plants2.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Kaggle Plant seedling Identificaton"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "%reload_ext autoreload\n",
    "%autoreload 2\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "from fastai.imports import *\n",
    "from fastai.torch_imports import *\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 *"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "import os as os\n",
    "from os import *\n",
    "import random"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "PATH = \"data/plants/\"\n",
    "arch = resnet50\n",
    "bs = 64"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['sample_submission.csv', 'test', 'valid', 'subm', 'models', 'train', 'tmp']"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "os.listdir(PATH)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "classes = !ls {PATH}/train"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['Black-grass',\n",
       " 'Charlock',\n",
       " 'Cleavers',\n",
       " 'Common Chickweed',\n",
       " 'Common wheat',\n",
       " 'Fat Hen',\n",
       " 'Loose Silky-bent',\n",
       " 'Maize',\n",
       " 'Scentless Mayweed',\n",
       " 'Shepherds Purse',\n",
       " 'Small-flowered Cranesbill',\n",
       " 'Sugar beet']"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "classes"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Creating Validation Set"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "files = os.listdir(f'{PATH}valid/Black-grass')[:5]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['2ee4dad8c.png', '1276dffba.png']"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "files"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "def move_to_valid(files, class_dir):\n",
    "    for i in range(len(files)):\n",
    "        !mv \"{PATH}/train/{class_dir}/{files[i]}\" \"{PATH}/valid/{class_dir}\"\n",
    "        "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "for classname in classes:\n",
    "    os.makedirs(f'{PATH}/valid/{classname}', exist_ok = True)\n",
    "    files = !ls \"{PATH}/train/{classname}\"\n",
    "    random.shuffle(files)\n",
    "    n_valid_files = int(len(files)*0.1)\n",
    "    valid_files = [files[n] for n in range(n_valid_files)]\n",
    "    move_to_valid(valid_files, classname)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f4fb59a9748>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "img = plt.imread(f'{PATH}valid/Black-grass/{files[0]}')\n",
    "plt.imshow(img);"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(531, 531, 3)"
      ]
     },
     "execution_count": 44,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "img.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [],
   "source": [
    "for classname in classes:\n",
    "    files = os.listdir(f'{PATH}train/{classname}')\n",
    "    size_d = {k: PIL.Image.open(f'{PATH}train/{classname}/' + k).size for k in files}\n",
    "    "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [],
   "source": [
    "row_sz, col_sz = list(zip(*size_d.values()))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [],
   "source": [
    "row_sz = np.array(row_sz); col_sz = np.array(col_sz)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([303, 297, 682, 731, 481])"
      ]
     },
     "execution_count": 30,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "row_sz[:5]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([303, 297, 682, 731, 481])"
      ]
     },
     "execution_count": 31,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "col_sz[:5]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(array([58., 30., 80., 78., 30., 27., 13., 12.,  8.,  2.]),\n",
       " array([  49. ,  215.6,  382.2,  548.8,  715.4,  882. , 1048.6, 1215.2, 1381.8, 1548.4, 1715. ]),\n",
       " <a list of 10 Patch objects>)"
      ]
     },
     "execution_count": 32,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fc92fa99a20>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.hist(row_sz)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(array([58., 30., 80., 78., 30., 27., 14., 12.,  7.,  2.]),\n",
       " array([  49. ,  215.6,  382.2,  548.8,  715.4,  882. , 1048.6, 1215.2, 1381.8, 1548.4, 1715. ]),\n",
       " <a list of 10 Patch objects>)"
      ]
     },
     "execution_count": 33,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fc92abd9cf8>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.hist(col_sz)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 72,
   "metadata": {},
   "outputs": [],
   "source": [
    "files = os.listdir(f'{PATH}test/')\n",
    "size_d = {k: PIL.Image.open(f'{PATH}test/' + k).size for k in files}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 73,
   "metadata": {},
   "outputs": [],
   "source": [
    "row_sz, col_sz = list(zip(*size_d.values()))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 74,
   "metadata": {},
   "outputs": [],
   "source": [
    "row_sz = np.array(row_sz); col_sz = np.array(col_sz)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 75,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(array([98., 86., 84., 89., 73., 74., 83., 76., 65., 66.]),\n",
