heuzef
/
jan24_cds_mushrooms
1
0
Forka 0
Codice Problemi 0 Pull Requests 0 Rilasci 0 Wiki Attività
Non puoi selezionare più di 25 argomenti Gli argomenti devono iniziare con una lettera o un numero, possono includere trattini ('-') e possono essere lunghi fino a 35 caratteri.
3 Commit
1 Ramo (Branch)
332 MiB
 
 
 
Ramo (Branch): master
Rami (Branch) Tag
${ item.name }
Crea branch ${ searchTerm }
da 'master'
${ noResults }
jan24_cds_mushrooms/notebooks/transfer_learning_ResNet50/ResNet50.ipynb

351 righe
21 KiB
Originale Permalink Blame Cronologia 

  1. {

  2.  "cells": [

  3.   {

  4.    "cell_type": "code",

  5.    "execution_count": 3,

  6.    "metadata": {},

  7.    "outputs": [

  8.     {

  9.      "name": "stdout",

  10.      "output_type": "stream",

  11.      "text": [

  12.       "Found 2199 images belonging to 2 classes.\n",

  13.       "Found 549 images belonging to 2 classes.\n",

  14.       "Training samples: 2199\n",

  15.       "Validation samples: 549\n"

  16.      ]

  17.     }

  18.    ],

  19.    "source": [

  20.     "import tensorflow as tf\n",

  21.     "from tensorflow.keras.preprocessing.image import ImageDataGenerator\n",

  22.     "\n",

  23.     "dataset_dir = 'C:/Users/vsavelev/GITHUB/DS_projet/jan24_cds_mushrooms/data'\n",

  24.     "\n",

  25.     "# Create ImageDataGenerator with validation split\n",

  26.     "datagen = ImageDataGenerator(rescale=1.0/255, validation_split=0.2)\n",

  27.     "\n",

  28.     "train_generator = datagen.flow_from_directory(\n",

  29.     "    dataset_dir,\n",

  30.     "    target_size=(224, 224),\n",

  31.     "    batch_size=32,\n",

  32.     "    class_mode='categorical',\n",

  33.     "    subset='training'  # Set as training data\n",

  34.     ")\n",

  35.     "\n",

  36.     "validation_generator = datagen.flow_from_directory(\n",

  37.     "    dataset_dir,\n",

  38.     "    target_size=(224, 224),\n",

  39.     "    batch_size=32,\n",

  40.     "    class_mode='categorical',\n",

  41.     "    subset='validation'  # Set as validation data\n",

  42.     ")\n",

  43.     "\n",

  44.     "print(f'Training samples: {train_generator.samples}')\n",

  45.     "print(f'Validation samples: {validation_generator.samples}')"

  46.    ]

  47.   },

  48.   {

  49.    "cell_type": "code",

  50.    "execution_count": 4,

  51.    "metadata": {},

  52.    "outputs": [

  53.     {

  54.      "data": {

  55.       "text/html": [

  56.        "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\">Model: \"sequential\"</span>\n",

  57.        "</pre>\n"

  58.       ],

  59.       "text/plain": [

  60.        "\u001b[1mModel: \"sequential\"\u001b[0m\n"

  61.       ]

  62.      },

  63.      "metadata": {},

  64.      "output_type": "display_data"

  65.     },

  66.     {

  67.      "data": {

  68.       "text/html": [

  69.        "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\">┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓\n",

  70.        "┃<span style=\"font-weight: bold\"> Layer (type)                    </span>┃<span style=\"font-weight: bold\"> Output Shape           </span>┃<span style=\"font-weight: bold\">       Param # </span>┃\n",

  71.        "┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩\n",

  72.        "│ resnet50 (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">Functional</span>)           │ ?                      │    <span style=\"color: #00af00; text-decoration-color: #00af00\">23,587,712</span> │\n",

  73.        "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",

  74.        "│ global_average_pooling2d        │ ?                      │   <span style=\"color: #00af00; text-decoration-color: #00af00\">0</span> (unbuilt) │\n",

  75.        "│ (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">GlobalAveragePooling2D</span>)        │                        │               │\n",

  76.        "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",

  77.        "│ dense (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">Dense</span>)                   │ ?                      │   <span style=\"color: #00af00; text-decoration-color: #00af00\">0</span> (unbuilt) │\n",

