Josh_Lesson1_Optimized.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Image Classification using CNN - streamlined w/Fastai"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Review: easy steps to train a world-class image classifier\n",
    "\n",
    "1. Enable data augmentation, and precompute=True\n",
    "1. Use `lr_find()` to find highest learning rate where loss is still clearly improving\n",
    "1. Train last layer from precomputed activations for 1-2 epochs\n",
    "1. Train last layer with data augmentation (i.e. precompute=False) for 2-3 epochs with cycle_len=1\n",
    "1. Unfreeze all layers\n",
    "1. Set earlier layers to 3x-10x lower learning rate than next higher layer\n",
    "1. Use Test Time Augmentation (TTA) for test predictions\n",
    "1. Use `lr_find()` again\n",
    "1. Train full network with cycle_mult=2 until over-fitting"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Put these at the top of every notebook, to get automatic reloading and inline plotting\n",
    "%reload_ext autoreload\n",
    "%autoreload 2\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [],
   "source": [
    "# This file contains all the main external libs we'll use\n",
    "from fastai.imports import *\n",
    "\n",
    "from fastai.transforms import *\n",
    "from fastai.conv_learner import *\n",
    "from fastai.model import *\n",
    "from fastai.dataset import *\n",
    "from fastai.sgdr import *\n",
    "from fastai.plots import *\n",
    "\n",
    "PATH = \"data/dogscats/\"\n",
    "\n",
    "sz=224\n",
    "\n",
    "arch=resnet34\n",
    "\n",
    "bs=28"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Uncomment the below if you need to reset your precomputed activations\n",
    "!rm -rf {PATH}tmp"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 1. Enable data augmentation  && Create Data Object"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "100%|██████████| 360/360 [00:59<00:00,  6.04it/s]\n",
      "100%|██████████| 32/32 [00:05<00:00,  5.90it/s]\n"
     ]
    }
   ],
   "source": [
    "# The transformation object\n",
    "tfms = tfms_from_model(resnet34, sz, aug_tfms=transforms_side_on, max_zoom=1.1)\n",
    "\n",
    "# The cleaned up object which holds our data\n",
    "data = ImageClassifierData.from_paths(PATH, tfms=tfms)\n",
    "\n",
    "# The Pretrained ConvLearner object\n",
    "learn = ConvLearner.pretrained(arch, data, precompute=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 2. Find highest learning rate where loss is still increasing"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "669e91533b004ce1926035a9c03d54d2",
       "version_major": 2,
       "version_minor": 0
      },
      "text/html": [
       "<p>Failed to display Jupyter Widget of type <code>HBox</code>.</p>\n",
       "<p>\n",
       "  If you're reading this message in the Jupyter Notebook or JupyterLab Notebook, it may mean\n",
       "  that the widgets JavaScript is still loading. If this message persists, it\n",
       "  likely means that the widgets JavaScript library is either not installed or\n",
       "  not enabled. See the <a href=\"https://ipywidgets.readthedocs.io/en/stable/user_install.html\">Jupyter\n",
       "  Widgets Documentation</a> for setup instructions.\n",
       "</p>\n",
       "<p>\n",
       "  If you're reading this message in another frontend (for example, a static\n",
       "  rendering on GitHub or <a href=\"https://nbviewer.jupyter.org/\">NBViewer</a>),\n",
       "  it may mean that your frontend doesn't currently support widgets.\n",
       "</p>\n"
      ],
      "text/plain": [
       "HBox(children=(IntProgress(value=0, description='Epoch', max=1), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      " 76%|███████▋  | 275/360 [00:05<00:01, 47.49it/s, loss=0.456] \n",
      "                                                             \r"
     ]
    }
   ],
   "source": [
    "lrf=learn.lr_find()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 3. Train last layer from pre-computed model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "0d349b8af4bb4085a872480e81b32145",
       "version_major": 2,
       "version_minor": 0
      },
      "text/html": [
       "<p>Failed to display Jupyter Widget of type <code>HBox</code>.</p>\n",
       "<p>\n",
       "  If you're reading this message in the Jupyter Notebook or JupyterLab Notebook, it may mean\n",
       "  that the widgets JavaScript is still loading. If this message persists, it\n",
       "  likely means that the widgets JavaScript library is either not installed or\n",
       "  not enabled. See the <a href=\"https://ipywidgets.readthedocs.io/en/stable/user_install.html\">Jupyter\n",
       "  Widgets Documentation</a> for setup instructions.\n",
       "</p>\n",
       "<p>\n",
       "  If you're reading this message in another frontend (for example, a static\n",
       "  rendering on GitHub or <a href=\"https://nbviewer.jupyter.org/\">NBViewer</a>),\n",