       " array([200. , 214.9, 229.8, 244.7, 259.6, 274.5, 289.4, 304.3, 319.2, 334.1, 349. ]),\n",
       " <a list of 10 Patch objects>)"
      ]
     },
     "execution_count": 75,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fc92817c518>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.hist(row_sz)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 76,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(array([98., 86., 84., 89., 73., 74., 83., 76., 65., 66.]),\n",
       " array([200. , 214.9, 229.8, 244.7, 259.6, 274.5, 289.4, 304.3, 319.2, 334.1, 349. ]),\n",
       " <a list of 10 Patch objects>)"
      ]
     },
     "execution_count": 76,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fc928020860>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.hist(col_sz)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "def get_data(sz, bs):\n",
    "    tfms = tfms_from_model(arch,sz, aug_tfms = transforms_side_on, max_zoom = 1.1)\n",
    "    data = ImageClassifierData.from_paths(PATH, bs=bs, tfms=tfms, test_name = 'test', num_workers=4)\n",
    "    return data"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Beginning with smaller pictures. This aids quick inital training, and prevents overfitting"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "sz = 128\n",
    "bs = 64"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "data = get_data(sz, bs)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "100%|██████████| 74/74 [00:19<00:00,  3.77it/s]\n",
      "100%|██████████| 1/1 [00:00<00:00,  4.06it/s]\n"
     ]
    }
   ],
   "source": [
    "learn = ConvLearner.pretrained(arch, data, precompute = True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Finding an optimal learning rate\n",
    "Experimenting with an exponentially growing learning rate. Plotting loss vs learning rate. Seeing at where the loss stops decreasing."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "f996ef9f968d4a08ac22e6f53d724ea4",
       "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": [
      " 61%|██████    | 45/74 [00:01<00:00, 40.09it/s, loss=11]  \n",
      "                                                        \r"
     ]
    }
   ],
   "source": [
    "lrf=learn.lr_find()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.sched.plot_lr()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 101,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fc90b43c6d8>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "learn.sched.plot()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Initial Traing\n",
    "training the last layer until we are near overfitting"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 103,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "14e651c8b5ae45ffa429ba4e1a250882",
       "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": [
      "epoch      trn_loss   val_loss   accuracy                  \n",
      "    0      0.770648   0.588945   0.824373  \n",
      "\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[0.5889454, 0.8243727617793613]"
      ]
     },
     "execution_count": 103,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "learn.fit(.1,1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 104,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "9c9234b0a5234e3d95f2548d1010bc14",
       "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": [
      "epoch      trn_loss   val_loss   accuracy                  \n",
      "    0      0.543364   0.498234   0.854055  \n",
      "\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[0.49823442, 0.8540546562936571]"
      ]
     },
     "execution_count": 104,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "learn.fit(.1,1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.precompute = False"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Unfreezing all the layers. Using differential learning rates, as we probably want to tweak the final layers, farther from the pretrained weights than the intial layers."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.unfreeze()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [],
   "source": [
    "lr = np.array([1e-3, 1e-2, 1e-1])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 110,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "f44c9a76865540e89c40603ba0ee2e74",
       "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": [
      "epoch      trn_loss   val_loss   accuracy                  \n",
      "    0      0.531517   0.232281   0.904402  \n",
      "\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[0.23228067, 0.9044018785158793]"
      ]
     },
     "execution_count": 110,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "learn.fit(lr, 1, cycle_len = 1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 113,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.save('128')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 114,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "0d3e9dd3a5774e6b94466da7660289e1",
       "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": [