  78.        "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",

  79.        "│ dropout (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">Dropout</span>)               │ ?                      │   <span style=\"color: #00af00; text-decoration-color: #00af00\">0</span> (unbuilt) │\n",

  80.        "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",

  81.        "│ dense_1 (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">Dense</span>)                 │ ?                      │   <span style=\"color: #00af00; text-decoration-color: #00af00\">0</span> (unbuilt) │\n",

  82.        "└─────────────────────────────────┴────────────────────────┴───────────────┘\n",

  83.        "</pre>\n"

  84.       ],

  85.       "text/plain": [

  86.        "┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓\n",

  87.        "┃\u001b[1m \u001b[0m\u001b[1mLayer (type)                   \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1mOutput Shape          \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1m      Param #\u001b[0m\u001b[1m \u001b[0m┃\n",

  88.        "┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩\n",

  89.        "│ resnet50 (\u001b[38;5;33mFunctional\u001b[0m)           │ ?                      │    \u001b[38;5;34m23,587,712\u001b[0m │\n",

  90.        "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",

  91.        "│ global_average_pooling2d        │ ?                      │   \u001b[38;5;34m0\u001b[0m (unbuilt) │\n",

  92.        "│ (\u001b[38;5;33mGlobalAveragePooling2D\u001b[0m)        │                        │               │\n",

  93.        "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",

  94.        "│ dense (\u001b[38;5;33mDense\u001b[0m)                   │ ?                      │   \u001b[38;5;34m0\u001b[0m (unbuilt) │\n",

  95.        "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",

  96.        "│ dropout (\u001b[38;5;33mDropout\u001b[0m)               │ ?                      │   \u001b[38;5;34m0\u001b[0m (unbuilt) │\n",

  97.        "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",

  98.        "│ dense_1 (\u001b[38;5;33mDense\u001b[0m)                 │ ?                      │   \u001b[38;5;34m0\u001b[0m (unbuilt) │\n",

  99.        "└─────────────────────────────────┴────────────────────────┴───────────────┘\n"

  100.       ]

  101.      },

  102.      "metadata": {},

  103.      "output_type": "display_data"

  104.     },

  105.     {

  106.      "data": {

  107.       "text/html": [

  108.        "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\"> Total params: </span><span style=\"color: #00af00; text-decoration-color: #00af00\">23,587,712</span> (89.98 MB)\n",

  109.        "</pre>\n"

  110.       ],

  111.       "text/plain": [

  112.        "\u001b[1m Total params: \u001b[0m\u001b[38;5;34m23,587,712\u001b[0m (89.98 MB)\n"

  113.       ]

  114.      },

  115.      "metadata": {},

  116.      "output_type": "display_data"

  117.     },

  118.     {

  119.      "data": {

  120.       "text/html": [

  121.        "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\"> Trainable params: </span><span style=\"color: #00af00; text-decoration-color: #00af00\">0</span> (0.00 B)\n",

  122.        "</pre>\n"

  123.       ],

  124.       "text/plain": [

  125.        "\u001b[1m Trainable params: \u001b[0m\u001b[38;5;34m0\u001b[0m (0.00 B)\n"

  126.       ]

  127.      },

  128.      "metadata": {},

  129.      "output_type": "display_data"

  130.     },

  131.     {

  132.      "data": {

  133.       "text/html": [

  134.        "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\"> Non-trainable params: </span><span style=\"color: #00af00; text-decoration-color: #00af00\">23,587,712</span> (89.98 MB)\n",

  135.        "</pre>\n"

  136.       ],

  137.       "text/plain": [

  138.        "\u001b[1m Non-trainable params: \u001b[0m\u001b[38;5;34m23,587,712\u001b[0m (89.98 MB)\n"

  139.       ]

  140.      },

  141.      "metadata": {},

  142.      "output_type": "display_data"

  143.     }

  144.    ],

  145.    "source": [

  146.     "from tensorflow.keras.applications import ResNet50\n",

  147.     "from tensorflow.keras import layers, models\n",

  148.     "\n",

  149.     "# Load and Configure the Pre-trained ResNet50 Model\n",

  150.     "base_model = ResNet50(weights='imagenet', include_top=False, input_shape=(224, 224, 3))\n",

