{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "%reload_ext autoreload\n",
    "%autoreload 2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from nb_005 import *"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# STL-10"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Basic data aug"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "PATH = Path('data/stl10')\n",
    "\n",
    "data_mean, data_std = map(tensor, ([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]))\n",
    "data_norm,data_denorm = normalize_funcs(data_mean,data_std)\n",
    "\n",
    "train_ds = ImageDataset.from_folder(PATH/'train')\n",
    "valid_ds = ImageDataset.from_folder(PATH/'valid')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "x=valid_ds[0][0]\n",
    "x.show()\n",
    "x.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "size=96\n",
    "tfms = get_transforms(do_flip=True, max_rotate=10, max_lighting=0.2, max_warp=0.15, max_zoom=1.2)\n",
    "# tfms = get_transforms(do_flip=True, max_rotate=10, max_lighting=0.2)\n",
    "tds = transform_datasets(train_ds, valid_ds, tfms, size=size)#, padding_mode='zeros')\n",
    "data = DataBunch.create(*tds, bs=32, num_workers=8, tfms=data_norm)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "(x,y) = next(iter(data.valid_dl))\n",
    "\n",
    "_,axs = plt.subplots(4,4,figsize=(12,12))\n",
    "for i,ax in enumerate(axs.flatten()): show_image(data_denorm(x[i].cpu()), ax)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "(x,y) = next(iter(data.train_dl))\n",
    "\n",
    "_,axs = plt.subplots(4,4,figsize=(12,12))\n",
    "for i,ax in enumerate(axs.flatten()): show_image(data_denorm(x[i].cpu()), ax)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "_,axs = plt.subplots(4,4,figsize=(12,12))\n",
    "for i,ax in enumerate(axs.flat): tds[0][1][0].show(ax)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Train"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from torchvision.models import resnet18, resnet34, resnet50\n",
    "arch = resnet50"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "lr = 2e-3"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn = ConvLearner(data, arch, 2 , wd=1e-2) #, train_bn=False #, callback_fns=[BnFreeze]\n",
    "#                     , opt_fn=partial(optim.SGD, momentum=0.9))\n",
    "learn.metrics = [accuracy]\n",
    "learn.split(lambda m: (m[0][6], m[1]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "lr_find(learn)\n",
    "learn.recorder.plot()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.fit_one_cycle(6, slice(lr))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.save('0')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Gradual unfreezing"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.load('0')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.unfreeze()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "lr = 2e-4"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.fit_one_cycle(6, slice(lr/100,lr), pct_start=0.05)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.save('1')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.recorder.plot_losses()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Fin"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import pandas as pd\n",
    "csv = pd.read_csv(PATH/'default.csv')\n",
    "is_valid = csv['2']=='valid'\n",
    "valid_df,train_df = csv[is_valid],csv[~is_valid]\n",
    "len(valid_df),len(train_df)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "len(valid_ds)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_fns,train_lbls,valid_fns,valid_lbls = map(np.array,\n",
    "    (train_df['0'],train_df['1'],valid_df['0'],valid_df['1']))\n",
    "\n",
    "train_fns = [PATH/o for o in train_fns]\n",
    "valid_fns = [PATH/o for o in valid_fns]\n",
    "\n",
    "train_ds = ImageDataset(train_fns,train_lbls)\n",
    "valid_ds = ImageDataset(valid_fns,valid_lbls, classes=train_ds.classes)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}