{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Import der benötigten Bibliotheken\n",
    "import numpy as np\n",
    "import tensorflow as tf\n",
    "from tensorflow.keras.applications import ResNet50\n",
    "from tensorflow.keras.preprocessing.image import ImageDataGenerator\n",
    "from tensorflow.keras.layers import Dense, GlobalAveragePooling2D, Dropout\n",
    "from tensorflow.keras.models import Model\n",
    "from tensorflow.keras.optimizers import Adam\n",
    "from tensorflow.keras.callbacks import ModelCheckpoint, EarlyStopping, ReduceLROnPlateau"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Vorbereitung der Daten\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Found 78 images belonging to 6 classes.\n",
      "Found 16 images belonging to 6 classes.\n"
     ]
    }
   ],
   "source": [
    "# Daten-Vorbereitung\n",
    "base_dir = 'C:\\Daten\\Studium Wirtschaftsinformatik\\Semester 6 TZ\\KI-Anwendungen\\Übungen\\Übung2\\Abschluss\\Datensätze\\Strohhüte'  # Pfad zum übergeordneten Ordner, der die Klassenordner enthält\n",
    "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  # Behalte die Aufteilung für Training und Validation bei\n",
    ")\n",
    "train_generator = datagen.flow_from_directory(\n",
    "    base_dir,\n",
    "    target_size=(224, 224),  # Assuming using ResNet input dimensions\n",
    "    batch_size=32,  # Adjust according to your system capability\n",
    "    class_mode='categorical',\n",
    "    subset='training'  # Use the 'subset' argument for splitting\n",
    ")\n",
    "\n",
    "validation_generator = datagen.flow_from_directory(\n",
    "    base_dir,\n",
    "    target_size=(224, 224),\n",
    "    batch_size=32,\n",
    "    class_mode='categorical',\n",
    "    subset='validation'\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Modell Setup"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Modell-Setup\n",
    "base_model = ResNet50(weights='imagenet', include_top=False, input_shape=(224, 224, 3))\n",
    "base_model.trainable = False  # Zuerst wird das Basismodell eingefroren\n",
    "\n",
    "x = GlobalAveragePooling2D()(base_model.output)\n",
    "x = Dense(1024, activation='relu')(x)\n",
    "x = Dropout(0.5)(x)  # Dropout hinzugefügt, um Overfitting zu reduzieren\n",
    "predictions = Dense(3, activation='softmax')(x)\n",
    "\n",
    "model = Model(inputs=base_model.input, outputs=predictions)\n",
    "model.compile(optimizer=Adam(learning_rate=0.001), loss='categorical_crossentropy', metrics=['accuracy'])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Training des Models"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/20\n"
     ]
    },
    {
     "ename": "ValueError",
     "evalue": "Arguments `target` and `output` must have the same shape. Received: target.shape=(None, 6), output.shape=(None, 3)",
     "output_type": "error",
     "traceback": [
      "\u001b[1;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[1;31mValueError\u001b[0m                                Traceback (most recent call last)",
      "Cell \u001b[1;32mIn[4], line 2\u001b[0m\n\u001b[0;32m      1\u001b[0m \u001b[38;5;66;03m# Trainieren des Modells\u001b[39;00m\n\u001b[1;32m----> 2\u001b[0m history \u001b[38;5;241m=\u001b[39m \u001b[43mmodel\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mfit\u001b[49m\u001b[43m(\u001b[49m\n\u001b[0;32m      3\u001b[0m \u001b[43m    \u001b[49m\u001b[43mtrain_generator\u001b[49m\u001b[43m,\u001b[49m\n\u001b[0;32m      4\u001b[0m \u001b[43m    \u001b[49m\u001b[43msteps_per_epoch\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mtrain_generator\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43msamples\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m/\u001b[39;49m\u001b[38;5;241;43m/\u001b[39;49m\u001b[43m \u001b[49m\u001b[43mtrain_generator\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mbatch_size\u001b[49m\u001b[43m,\u001b[49m\n\u001b[0;32m      5\u001b[0m \u001b[43m    \u001b[49m\u001b[43mvalidation_data\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mvalidation_generator\u001b[49m\u001b[43m,\u001b[49m\n\u001b[0;32m      6\u001b[0m \u001b[43m    \u001b[49m\u001b[43mvalidation_steps\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mvalidation_generator\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43msamples\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m/\u001b[39;49m\u001b[38;5;241;43m/\u001b[39;49m\u001b[43m \u001b[49m\u001b[43mvalidation_generator\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mbatch_size\u001b[49m\u001b[43m,\u001b[49m\n\u001b[0;32m      7\u001b[0m \u001b[43m    \u001b[49m\u001b[43mepochs\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[38;5;241;43m20\u001b[39;49m\u001b[43m,\u001b[49m\n\u001b[0;32m      8\u001b[0m \u001b[43m    \u001b[49m\u001b[43mcallbacks\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43m[\u001b[49m\n\u001b[0;32m      9\u001b[0m \u001b[43m        \u001b[49m\u001b[43mModelCheckpoint\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;124;43m'\u001b[39;49m\u001b[38;5;124;43mbest_model.keras\u001b[39;49m\u001b[38;5;124;43m'\u001b[39;49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43msave_best_only\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[38;5;28;43;01mTrue\u001b[39;49;00m\u001b[43m)\u001b[49m\u001b[43m,\u001b[49m\n\u001b[0;32m     10\u001b[0m \u001b[43m        \u001b[49m\u001b[43mEarlyStopping\u001b[49m\u001b[43m(\u001b[49m\u001b[43mmonitor\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[38;5;124;43m'\u001b[39;49m\u001b[38;5;124;43mval_loss\u001b[39;49m\u001b[38;5;124;43m'\u001b[39;49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mpatience\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[38;5;241;43m5\u001b[39;49m\u001b[43m)\u001b[49m\u001b[43m,\u001b[49m\n\u001b[0;32m     11\u001b[0m \u001b[43m        \u001b[49m\u001b[43mReduceLROnPlateau\u001b[49m\u001b[43m(\u001b[49m\u001b[43mmonitor\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[38;5;124;43m'\u001b[39;49m\u001b[38;5;124;43mval_loss\u001b[39;49m\u001b[38;5;124;43m'\u001b[39;49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mfactor\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[38;5;241;43m0.2\u001b[39;49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mpatience\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[38;5;241;43m2\u001b[39;49m\u001b[43m)\u001b[49m\n\u001b[0;32m     12\u001b[0m \u001b[43m    \u001b[49m\u001b[43m]\u001b[49m\n\u001b[0;32m     13\u001b[0m \u001b[43m)\u001b[49m\n",
      "File \u001b[1;32mc:\\Users\\Jeremy Kuwegu\\anaconda3\\envs\\kia\\lib\\site-packages\\keras\\src\\utils\\traceback_utils.py:122\u001b[0m, in \u001b[0;36mfilter_traceback.<locals>.error_handler\u001b[1;34m(*args, **kwargs)\u001b[0m\n\u001b[0;32m    119\u001b[0m     filtered_tb \u001b[38;5;241m=\u001b[39m _process_traceback_frames(e\u001b[38;5;241m.\u001b[39m__traceback__)\n\u001b[0;32m    120\u001b[0m     \u001b[38;5;66;03m# To get the full stack trace, call:\u001b[39;00m\n\u001b[0;32m    121\u001b[0m     \u001b[38;5;66;03m# `keras.config.disable_traceback_filtering()`\u001b[39;00m\n\u001b[1;32m--> 122\u001b[0m     \u001b[38;5;28;01mraise\u001b[39;00m e\u001b[38;5;241m.\u001b[39mwith_traceback(filtered_tb) \u001b[38;5;28;01mfrom\u001b[39;00m \u001b[38;5;28;01mNone\u001b[39;00m\n\u001b[0;32m    123\u001b[0m \u001b[38;5;28;01mfinally\u001b[39;00m:\n\u001b[0;32m    124\u001b[0m     \u001b[38;5;28;01mdel\u001b[39;00m filtered_tb\n",