  151.     "\"\"\"\n",

  152.     "weights='imagenet': Loads the pre-trained weights from the ImageNet dataset.\n",

  153.     "include_top=False: Excludes the top fully-connected layers of the ResNet50 model, enabling you to add your own custom layers.\n",

  154.     "input_shape=(224, 224, 3): Specifies the input shape of the images (224x224 pixels, with 3 color channels - RGB).\n",

  155.     "\"\"\"\n",

  156.     "\n",

  157.     "# Freeze the base model (to freeze the pre-trained layers)\n",

  158.     "base_model.trainable = False\n",

  159.     "\n",

  160.     "# Add custom layers on top of the base model\n",

  161.     "model = models.Sequential([   #allows to stack layers linearly\n",

  162.     "    base_model,\n",

  163.     "    layers.GlobalAveragePooling2D(),\n",

  164.     "    layers.Dense(1024, activation='relu'),\n",

  165.     "    layers.Dropout(0.5),\n",

  166.     "    layers.Dense(train_generator.num_classes, activation='softmax')\n",

  167.     "])\n",

  168.     "\n",

  169.     "\"\"\"\n",

  170.     "GlobalAveragePooling2D(): Reduces each feature map to a single number by taking the average, \n",

  171.     "which helps to reduce the size of the model and prevent overfitting.\n",

  172.     "Dense(1024, activation='relu'): Adds a fully connected layer with 1024 units and ReLU activation function.\n",

  173.     "Dropout(0.5): Adds a dropout layer with a 50% dropout rate to prevent overfitting by randomly setting half of the input units \n",

  174.     "to 0 at each update during training.\n",

  175.     "Dense(train_generator.num_classes, activation='softmax'): \n",

  176.     "Adds the final output layer with units equal to the number of classes in your dataset, using the softmax activation function for multi-class classification.\n",

  177.     "\"\"\"\n",

  178.     "model.compile(optimizer=tf.keras.optimizers.Adam(),\n",

  179.     "              loss='categorical_crossentropy',\n",

  180.     "              metrics=['accuracy'])\n",

  181.     "\n",

  182.     "\"\"\"\n",

  183.     "optimizer=tf.keras.optimizers.Adam(): Uses the Adam optimizer, which is an adaptive learning rate optimization algorithm.\n",

  184.     "loss='categorical_crossentropy': Uses categorical cross-entropy as the loss function, suitable for multi-class classification.\n",

  185.     "metrics=['accuracy']: Tracks accuracy as the metric to evaluate the model's performance during training and testing.\n",

  186.     "\"\"\"\n",

  187.     "model.summary()\n",

  188.     "\n"

  189.    ]

  190.   },

  191.   {

  192.    "cell_type": "code",

  193.    "execution_count": 5,

  194.    "metadata": {},

  195.    "outputs": [

  196.     {

  197.      "name": "stdout",

  198.      "output_type": "stream",

  199.      "text": [

  200.       "Epoch 1/10\n"

  201.      ]

  202.     },

  203.     {

  204.      "name": "stderr",

  205.      "output_type": "stream",

  206.      "text": [

  207.       "c:\\Users\\vsavelev\\AppData\\Local\\anaconda3\\Lib\\site-packages\\keras\\src\\trainers\\data_adapters\\py_dataset_adapter.py:121: UserWarning: Your `PyDataset` class should call `super().__init__(**kwargs)` in its constructor. `**kwargs` can include `workers`, `use_multiprocessing`, `max_queue_size`. Do not pass these arguments to `fit()`, as they will be ignored.\n",

  208.       "  self._warn_if_super_not_called()\n"

  209.      ]

  210.     },

  211.     {

  212.      "name": "stdout",

  213.      "output_type": "stream",

  214.      "text": [

  215.       "\u001b[1m68/68\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m135s\u001b[0m 2s/step - accuracy: 0.8657 - loss: 0.5402 - val_accuracy: 0.8860 - val_loss: 0.3811\n",

  216.       "Epoch 2/10\n",

  217.       "\u001b[1m 1/68\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m1:36\u001b[0m 1s/step - accuracy: 0.8125 - loss: 0.5545"

  218.      ]