       "  it may mean that your frontend doesn't currently support widgets.\n",
       "</p>\n"
      ],
      "text/plain": [
       "HBox(children=(IntProgress(value=0, description='Epoch', max=1), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[ 0.       0.06501  0.0276   0.98975]                         \n",
      "\n"
     ]
    }
   ],
   "source": [
    "learn.fit(1e-2, 1)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 4. Train last layer w/Data Aug."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.precompute=False"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "8864a6d3308a4be6b795e17e9554819b",
       "version_major": 2,
       "version_minor": 0
      },
      "text/html": [
       "<p>Failed to display Jupyter Widget of type <code>HBox</code>.</p>\n",
       "<p>\n",
       "  If you're reading this message in the Jupyter Notebook or JupyterLab Notebook, it may mean\n",
       "  that the widgets JavaScript is still loading. If this message persists, it\n",
       "  likely means that the widgets JavaScript library is either not installed or\n",
       "  not enabled. See the <a href=\"https://ipywidgets.readthedocs.io/en/stable/user_install.html\">Jupyter\n",
       "  Widgets Documentation</a> for setup instructions.\n",
       "</p>\n",
       "<p>\n",
       "  If you're reading this message in another frontend (for example, a static\n",
       "  rendering on GitHub or <a href=\"https://nbviewer.jupyter.org/\">NBViewer</a>),\n",
       "  it may mean that your frontend doesn't currently support widgets.\n",
       "</p>\n"
      ],
      "text/plain": [
       "HBox(children=(IntProgress(value=0, description='Epoch', max=2), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[ 0.       0.0426   0.02729  0.98975]                         \n",
      "[ 1.       0.04244  0.02747  0.99072]                         \n",
      "\n"
     ]
    }
   ],
   "source": [
    "learn.fit(1e-2, 2, cycle_len=1)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 5. Unfreeze all layers"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.unfreeze()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 6. Set earlier layers to 3x-10x lower learning rate than next higher layer"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "219d868ac96d4efbb29b4560bd95b18b",
       "version_major": 2,
       "version_minor": 0
      },
      "text/html": [
       "<p>Failed to display Jupyter Widget of type <code>HBox</code>.</p>\n",
       "<p>\n",
       "  If you're reading this message in the Jupyter Notebook or JupyterLab Notebook, it may mean\n",
       "  that the widgets JavaScript is still loading. If this message persists, it\n",
       "  likely means that the widgets JavaScript library is either not installed or\n",
       "  not enabled. See the <a href=\"https://ipywidgets.readthedocs.io/en/stable/user_install.html\">Jupyter\n",
       "  Widgets Documentation</a> for setup instructions.\n",
       "</p>\n",
       "<p>\n",
       "  If you're reading this message in another frontend (for example, a static\n",
       "  rendering on GitHub or <a href=\"https://nbviewer.jupyter.org/\">NBViewer</a>),\n",
       "  it may mean that your frontend doesn't currently support widgets.\n",
       "</p>\n"
      ],
      "text/plain": [
       "HBox(children=(IntProgress(value=0, description='Epoch', max=7), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[ 0.       0.04892  0.02743  0.98682]                         \n",
      "[ 1.       0.03891  0.02143  0.98828]                         \n",
      "[ 2.       0.02655  0.02041  0.99121]                         \n",
      "[ 3.       0.02709  0.01984  0.99121]                         \n",
      "[ 4.       0.02545  0.01652  0.99365]                         \n",
      "[ 5.       0.02179  0.01746  0.99316]                         \n",
      " 16%|█▋        | 59/360 [00:29<02:32,  1.97it/s, loss=0.0162]"
     ]
    }
   ],
   "source": [
    "lr=np.array([1e-4,1e-3,1e-2]) #differential learning rates\n",
    "\n",
    "learn.fit(lr, 3, cycle_len=1, cycle_mult=2) #cycle_mult=2 needed for resnet34"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.sched.plot()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.save('224_all')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.load('224_all')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 7. Use Test Time Augmentation for Training Test Predictions"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Check predictions at this point\n",
    "log_preds,y = learn.TTA()\n",
    "accuracy(log_preds,y)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 8. Find the optimal learning rate again (lr_find)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 9. Train full network with cycle_mult=2 until over-fitting"
   ]
  },
  {
   "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": "3.6.4"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}