{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Import erfolgreich\n"
     ]
    }
   ],
   "source": [
    "from tensorflow.keras.models import Sequential\n",
    "from tensorflow.keras.layers import Dense, GlobalAveragePooling2D\n",
    "from tensorflow.keras.applications import ResNet50\n",
    "from tensorflow.keras.callbacks import EarlyStopping, ModelCheckpoint, ReduceLROnPlateau\n",
    "import tensorflow as tf\n",
    "print(\"Import erfolgreich\")\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Found 361 images belonging to 3 classes.\n",
      "Found 89 images belonging to 3 classes.\n",
      "Aktueller Arbeitspfad: c:\\Users\\lukas\\Studium\\6. Semester\\KI Anwendungen\\PIcture\n",
      "Verzeichnisinhalt: ['.history', '.ipynb_checkpoints', 'app.py', 'main.ipynb', 'pokemon']\n"
     ]
    }
   ],
   "source": [
    "from tensorflow.keras.preprocessing.image import ImageDataGenerator\n",
    "from sklearn.model_selection import train_test_split\n",
    "import os\n",
    "\n",
    "base_dir = 'pokemon'\n",
    "\n",
    "image_size = (224, 224)\n",
    "batch_size = 32\n",
    "\n",
    "train_datagen = ImageDataGenerator(\n",
    "    rescale=1./255,\n",
    "    rotation_range=40,\n",
    "    width_shift_range=0.2,\n",
    "    height_shift_range=0.2,\n",
    "    shear_range=0.2,\n",
    "    zoom_range=0.2,\n",
    "    horizontal_flip=True,\n",
    "    fill_mode='nearest',\n",
    "    validation_split=0.2  \n",
    ")\n",
    "\n",
    "train_generator = train_datagen.flow_from_directory(\n",
    "    base_dir,\n",
    "    target_size=image_size,\n",
    "    batch_size=batch_size,\n",
    "    class_mode='categorical',\n",
    "    subset='training' \n",
    ")\n",
    "\n",
    "validation_generator = train_datagen.flow_from_directory(\n",
    "    base_dir,\n",
    "    target_size=image_size,\n",
    "    batch_size=batch_size,\n",
    "    class_mode='categorical',\n",
    "    subset='validation' \n",
    ")\n",
    "\n",
    "# Aktuellen Arbeitspfad drucken\n",
    "print(\"Aktueller Arbeitspfad:\", os.getcwd())\n",
    "\n",
    "# Inhalte im aktuellen Verzeichnis auflisten\n",
    "print(\"Verzeichnisinhalt:\", os.listdir())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "base_model = ResNet50(weights='imagenet', include_top=False, input_shape=(224, 224, 3))\n",
    "\n",
    "base_model.trainable = False\n",
    "\n",
    "model = Sequential([\n",
    "    base_model,\n",
    "    GlobalAveragePooling2D(),\n",
    "    Dense(512, activation='relu'),\n",
    "    Dense(3, activation='softmax')  \n",
    "])\n",
    "\n",
    "model.compile(optimizer='adam', loss='categorical_crossentropy', metrics=['accuracy'])\n",
    "\n",
    "early_stopping = EarlyStopping(monitor='val_loss', patience=5, restore_best_weights=True)\n",
    "model_checkpoint = ModelCheckpoint('best_model.keras', save_best_only=True)\n",
    "reduce_lr = ReduceLROnPlateau(monitor='val_loss', factor=0.2, patience=2, min_lr=1e-7)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/10\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "c:\\Users\\lukas\\anaconda3\\envs\\kia\\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",
      "  self._warn_if_super_not_called()\n",
      "c:\\Users\\lukas\\anaconda3\\envs\\kia\\lib\\site-packages\\PIL\\Image.py:1000: UserWarning: Palette images with Transparency expressed in bytes should be converted to RGBA images\n",
      "  warnings.warn(\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[1m12/12\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m82s\u001b[0m 5s/step - accuracy: 0.3422 - loss: 1.4997 - val_accuracy: 0.3258 - val_loss: 1.1893 - learning_rate: 0.0010\n",
      "Epoch 2/10\n",
      "\u001b[1m12/12\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m60s\u001b[0m 4s/step - accuracy: 0.3986 - loss: 1.1751 - val_accuracy: 0.4045 - val_loss: 1.1911 - learning_rate: 0.0010\n",
      "Epoch 3/10\n",
      "\u001b[1m12/12\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m59s\u001b[0m 4s/step - accuracy: 0.4265 - loss: 1.1501 - val_accuracy: 0.5393 - val_loss: 1.0268 - learning_rate: 0.0010\n",
      "Epoch 4/10\n",