      "File \u001b[1;32mc:\\Users\\Jeremy Kuwegu\\anaconda3\\envs\\kia\\lib\\site-packages\\keras\\src\\backend\\tensorflow\\nn.py:554\u001b[0m, in \u001b[0;36mcategorical_crossentropy\u001b[1;34m(target, output, from_logits, axis)\u001b[0m\n\u001b[0;32m    552\u001b[0m \u001b[38;5;28;01mfor\u001b[39;00m e1, e2 \u001b[38;5;129;01min\u001b[39;00m \u001b[38;5;28mzip\u001b[39m(target\u001b[38;5;241m.\u001b[39mshape, output\u001b[38;5;241m.\u001b[39mshape):\n\u001b[0;32m    553\u001b[0m     \u001b[38;5;28;01mif\u001b[39;00m e1 \u001b[38;5;129;01mis\u001b[39;00m \u001b[38;5;129;01mnot\u001b[39;00m \u001b[38;5;28;01mNone\u001b[39;00m \u001b[38;5;129;01mand\u001b[39;00m e2 \u001b[38;5;129;01mis\u001b[39;00m \u001b[38;5;129;01mnot\u001b[39;00m \u001b[38;5;28;01mNone\u001b[39;00m \u001b[38;5;129;01mand\u001b[39;00m e1 \u001b[38;5;241m!=\u001b[39m e2:\n\u001b[1;32m--> 554\u001b[0m         \u001b[38;5;28;01mraise\u001b[39;00m \u001b[38;5;167;01mValueError\u001b[39;00m(\n\u001b[0;32m    555\u001b[0m             \u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mArguments `target` and `output` must have the same shape. \u001b[39m\u001b[38;5;124m\"\u001b[39m\n\u001b[0;32m    556\u001b[0m             \u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mReceived: \u001b[39m\u001b[38;5;124m\"\u001b[39m\n\u001b[0;32m    557\u001b[0m             \u001b[38;5;124mf\u001b[39m\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mtarget.shape=\u001b[39m\u001b[38;5;132;01m{\u001b[39;00mtarget\u001b[38;5;241m.\u001b[39mshape\u001b[38;5;132;01m}\u001b[39;00m\u001b[38;5;124m, output.shape=\u001b[39m\u001b[38;5;132;01m{\u001b[39;00moutput\u001b[38;5;241m.\u001b[39mshape\u001b[38;5;132;01m}\u001b[39;00m\u001b[38;5;124m\"\u001b[39m\n\u001b[0;32m    558\u001b[0m         )\n\u001b[0;32m    560\u001b[0m output, from_logits \u001b[38;5;241m=\u001b[39m _get_logits(\n\u001b[0;32m    561\u001b[0m     output, from_logits, \u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mSoftmax\u001b[39m\u001b[38;5;124m\"\u001b[39m, \u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mcategorical_crossentropy\u001b[39m\u001b[38;5;124m\"\u001b[39m\n\u001b[0;32m    562\u001b[0m )\n\u001b[0;32m    563\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m from_logits:\n",
      "\u001b[1;31mValueError\u001b[0m: Arguments `target` and `output` must have the same shape. Received: target.shape=(None, 6), output.shape=(None, 3)"
     ]
    }
   ],
   "source": [
    "# Trainieren des Modells\n",
    "history = model.fit(\n",
    "    train_generator,\n",
    "    steps_per_epoch=train_generator.samples // train_generator.batch_size,\n",
    "    validation_data=validation_generator,\n",
    "    validation_steps=validation_generator.samples // validation_generator.batch_size,\n",
    "    epochs=20,\n",
    "    callbacks=[\n",
    "        ModelCheckpoint('best_model.keras', save_best_only=True),\n",
    "        EarlyStopping(monitor='val_loss', patience=5),\n",
    "        ReduceLROnPlateau(monitor='val_loss', factor=0.2, patience=2)\n",
    "    ]\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Fine Tuning des Modells"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/10\n",
      "\u001b[1m9/9\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m148s\u001b[0m 12s/step - accuracy: 0.7070 - loss: 1.0460 - val_accuracy: 0.6094 - val_loss: 0.9729\n",
      "Epoch 2/10\n",
