLgeOUlhncBDqH+tE3qvW5c1mNcRGxBpRSSqlqnSosObfqRcq+I6zff4RTRaUAhPh50dc+tb1vVADd2/vhUc+3GtFzUEopparV0sudwZ0CGdwpEIDSMsO27OOk2C8gTt575NzK7d4ebvQMb31upuCAmLYNOr1de1BKKaUuKvtYAcn78s+F1tlhwa8fGUq39n51+mztQSmllLpsof7eXN+jPdf3aA/A6SJrWLBzqG+D/l4NKKWUUrXSwtOdQR0DG/z3OPckeKWUUs2WBpRSSimnpAGllFLKKWlAKaWUckoaUEoppZySBpRSSimnpAGllFLKKWlAKaWUckoaUEoppZyS063FJyK5wL56+KhA4HA9fE5T0JyOFfR4XVlzOlZoXsd7oWONMsYEVfcGpwuo+iIiyRdagNDVNKdjBT1eV9acjhWa1/FezrHqEJ9SSimnpAGllFLKKblyQL3p6AIaUXM6VtDjdWXN6ViheR1vrY/VZc9BKaWUatpcuQellFKqCdOAUkop5ZRcLqBEZIyIbBeRXSLylKPraWgisldENovIBhFJdnQ99U1E3hWRQyKSWm5bGxFZLCI77c8BjqyxvlzgWJ8WkSz797tBRK51ZI31SUQiRGSJiKSJyBYRedS+3eW+34scq0t+vyLiLSJrRWSj/XifsW+PEZE19r+fZ4uI50U/x5XOQYmIDdgBjAIygSRgsjEmzaGFNSAR2QskGmNc8mI/ERkGnARmGmPi7dv+BuQbY563/yMkwBjzpCPrrA8XONangZPGmH84sraGICLtgHbGmHUi4gukADcBd+Fi3+9FjnUiLvj9iogALY0xJ0XEA1gJPAo8DnxmjJklIq8DG40xr13oc1ytB9Uf2GWMSTfGFAGzgHEOrknVgTFmOZBfafM44AP7zx9g/Y/e5F3gWF2WMeagMWad/ecTwFYgDBf8fi9yrC7JWE7aX3rYHwa4Cphr337J79bVAioMyCj3OhMX/kNgZ4BvRSRFRKY7uphGEmKMOWj/ORsIcWQxjeAhEdlkHwJs8sNd1RGRaKA3sAYX/34rHSu46PcrIjYR2QAcAhYDu4GjxpgS+y6X/PvZ1QKqORpijOkDjAUetA8TNRvGGqN2nXHqql4DOgK9gIPA/zm0mgYgIq2AT4FfGmOOl29zte+3mmN12e/XGFNqjOkFhGONbnWp7We4WkBlARHlXofbt7ksY0yW/fkQMA/rD4Kry7GP6Z8d2z/k4HoajDEmx/4/ehnwFi72/drPT3wKfGiM+cy+2SW/3+qO1dW/XwBjzFFgCXAF0FpE3O1Nl/z72dUCKgmItc8U8QRuA+Y7uKYGIyIt7SdcEZGWwGgg9eLvcgnzgan2n6cCXziwlgZ19i9qu5txoe/XfiL9HWCrMeaFck0u9/1e6Fhd9fsVkSARaW3/2Qdr4tpWrKCaYN/tkt+tS83iA7BP03wRsAHvGmOec2xFDUdEOmD1mgDcgY9c7XhF5GNgONZS/TnAn4DPgTlAJNatWSYaY5r85IILHOtwrOEfA+wF7it3fqZJE5EhwApgM1Bm3/xbrHMzLvX9XuRYJ+OC36+I9MCaBGHD6gjNMcbMsP+dNQtoA6wHbjfGFF7wc1wtoJRSSrkGVxviU0op5SI0oJRSSjklDSillFJOSQNKKaWUU9KAUkop5ZQ0oJRSSjklDSillFJO6f8BOniLUzjBivQAAAAASUVORK5CYII=\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": [
    "import pandas as pd\n",
    "history_frame = pd.DataFrame(history.history)\n",
    "history_frame.loc[:, ['loss', 'val_loss']].plot()\n",
    "history_frame.loc[:, ['binary_accuracy', 'val_binary_accuracy']].plot();"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "49f39fcc",
   "metadata": {
    "papermill": {
     "duration": 0.43551,
     "end_time": "2022-03-13T16:25:54.258582",
     "exception": false,
     "start_time": "2022-03-13T16:25:53.823072",
     "status": "completed"
    },
    "tags": []
   },
   "source": [
    "# 4) Examine Loss and Accuracy #\n",
    "\n",
    "Do you notice a difference between these learning curves and the curves for VGG16 from the tutorial? What does this difference tell you about what this model (InceptionV2) learned compared to VGG16? Are there ways in which one is better than the other? Worse?\n",
    "\n",
    "After you've thought about it, run the cell below to see the answer."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "2e9aec55",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-03-13T16:25:55.082465Z",
     "iopub.status.busy": "2022-03-13T16:25:55.081607Z",
     "iopub.status.idle": "2022-03-13T16:25:55.086026Z",
     "shell.execute_reply": "2022-03-13T16:25:55.086515Z",
     "shell.execute_reply.started": "2022-03-13T16:21:10.389314Z"
    },
    "papermill": {
     "duration": 0.346318,
     "end_time": "2022-03-13T16:25:55.086691",
     "exception": false,
     "start_time": "2022-03-13T16:25:54.740373",
     "status": "completed"
    },
    "tags": []
   },
   "outputs": [
    {
     "data": {
      "application/javascript": [