  219.     },

  220.     {

  221.      "name": "stderr",

  222.      "output_type": "stream",

  223.      "text": [

  224.       "c:\\Users\\vsavelev\\AppData\\Local\\anaconda3\\Lib\\contextlib.py:158: UserWarning: Your input ran out of data; interrupting training. Make sure that your dataset or generator can generate at least `steps_per_epoch * epochs` batches. You may need to use the `.repeat()` function when building your dataset.\n",

  225.       "  self.gen.throw(typ, value, traceback)\n"

  226.      ]

  227.     },

  228.     {

  229.      "name": "stdout",

  230.      "output_type": "stream",

  231.      "text": [

  232.       "\u001b[1m68/68\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 7ms/step - accuracy: 0.8125 - loss: 0.5545 - val_accuracy: 1.0000 - val_loss: 0.2601\n",

  233.       "Epoch 3/10\n",

  234.       "\u001b[1m68/68\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m147s\u001b[0m 2s/step - accuracy: 0.8812 - loss: 0.3830 - val_accuracy: 0.8860 - val_loss: 0.3604\n",

  235.       "Epoch 4/10\n",

  236.       "\u001b[1m68/68\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m3s\u001b[0m 7ms/step - accuracy: 0.9375 - loss: 0.2276 - val_accuracy: 1.0000 - val_loss: 0.1036\n",

  237.       "Epoch 5/10\n",

  238.       "\u001b[1m68/68\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m186s\u001b[0m 3s/step - accuracy: 0.8776 - loss: 0.3826 - val_accuracy: 0.8860 - val_loss: 0.3568\n",

  239.       "Epoch 6/10\n",

  240.       "\u001b[1m68/68\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m3s\u001b[0m 9ms/step - accuracy: 0.9688 - loss: 0.1941 - val_accuracy: 1.0000 - val_loss: 0.1264\n",

  241.       "Epoch 7/10\n",

  242.       "\u001b[1m68/68\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m199s\u001b[0m 3s/step - accuracy: 0.8844 - loss: 0.3665 - val_accuracy: 0.8879 - val_loss: 0.3532\n",

  243.       "Epoch 8/10\n",

  244.       "\u001b[1m68/68\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m3s\u001b[0m 10ms/step - accuracy: 0.9062 - loss: 0.2978 - val_accuracy: 0.8000 - val_loss: 0.5479\n",

  245.       "Epoch 9/10\n",

  246.       "\u001b[1m68/68\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m190s\u001b[0m 3s/step - accuracy: 0.8929 - loss: 0.3426 - val_accuracy: 0.8879 - val_loss: 0.3531\n",

  247.       "Epoch 10/10\n",

  248.       "\u001b[1m68/68\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m3s\u001b[0m 10ms/step - accuracy: 0.8125 - loss: 0.4996 - val_accuracy: 0.8000 - val_loss: 0.5142\n"

  249.      ]

  250.     }

  251.    ],

  252.    "source": [

  253.     "history = model.fit(\n",

  254.     "    train_generator,\n",

  255.     "    steps_per_epoch=train_generator.samples // train_generator.batch_size,\n",

  256.     "    validation_data=validation_generator,\n",

  257.     "    validation_steps=validation_generator.samples // validation_generator.batch_size,\n",

  258.     "    epochs=10\n",

  259.     ")\n",

  260.     "\n",

  261.     "#This specifies the number of complete passes through the training dataset. Here, the model will train for 10 epochs."

  262.    ]

  263.   },

  264.   {

  265.    "cell_type": "code",

  266.    "execution_count": 6,

  267.    "metadata": {},

  268.    "outputs": [

  269.     {

  270.      "name": "stdout",

  271.      "output_type": "stream",

  272.      "text": [

  273.       "Epoch 1/10\n",

  274.       "\u001b[1m68/68\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m330s\u001b[0m 4s/step - accuracy: 0.8855 - loss: 0.4046 - val_accuracy: 0.8879 - val_loss: 0.3791\n",

  275.       "Epoch 2/10\n",

  276.       "\u001b[1m68/68\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m5s\u001b[0m 10ms/step - accuracy: 0.8750 - loss: 0.3539 - val_accuracy: 0.8000 - val_loss: 0.5808\n",

  277.       "Epoch 3/10\n",

  278.       "\u001b[1m68/68\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m289s\u001b[0m 4s/step - accuracy: 0.8960 - loss: 0.2870 - val_accuracy: 0.8860 - val_loss: 0.3903\n",