      "epoch      trn_loss   val_loss   accuracy                  \n",
      "    0      0.253917   0.183475   0.932236  \n",
      "\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[0.18347462, 0.9322356647915311]"
      ]
     },
     "execution_count": 114,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "learn.fit(lr, 1, cycle_len = 1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 117,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "3279c1a365df4175ba7b293c9fcdf565",
       "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=3), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch      trn_loss   val_loss   accuracy                  \n",
      "    0      0.210482   0.148638   0.946125  \n",
      "    1      0.202127   0.152386   0.944388                  \n",
      "    2      0.150931   0.114867   0.958277                  \n",
      "\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[0.114866905, 0.9582773314581977]"
      ]
     },
     "execution_count": 117,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "learn.fit(lr, 2, cycle_len = 1, cycle_mult = 2)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Stochastit Gradient descent with restarts. This jumps out of local optima for a more general fit"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 118,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fc90b347080>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "learn.sched.plot_lr()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 119,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "77c9684bb4294a1f9fb84a759cf29607",
       "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=3), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch      trn_loss   val_loss   accuracy                  \n",
      "    0      0.14166    0.151169   0.947861  \n",
      "    1      0.116485   0.102146   0.958221                  \n",
      "    2      0.084983   0.097523   0.958277                   \n",
      "\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[0.09752311, 0.9582773314581977]"
      ]
     },
     "execution_count": 119,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "learn.fit(lr, 1, cycle_len = 3)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 124,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "2aee5e7ed6ed46d6badcec67c67dbed1",
       "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=4), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "  5%|▍         | 3/66 [00:05<01:57,  1.87s/it, loss=0.0743]\n",
      "  6%|▌         | 4/66 [00:07<01:52,  1.81s/it, loss=0.0639]"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Exception in thread Thread-86:\n",
      "Traceback (most recent call last):\n",
      "  File \"/home/paperspace/anaconda3/envs/fastai/lib/python3.6/threading.py\", line 916, in _bootstrap_inner\n",
      "    self.run()\n",
      "  File \"/home/paperspace/anaconda3/envs/fastai/lib/python3.6/site-packages/tqdm/_tqdm.py\", line 144, in run\n",
      "    for instance in self.tqdm_cls._instances:\n",
      "  File \"/home/paperspace/anaconda3/envs/fastai/lib/python3.6/_weakrefset.py\", line 60, in __iter__\n",
      "    for itemref in self.data:\n",
      "RuntimeError: Set changed size during iteration\n",
      "\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch      trn_loss   val_loss   accuracy                   \n",
      "    0      0.084073   0.14562    0.951333  \n",
      "    1      0.073789   0.125953   0.956597                   \n",
      "    2      0.048128   0.113907   0.954861                   \n",
      "    3      0.037071   0.112343   0.96875                    \n",
      "\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[0.11234309, 0.96875]"
      ]
     },
     "execution_count": 124,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "learn.fit(lr, 1, cycle_len = 4)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Sequential(\n",
       "  (0): Conv2d (3, 64, kernel_size=(7, 7), stride=(2, 2), padding=(3, 3), bias=False)\n",
       "  (1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
       "  (2): ReLU(inplace)\n",
       "  (3): MaxPool2d(kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), dilation=(1, 1))\n",
       "  (4): Sequential(\n",
       "    (0): Bottleneck(\n",
       "      (conv1): Conv2d (64, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
       "      (conv2): Conv2d (64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
       "      (conv3): Conv2d (64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
       "      (relu): ReLU(inplace)\n",
       "      (downsample): Sequential(\n",
       "        (0): Conv2d (64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
       "      )\n",
       "    )\n",
       "    (1): Bottleneck(\n",
       "      (conv1): Conv2d (256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
       "      (conv2): Conv2d (64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
       "      (conv3): Conv2d (64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
       "      (relu): ReLU(inplace)\n",
       "    )\n",
       "    (2): Bottleneck(\n",
