6_Overfitting_and_Underfitting.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Todo\n",
    "\n",
    "- Define overfitting and underfitting \n",
    "- signal vs noise\n",
    "- learning curves\n",
    "- model capacity\n",
    "- early stopping"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "from tensorflow.keras.callbacks import EarlyStopping\n",
    "\n",
    "early_stopping = EarlyStopping(\n",
    "    min_delta=0.001, # minimium amount of change to count as an improvement\n",
    "    patience=20, # how many epochs to wait before stopping\n",
    "    restore_best_weights=True,\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "from tensorflow import keras\n",
    "from tensorflow.keras import layers, callbacks\n",
    "\n",
    "early_stopping = callbacks.EarlyStopping(\n",
    "    min_delta=0.001, # minimium amount of change to count as an improvement\n",
    "    patience=20, # how many epochs to wait before stopping\n",
    "    restore_best_weights=True,\n",
    ")\n",
    "\n",
    "model = keras.Sequential([\n",
    "    layers.Dense(512, activation='relu', input_shape=[11]),\n",
    "    layers.Dense(512, activation='relu'),\n",
    "    layers.Dense(512, activation='relu'),\n",
    "    layers.Dense(1),\n",
    "])\n",
    "model.compile(\n",
    "    optimizer='adam',\n",
    "    loss='mae',\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "## Use previous datasets to test the function below"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "# history = model.fit(\n",
    "#     X_train, y_train,\n",
    "#     validation_data=(X_valid, y_valid),\n",
    "#     batch_size=256,\n",
    "#     epochs=500,\n",
    "#     callbacks=[early_stopping], # put your callbacks in a list\n",
    "#     verbose=0,  # turn off training log\n",
    "# )\n",
    "\n",
    "# history_df = pd.DataFrame(history.history)\n",
    "# history_df.loc[:, ['loss', 'val_loss']].plot();\n",
    "# print(\"Minimum validation loss: {}\".format(history_df['val_loss'].min()))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "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": "3.8.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}