      "\u001b[1m12/12\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m51s\u001b[0m 4s/step - accuracy: 0.5038 - loss: 1.0047 - val_accuracy: 0.6180 - val_loss: 0.9316 - learning_rate: 0.0010\n",
      "Epoch 5/10\n",
      "\u001b[1m12/12\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m47s\u001b[0m 3s/step - accuracy: 0.4520 - loss: 1.0198 - val_accuracy: 0.4831 - val_loss: 0.9292 - learning_rate: 0.0010\n",
      "Epoch 6/10\n",
      "\u001b[1m12/12\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m45s\u001b[0m 3s/step - accuracy: 0.5053 - loss: 0.9459 - val_accuracy: 0.6292 - val_loss: 0.9976 - learning_rate: 0.0010\n",
      "Epoch 7/10\n",
      "\u001b[1m12/12\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m46s\u001b[0m 3s/step - accuracy: 0.5592 - loss: 0.9549 - val_accuracy: 0.5618 - val_loss: 0.8820 - learning_rate: 0.0010\n",
      "Epoch 8/10\n",
      "\u001b[1m12/12\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m53s\u001b[0m 4s/step - accuracy: 0.4919 - loss: 0.9838 - val_accuracy: 0.5281 - val_loss: 0.8794 - learning_rate: 0.0010\n",
      "Epoch 9/10\n",
      "\u001b[1m12/12\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m60s\u001b[0m 4s/step - accuracy: 0.5190 - loss: 0.9214 - val_accuracy: 0.6180 - val_loss: 0.8400 - learning_rate: 0.0010\n",
      "Epoch 10/10\n",
      "\u001b[1m12/12\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m51s\u001b[0m 4s/step - accuracy: 0.5295 - loss: 0.9299 - val_accuracy: 0.5730 - val_loss: 0.8847 - learning_rate: 0.0010\n"
     ]
    }
   ],
   "source": [
    "history = model.fit(\n",
    "    train_generator,\n",
    "    epochs=10,\n",
    "    validation_data=validation_generator,\n",
    "    callbacks=[early_stopping, model_checkpoint, reduce_lr]\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 10/20\n",
      "\u001b[1m12/12\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m101s\u001b[0m 7s/step - accuracy: 0.4401 - loss: 3.5201 - val_accuracy: 0.4607 - val_loss: 0.9513 - learning_rate: 1.0000e-05\n",
      "Epoch 11/20\n",
      "\u001b[1m12/12\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m78s\u001b[0m 6s/step - accuracy: 0.5555 - loss: 2.0987 - val_accuracy: 0.4157 - val_loss: 1.0550 - learning_rate: 1.0000e-05\n",
      "Epoch 12/20\n",
      "\u001b[1m12/12\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m73s\u001b[0m 6s/step - accuracy: 0.4555 - loss: 1.7529 - val_accuracy: 0.4157 - val_loss: 1.2538 - learning_rate: 1.0000e-05\n",
      "Epoch 13/20\n",
      "\u001b[1m12/12\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m72s\u001b[0m 6s/step - accuracy: 0.5554 - loss: 1.0633 - val_accuracy: 0.4157 - val_loss: 1.4237 - learning_rate: 2.0000e-06\n",
      "Epoch 14/20\n",
      "\u001b[1m12/12\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m75s\u001b[0m 6s/step - accuracy: 0.5212 - loss: 1.2389 - val_accuracy: 0.4157 - val_loss: 1.7295 - learning_rate: 2.0000e-06\n"
     ]
    }
   ],
   "source": [
    "base_model.trainable = True\n",
    "fine_tune_at = 100\n",
    "\n",
    "# Set the earlier layers to not be trainable\n",
    "for layer in base_model.layers[:fine_tune_at]:\n",
    "    layer.trainable = False\n",
    "\n",
    "# Ensure you include a loss function here\n",
    "model.compile(optimizer=tf.keras.optimizers.Adam(1e-5), \n",
    "              loss='categorical_crossentropy',  # This should be the loss function you used initially\n",
    "              metrics=['accuracy'])\n",
    "\n",
    "fine_tune_epochs = 10\n",
    "total_epochs = history.epoch[-1] + fine_tune_epochs + 1\n",
    "\n",
    "history_fine = model.fit(\n",
    "    train_generator,\n",
    "    epochs=total_epochs,\n",
    "    initial_epoch=history.epoch[-1],\n",
    "    validation_data=validation_generator,\n",
    "    callbacks=[early_stopping, model_checkpoint, reduce_lr]\n",
    ")\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "model.save('pokemon_classifier_model.keras')\n",
    "\n"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "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": "3.9.19"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}