       "      (conv1): Conv2d (256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
       "      (conv2): Conv2d (64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
       "      (conv3): Conv2d (64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
       "      (relu): ReLU(inplace)\n",
       "    )\n",
       "  )\n",
       "  (5): Sequential(\n",
       "    (0): Bottleneck(\n",
       "      (conv1): Conv2d (256, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
       "      (conv2): Conv2d (128, 128, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)\n",
       "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
       "      (conv3): Conv2d (128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
       "      (relu): ReLU(inplace)\n",
       "      (downsample): Sequential(\n",
       "        (0): Conv2d (256, 512, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
       "        (1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
       "      )\n",
       "    )\n",
       "    (1): Bottleneck(\n",
       "      (conv1): Conv2d (512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
       "      (conv2): Conv2d (128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
       "      (conv3): Conv2d (128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
       "      (relu): ReLU(inplace)\n",
       "    )\n",
       "    (2): Bottleneck(\n",
       "      (conv1): Conv2d (512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
       "      (conv2): Conv2d (128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
       "      (conv3): Conv2d (128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
       "      (relu): ReLU(inplace)\n",
       "    )\n",
       "    (3): Bottleneck(\n",
       "      (conv1): Conv2d (512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
       "      (conv2): Conv2d (128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
       "      (conv3): Conv2d (128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
       "      (relu): ReLU(inplace)\n",
       "    )\n",
       "  )\n",
       "  (6): Sequential(\n",
       "    (0): Bottleneck(\n",
       "      (conv1): Conv2d (512, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
       "      (conv2): Conv2d (256, 256, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)\n",
       "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
       "      (conv3): Conv2d (256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
       "      (relu): ReLU(inplace)\n",
       "      (downsample): Sequential(\n",
       "        (0): Conv2d (512, 1024, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
       "        (1): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
       "      )\n",
       "    )\n",
       "    (1): Bottleneck(\n",
       "      (conv1): Conv2d (1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
       "      (conv2): Conv2d (256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
       "      (conv3): Conv2d (256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
       "      (relu): ReLU(inplace)\n",
       "    )\n",
       "    (2): Bottleneck(\n",
       "      (conv1): Conv2d (1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
       "      (conv2): Conv2d (256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
       "      (conv3): Conv2d (256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
       "      (relu): ReLU(inplace)\n",
       "    )\n",
       "    (3): Bottleneck(\n",
       "      (conv1): Conv2d (1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
       "      (conv2): Conv2d (256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
       "      (conv3): Conv2d (256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
       "      (relu): ReLU(inplace)\n",
       "    )\n",
       "    (4): Bottleneck(\n",
       "      (conv1): Conv2d (1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
       "      (conv2): Conv2d (256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
       "      (conv3): Conv2d (256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
       "      (relu): ReLU(inplace)\n",
       "    )\n",
       "    (5): Bottleneck(\n",
       "      (conv1): Conv2d (1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
       "      (conv2): Conv2d (256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
       "      (conv3): Conv2d (256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
       "      (relu): ReLU(inplace)\n",
       "    )\n",
       "  )\n",
       "  (7): Sequential(\n",
       "    (0): Bottleneck(\n",
       "      (conv1): Conv2d (1024, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
       "      (conv2): Conv2d (512, 512, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)\n",
       "      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
       "      (conv3): Conv2d (512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True)\n",
       "      (relu): ReLU(inplace)\n",
       "      (downsample): Sequential(\n",
       "        (0): Conv2d (1024, 2048, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