      "\u001b[1m9/9\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 135ms/step - accuracy: 1.0000 - loss: 0.0769 - val_accuracy: 0.5714 - val_loss: 1.0434\n",
      "Epoch 3/10\n",
      "\u001b[1m9/9\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m100s\u001b[0m 10s/step - accuracy: 0.9677 - loss: 0.1108 - val_accuracy: 0.5469 - val_loss: 0.9639\n",
      "Epoch 4/10\n",
      "\u001b[1m9/9\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m10s\u001b[0m 111ms/step - accuracy: 1.0000 - loss: 0.0381 - val_accuracy: 0.7143 - val_loss: 0.9019\n",
      "Epoch 5/10\n",
      "\u001b[1m9/9\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m96s\u001b[0m 10s/step - accuracy: 0.9992 - loss: 0.0220 - val_accuracy: 0.2969 - val_loss: 1.1206\n",
      "Epoch 6/10\n",
      "\u001b[1m9/9\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m12s\u001b[0m 164ms/step - accuracy: 1.0000 - loss: 0.0226 - val_accuracy: 0.1429 - val_loss: 1.1233\n",
      "Epoch 7/10\n",
      "\u001b[1m9/9\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m97s\u001b[0m 10s/step - accuracy: 1.0000 - loss: 0.0062 - val_accuracy: 0.1719 - val_loss: 1.4363\n",
      "Epoch 8/10\n",
      "\u001b[1m9/9\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m11s\u001b[0m 103ms/step - accuracy: 0.9688 - loss: 0.0287 - val_accuracy: 0.1429 - val_loss: 1.4406\n",
      "Epoch 9/10\n",
      "\u001b[1m9/9\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m98s\u001b[0m 10s/step - accuracy: 0.9907 - loss: 0.0160 - val_accuracy: 0.2344 - val_loss: 1.4151\n",
      "Epoch 10/10\n",
      "\u001b[1m9/9\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m11s\u001b[0m 101ms/step - accuracy: 1.0000 - loss: 0.0038 - val_accuracy: 0.0000e+00 - val_loss: 1.6847\n"
     ]
    }
   ],
   "source": [
    "# Fine-Tuning des Modells\n",
    "for layer in base_model.layers:\n",
    "    layer.trainable = True\n",
    "\n",
    "model.compile(optimizer=Adam(learning_rate=0.0001), loss='categorical_crossentropy', metrics=['accuracy'])\n",
    "history_fine = model.fit(\n",
    "    train_generator,\n",
    "    steps_per_epoch=train_generator.samples // train_generator.batch_size,\n",
    "    validation_data=validation_generator,\n",
    "    validation_steps=validation_generator.samples // validation_generator.batch_size,\n",
    "    epochs=10\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Bewertung und Ergebnisse"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[1m2/2\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m7s\u001b[0m 3s/step - accuracy: 0.2708 - loss: 1.4508\n",
      "Performance vor dem Fine-Tuning: 0.640625\n",
      "Performance nach dem Fine-Tuning: 0.0\n"
     ]
    }
   ],
   "source": [
    "# Ergebnisse bewerten\n",
    "eval_result = model.evaluate(validation_generator, steps=validation_generator.samples // validation_generator.batch_size)\n",
    "print(f'Performance vor dem Fine-Tuning: {history.history[\"val_accuracy\"][-1]}')\n",
    "print(f'Performance nach dem Fine-Tuning: {history_fine.history[\"val_accuracy\"][-1]}')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "WARNING:absl:You are saving your model as an HDF5 file via `model.save()` or `keras.saving.save_model(model)`. This file format is considered legacy. We recommend using instead the native Keras format, e.g. `model.save('my_model.keras')` or `keras.saving.save_model(model, 'my_model.keras')`. \n"
     ]
    }
   ],
   "source": [
    "model.save('mein_modell.h5') "
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "kia",
   "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
}