04-convolutional-neural-network.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "%reload_ext autoreload\n",
    "%autoreload 2\n",
    "%matplotlib inline\n",
    "\n",
    "from fastai.vision import *\n",
    "from fastai.metrics import *"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Convolutional Neural Network"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Nachdem die Klassifikation unserer Bilder mit einem einfachen Netz wahrscheinlich nicht sehr erfolgreich war, probieren wir nun ein CNN."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Erster Versuch wieder mit MNIST"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Wir starten unsere Versuche mit einem Convolutional Neural Network wieder mit dem MNIST-Datensatz - den handgeschriebenen Ziffern."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "MNIST_path = untar_data(URLs.MNIST)\n",
    "MNIST_path.ls()\n",
    "data = (ImageList.from_folder(MNIST_path)\n",
    "        .split_by_folder(train='training', valid='testing')\n",
    "        .label_from_folder()\n",
    "        .transform(size=28)\n",
    "        .databunch(bs=64))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Wir erstellen das Netzwerk diesmal nicht von Grund auf, sondern nutzen eine Standard-Architektur: ein kleines Residual Network mit 18 layers namens resnet18. Dieses Netzwerk erwartet drei Channel (RGB) als Input. Der Schalter `pretrained=False` sagt, dass wir unser Netzwerk selbst trainieren möchten und nicht auf einem vortrainierten Modell aufsetzen wollen."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn = cnn_learner(data, models.resnet18, pretrained=False, metrics=error_rate)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Wir trainieren das Netz mit den Standard-Einstellungen und freuen uns hoffentlich über eine Fehlerrate kleiner als 1%."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: left;\">\n",
       "      <th>epoch</th>\n",
       "      <th>train_loss</th>\n",
       "      <th>valid_loss</th>\n",
       "      <th>error_rate</th>\n",
       "      <th>time</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>0</td>\n",
       "      <td>0.141837</td>\n",
       "      <td>0.217335</td>\n",
       "      <td>0.058200</td>\n",
       "      <td>00:37</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>0.085851</td>\n",
       "      <td>0.048170</td>\n",
       "      <td>0.013400</td>\n",
       "      <td>00:37</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>2</td>\n",
       "      <td>0.042089</td>\n",
       "      <td>0.028651</td>\n",
       "      <td>0.008300</td>\n",
       "      <td>00:37</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>3</td>\n",
       "      <td>0.020895</td>\n",
       "      <td>0.023651</td>\n",
       "      <td>0.007200</td>\n",
       "      <td>00:37</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "learn.fit_one_cycle(4)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Wieder ein paar Beispiele für Fehlklassifikationen..."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 648x504 with 16 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "interp = ClassificationInterpretation.from_learner(learn)\n",
    "interp.plot_top_losses(16, figsize=(9,7))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Aufgaben und Ideen: eigene Bilder\n",
    "\n",
    "1. Erstelle ein Netz, das Deine Bilder verarbeiten kann.\n",
    "2. Trainiere das Netz.\n",
    "3. Wie sind die Ergebnisse?\n",
    "4. Was passiert wenn Du eine andere Architektur nutzt (z.B. resnet34, resnet50, vgg16_bn, densenet121)?\n",
    "6. Was passiert wenn Du noch ein paar Epochen länger trainierst?"
   ]
  }
 ],
 "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": "3.7.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}