       "        (1): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True)\n",
       "      )\n",
       "    )\n",
       "    (1): Bottleneck(\n",
       "      (conv1): Conv2d (2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
       "      (conv2): Conv2d (512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
       "      (conv3): Conv2d (512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True)\n",
       "      (relu): ReLU(inplace)\n",
       "    )\n",
       "    (2): Bottleneck(\n",
       "      (conv1): Conv2d (2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
       "      (conv2): Conv2d (512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
       "      (conv3): Conv2d (512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True)\n",
       "      (relu): ReLU(inplace)\n",
       "    )\n",
       "  )\n",
       "  (8): AdaptiveConcatPool2d(\n",
       "    (ap): AdaptiveAvgPool2d(output_size=(1, 1))\n",
       "    (mp): AdaptiveMaxPool2d(output_size=(1, 1))\n",
       "  )\n",
       "  (9): Flatten(\n",
       "  )\n",
       "  (10): BatchNorm1d(4096, eps=1e-05, momentum=0.1, affine=True)\n",
       "  (11): Dropout(p=0.25)\n",
       "  (12): Linear(in_features=4096, out_features=512)\n",
       "  (13): ReLU()\n",
       "  (14): BatchNorm1d(512, eps=1e-05, momentum=0.1, affine=True)\n",
       "  (15): Dropout(p=0.5)\n",
       "  (16): Linear(in_features=512, out_features=12)\n",
       "  (17): LogSoftmax()\n",
       ")"
      ]
     },
     "execution_count": 25,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "learn"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 126,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "37ca6e3f4b97428ea3ba81ca5b12e221",
       "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=6), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch      trn_loss   val_loss   accuracy                   \n",
      "    0      0.071247   0.208471   0.935652  \n",
      "    1      0.095994   0.139113   0.958333                   \n",
      "    2      0.07657    0.143133   0.956597                   \n",
      "    3      0.052554   0.117813   0.968694                   \n",
      "    4      0.025257   0.110808   0.972222                   \n",
      "    5      0.01792    0.112377   0.970486                   \n",
      "\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[0.1123773, 0.9704861111111112]"
      ]
     },
     "execution_count": 126,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "learn.fit(lr, 1, cycle_len = 6)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Resizing\n",
    "Now resizeing to our final size. This size is about the average size in the test set. We will repeat the same process as before"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [],
   "source": [
    "sz = 256\n",
    "data = get_data(256, bs)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.set_data(get_data(sz,bs))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 64,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.freeze()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 71,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "f9d85c03562540cc831b99017c260982",
       "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": [
      "epoch      trn_loss   val_loss   accuracy                   \n",
      "    0      0.023298   0.008699   0.996875  \n",
      "\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[0.008699147, 0.996875]"
      ]
     },
     "execution_count": 71,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "learn.fit(lr, 1, cycle_len = 1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 74,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.save('256')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.load('256')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Now using all of our data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 95,
   "metadata": {},
   "outputs": [],
   "source": [
    "def move_to_train(files, class_dir):\n",
    "    for i in range(len(files)):\n",
    "        !mv \"{PATH}/valid/{class_dir}/{files[i]}\" \"{PATH}/train/{class_dir}\"\n",
    "        "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "for classname in classes:\n",
    "    files = !ls \"{PATH}/valid/{classname}\"\n",
    "    random.shuffle(files)\n",
    "    n_train_files = int(len(files)-2)\n",
    "    train_files = [files[n] for n in range(n_train_files)]\n",
    "    move_to_train(train_files, classname)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [],
   "source": [
    "data = get_data(256, bs)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.set_data(get_data(sz,bs))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "4c26b43a4f2f40b9a7cc19cdae42c990",
       "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": [
      "epoch      trn_loss   val_loss   accuracy                   \n",
      "    0      0.024479   0.001427   1.0       \n",
      "    1      0.025623   0.001338   1.0                        \n",
      "\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[0.0013381243, 1.0]"
      ]
     },