  279.       "Epoch 4/10\n",

  280.       "\u001b[1m68/68\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 7ms/step - accuracy: 0.9688 - loss: 0.1397 - val_accuracy: 1.0000 - val_loss: 0.2601\n",

  281.       "Epoch 5/10\n",

  282.       "\u001b[1m68/68\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m283s\u001b[0m 4s/step - accuracy: 0.9162 - loss: 0.2052 - val_accuracy: 0.8787 - val_loss: 0.4098\n",

  283.       "Epoch 6/10\n",

  284.       "\u001b[1m68/68\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 9ms/step - accuracy: 0.8438 - loss: 0.3068 - val_accuracy: 0.6000 - val_loss: 0.7204\n",

  285.       "Epoch 7/10\n",

  286.       "\u001b[1m68/68\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m290s\u001b[0m 4s/step - accuracy: 0.9429 - loss: 0.1480 - val_accuracy: 0.8382 - val_loss: 0.4598\n",

  287.       "Epoch 8/10\n",

  288.       "\u001b[1m68/68\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 6ms/step - accuracy: 0.9062 - loss: 0.1594 - val_accuracy: 1.0000 - val_loss: 0.2145\n",

  289.       "Epoch 9/10\n",

  290.       "\u001b[1m68/68\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m299s\u001b[0m 4s/step - accuracy: 0.9658 - loss: 0.1009 - val_accuracy: 0.8548 - val_loss: 0.4892\n",

  291.       "Epoch 10/10\n",

  292.       "\u001b[1m68/68\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 6ms/step - accuracy: 1.0000 - loss: 0.0651 - val_accuracy: 0.8000 - val_loss: 0.9206\n"

  293.      ]

  294.     }

  295.    ],

  296.    "source": [

  297.     "# Unfreeze some layers\n",

  298.     "base_model.trainable = True\n",

  299.     "fine_tune_at = 100  # fine-tune from this layer onwards\n",

  300.     "\n",

  301.     "for layer in base_model.layers[:fine_tune_at]:\n",

  302.     "    layer.trainable = False\n",

  303.     "\n",

  304.     "model.compile(optimizer=tf.keras.optimizers.Adam(learning_rate=1e-5),\n",

  305.     "              loss='categorical_crossentropy', #The categorical cross-entropy loss function is used because this is a multi-class classification problem\n",

  306.     "              metrics=['accuracy'])\n",

  307.     "\n",

  308.     "\"\"\"\n",

  309.     "The Adam optimizer is used with a very small learning rate (1e-5). Fine-tuning typically \n",

  310.     "uses a smaller learning rate to prevent large updates to the weights, which could potentially destroy the learned features in the pre-trained model.\n",

  311.     "\"\"\"\n",

  312.     "history_fine = model.fit(\n",

  313.     "    train_generator,\n",

  314.     "    steps_per_epoch=train_generator.samples // train_generator.batch_size,\n",

  315.     "    validation_data=validation_generator,\n",

  316.     "    validation_steps=validation_generator.samples // validation_generator.batch_size,\n",

  317.     "    epochs=10\n",

  318.     ")\n"

  319.    ]

  320.   },

  321.   {

  322.    "cell_type": "code",

  323.    "execution_count": 7,

  324.    "metadata": {},

  325.    "outputs": [

  326.     {

  327.      "name": "stdout",

  328.      "output_type": "stream",

  329.      "text": [

  330.       "\u001b[1m18/18\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m36s\u001b[0m 2s/step - accuracy: 0.8559 - loss: 0.4854\n",

  331.       "Validation loss: 0.4950728416442871\n",

  332.       "Validation accuracy: 0.8561019897460938\n"

  333.      ]

  334.     }

  335.    ],

  336.    "source": [

  337.     "loss, accuracy = model.evaluate(validation_generator)\n",

  338.     "print(f'Validation loss: {loss}')\n",

  339.     "print(f'Validation accuracy: {accuracy}')\n"

  340.    ]

  341.   }

  342.  ],

  343.  "metadata": {

  344.   "language_info": {

  345.    "name": "python"

  346.   }

  347.  },

  348.  "nbformat": 4,

  349.  "nbformat_minor": 2

  350. }