       "parent.postMessage({\"jupyterEvent\": \"custom.exercise_interaction\", \"data\": {\"outcomeType\": 1, \"valueTowardsCompletion\": 0.25, \"interactionType\": 1, \"questionType\": 4, \"questionId\": \"4_Q4\", \"learnToolsVersion\": \"0.3.4\", \"failureMessage\": \"\", \"exceptionClass\": \"\", \"trace\": \"\"}}, \"*\")"
      ],
      "text/plain": [
       "<IPython.core.display.Javascript object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/markdown": [
       "<span style=\"color:#33cc33\">Correct:</span> \n",
       "\n",
       "That the training loss and validation loss stay fairly close is evidence that the model isn't just memorizing the training data, but rather learning general properties of the two classes. But, because this model converges at a loss greater than the VGG16 model, it's likely that it is underfitting some, and could benefit from some extra capacity.\n"
      ],
      "text/plain": [
       "Correct: \n",
       "\n",
       "That the training loss and validation loss stay fairly close is evidence that the model isn't just memorizing the training data, but rather learning general properties of the two classes. But, because this model converges at a loss greater than the VGG16 model, it's likely that it is underfitting some, and could benefit from some extra capacity."
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# View the solution (Run this code cell to receive credit!)\n",
    "q_4.check()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "554f4f55",
   "metadata": {
    "papermill": {
     "duration": 0.324563,
     "end_time": "2022-03-13T16:25:55.735441",
     "exception": false,
     "start_time": "2022-03-13T16:25:55.410878",
     "status": "completed"
    },
    "tags": []
   },
   "source": [
    "# Conclusion #\n",
    "\n",
    "In this first lesson, you learned the basics of **convolutional image classifiers**, that they consist of a **base** for extracting features from images, and a **head** which uses the features to decide the image's class. You also saw how to build a classifier with **transfer learning** on pretrained base. "
   ]
  },
  {
   "cell_type": "markdown",
   "id": "67027540",
   "metadata": {
    "papermill": {
     "duration": 0.325209,
     "end_time": "2022-03-13T16:25:56.386464",
     "exception": false,
     "start_time": "2022-03-13T16:25:56.061255",
     "status": "completed"
    },
    "tags": []
   },
   "source": [
    "# Keep Going #\n",
    "\n",
    "Move on to [**Lesson 2**](https://www.kaggle.com/ryanholbrook/convolution-and-relu) for a detailed look at how the base does this feature extraction. (It's really cool!)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "8eed55c5",
   "metadata": {
    "papermill": {
     "duration": 0.327289,
     "end_time": "2022-03-13T16:25:57.037224",
     "exception": false,
     "start_time": "2022-03-13T16:25:56.709935",
     "status": "completed"
    },
    "tags": []
   },
   "source": [
    "---\n",
    "\n",
    "\n",
    "\n",
    "\n",
    "*Have questions or comments? Visit the [course discussion forum](https://www.kaggle.com/learn/computer-vision/discussion) to chat with other learners.*"
   ]
  }
 ],
 "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": "3.7.12"
  },
  "papermill": {
   "default_parameters": {},
   "duration": 203.563258,
   "end_time": "2022-03-13T16:26:00.990853",
   "environment_variables": {},
   "exception": null,
   "input_path": "__notebook__.ipynb",
   "output_path": "__notebook__.ipynb",
   "parameters": {},
   "start_time": "2022-03-13T16:22:37.427595",
   "version": "2.3.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}