     "execution_count": 26,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "learn.fit(lr, 2, cycle_len = 1)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Submitting Test time averaged Predictions\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.save('256_all_data')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.load('256_all_data')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Underfitting, so I lowered dropout"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.ps = 0.1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "2d6d5344d27f42fba1165d806a50f5f3",
       "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=3), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch      trn_loss   val_loss   accuracy                    \n",
      "    0      0.022596   0.000773   1.0       \n",
      "    1      0.024605   0.000674   1.0                        \n",
      "    2      0.020949   0.000658   1.0                        \n",
      "\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[0.00065847236, 1.0]"
      ]
     },
     "execution_count": 35,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "learn.fit(lr, 1, cycle_len = 3)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.save('256_all_data_0.1')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "                                             \r"
     ]
    }
   ],
   "source": [
    "log_preds, y = learn.TTA(is_test = True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [],
   "source": [
    "probs = np.mean(np.exp(log_preds),0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(794, 12)"
      ]
     },
     "execution_count": 38,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "probs.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [],
   "source": [
    "preds = np.argmax(probs, axis = 1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(794,)"
      ]
     },
     "execution_count": 40,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "preds.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([ 4, 10, 10,  7, 10])"
      ]
     },
     "execution_count": 41,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "preds[:5]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [],
   "source": [
    "pred_classes = []\n",
    "for p in preds:\n",
    "    pred_classes.append(classes[p])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<bound method NDFrame.head of               file     species\n",
       "0    0021e90e4.png  Sugar beet\n",
       "1    003d61042.png  Sugar beet\n",
       "2    007b3da8b.png  Sugar beet\n",
       "3    0086a6340.png  Sugar beet\n",
       "4    00c47e980.png  Sugar beet\n",
       "5    00d090cde.png  Sugar beet\n",
       "6    00ef713a8.png  Sugar beet\n",
       "7    01291174f.png  Sugar beet\n",
       "8    026716f9b.png  Sugar beet\n",
       "9    02cfeb38d.png  Sugar beet\n",
       "10   03566743d.png  Sugar beet\n",
       "11   03a2ee656.png  Sugar beet\n",
       "12   03e322a29.png  Sugar beet\n",
       "13   03ef36742.png  Sugar beet\n",
       "14   043449b0b.png  Sugar beet\n",
       "15   0437393b1.png  Sugar beet\n",
       "16   04814f36d.png  Sugar beet\n",
       "17   05341a8a6.png  Sugar beet\n",
       "18   060450d79.png  Sugar beet\n",
       "19   060f1dc84.png  Sugar beet\n",
       "20   0625f063b.png  Sugar beet\n",
       "21   063363305.png  Sugar beet\n",
       "22   06d12f6fa.png  Sugar beet\n",
       "23   071cb3ece.png  Sugar beet\n",
       "24   0751c0bbc.png  Sugar beet\n",
       "25   07e62f903.png  Sugar beet\n",
       "26   085974290.png  Sugar beet\n",
       "27   0885e7690.png  Sugar beet\n",
       "28   089ad62a7.png  Sugar beet\n",
       "29   08d591441.png  Sugar beet\n",
       "..             ...         ...\n",
       "764  f593c9cf0.png  Sugar beet\n",
       "765  f66ae4070.png  Sugar beet\n",
       "766  f6d250856.png  Sugar beet\n",
       "767  f8318faf1.png  Sugar beet\n",
       "768  f85ed9b6d.png  Sugar beet\n",
       "769  f9b6bfb00.png  Sugar beet\n",
       "770  f9ea23fb5.png  Sugar beet\n",
       "771  f9f35cbd4.png  Sugar beet\n",
       "772  fa5fd1384.png  Sugar beet\n",
       "773  fa9f3a8f9.png  Sugar beet\n",
       "774  fadc6adbc.png  Sugar beet\n",
       "775  fb022edf9.png  Sugar beet\n",
       "776  fba8fc78a.png  Sugar beet\n",
       "777  fbf88b6be.png  Sugar beet\n",
       "778  fc3e58836.png  Sugar beet\n",
       "779  fc6f686fb.png  Sugar beet\n",
       "780  fd253a74e.png  Sugar beet\n",
       "781  fd87b36ae.png  Sugar beet\n",
       "782  fd925f542.png  Sugar beet\n",
       "783  fda0b5c38.png  Sugar beet\n",
       "784  fda39e16f.png  Sugar beet\n",
       "785  fdea6b119.png  Sugar beet\n",
       "786  fe29629fb.png  Sugar beet\n",
       "787  fe9e87b78.png  Sugar beet\n",
       "788  fea1d13d6.png  Sugar beet\n",
       "789  fea355851.png  Sugar beet\n",
       "790  fea3da57c.png  Sugar beet\n",
       "791  fef2ade8c.png  Sugar beet\n",
       "792  ff65bc002.png  Sugar beet\n",
       "793  ffc6f8527.png  Sugar beet\n",
       "\n",
       "[794 rows x 2 columns]>"
      ]
     },
     "execution_count": 43,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_test = pd.read_csv(f'{PATH}sample_submission.csv')\n",
    "df_test.head"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {},
   "outputs": [],
   "source": [
    "df = pd.DataFrame(pred_classes)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {},
   "outputs": [],
   "source": [
    "df.insert(0,'file', [o[5:] for o in data.test_ds.fnames])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<bound method NDFrame.head of               file                          0\n",
       "0    79e5ea8fa.png               Common wheat\n",
       "1    3a909ead8.png  Small-flowered Cranesbill\n",
       "2    599691cd9.png  Small-flowered Cranesbill\n",
       "3    5ca2687a4.png                      Maize\n",
       "4    3dd52bd2a.png  Small-flowered Cranesbill\n",
       "5    1541bdb2e.png                      Maize\n",
       "6    39d569be4.png           Loose Silky-bent\n",
       "7    3d65168c2.png           Common Chickweed\n",
       "8    995c7ab1e.png               Common wheat\n",
       "9    99569b224.png               Common wheat\n",
       "10   8d6acbe9b.png           Common Chickweed\n",
       "11   aee6fa3df.png           Loose Silky-bent\n",
       "12   5b63dcc21.png           Common Chickweed\n",
       "13   219fd68d5.png          Scentless Mayweed\n",
       "14   69d1669f8.png                   Charlock\n",
       "15   82b5f4d33.png            Shepherds Purse\n",
       "16   35ebe165c.png                Black-grass\n",
       "17   aad8375e0.png            Shepherds Purse\n",
       "18   1376f3b63.png           Loose Silky-bent\n",
       "19   728eabae1.png                   Charlock\n",
       "20   0c51bf229.png                      Maize\n",
       "21   cf3a8b2fd.png           Loose Silky-bent\n",
       "22   60f0bc617.png  Small-flowered Cranesbill\n",
       "23   4e18ab737.png                   Cleavers\n",
       "24   3fbd0fc6a.png           Common Chickweed\n",
       "25   cf46d09c5.png           Common Chickweed\n",
       "26   558aa7deb.png           Loose Silky-bent\n",
       "27   b03397525.png           Loose Silky-bent\n",
       "28   1bf9b94a6.png          Scentless Mayweed\n",
       "29   2a667e099.png                    Fat Hen\n",
       "..             ...                        ...\n",
       "764  809eb0b82.png                   Charlock\n",
       "765  19e58cc5e.png           Common Chickweed\n",
       "766  e3f50adfc.png                   Cleavers\n",
       "767  afcf6abd5.png  Small-flowered Cranesbill\n",
       "768  b0acaff4a.png                Black-grass\n",
       "769  c7051c902.png           Common Chickweed\n",
       "770  06d12f6fa.png            Shepherds Purse\n",
       "771  ede6b84b4.png                   Cleavers\n",
       "772  19b51843a.png           Common Chickweed\n",
       "773  258b1183c.png                   Cleavers\n",
       "774  5e6a237f2.png                 Sugar beet\n",
       "775  e5297b675.png           Loose Silky-bent\n",
       "776  8d6991365.png                Black-grass\n",
       "777  099b961ec.png            Shepherds Purse\n",
       "778  a254d71f6.png                Black-grass\n",
       "779  d350a25fa.png           Common Chickweed\n",
       "780  116b136de.png                 Sugar beet\n",
       "781  39b740f7e.png                Black-grass\n",
       "782  c832e4302.png                    Fat Hen\n",
       "783  406ecb5c5.png                   Charlock\n",
       "784  fe29629fb.png          Scentless Mayweed\n",
       "785  16fd2e01a.png  Small-flowered Cranesbill\n",
       "786  aa7d098d1.png           Common Chickweed\n",
       "787  88d8a4508.png                      Maize\n",
       "788  fea355851.png           Loose Silky-bent\n",
       "789  b9062c1c8.png                 Sugar beet\n",
       "790  446f7da01.png           Loose Silky-bent\n",
       "791  a8388a37f.png                    Fat Hen\n",
       "792  00ef713a8.png           Common Chickweed\n",
       "793  0437393b1.png                    Fat Hen\n",
       "\n",
       "[794 rows x 2 columns]>"
      ]
     },
     "execution_count": 46,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df.head"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {},
   "outputs": [],
   "source": [
    "df.columns = ['file', 'species']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {},
   "outputs": [],
   "source": [
    "SUBM = f'{PATH}subm/'\n",
    "os.makedirs(SUBM, exist_ok = True)\n",
    "df.to_csv(f'{SUBM}subm.gz', compression = 'gzip', index = False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<a href='data/plants/subm/subm.gz' target='_blank'>data/plants/subm/subm.gz</a><br>"
      ],
      "text/plain": [
       "/home/paperspace/fastai/courses/dl1/data/plants/subm/subm.gz"
      ]
     },
     "execution_count": 49,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "FileLink(f'{SUBM}subm.gz')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Kaggle results: F1 Score 0.97984\n",
    "85 on the leaderboard out of 543\n"
   ]
  },
  {
   "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
}