Trees.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 218,
   "metadata": {},
   "outputs": [],
   "source": [
    "import torch\n",
    "from torch import nn\n",
    "from torch.autograd import Variable\n",
    "from torchvision import datasets, transforms\n",
    "import torchvision\n",
    "import matplotlib.pyplot as plt\n",
    "import numpy as np\n",
    "from torch import optim\n",
    "from sklearn.metrics import accuracy_score\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "bsz = 10\n",
    "\n",
    "train_loader = torch.utils.data.DataLoader(\n",
    "        datasets.MNIST('../data', train=True, download=True,\n",
    "                       transform=transforms.Compose([\n",
    "                           transforms.ToTensor(),\n",
    "                           transforms.Normalize((0.1307,), (0.3081,))\n",
    "                       ])),\n",
    "        batch_size=bsz, shuffle=True)\n",
    "test_loader = torch.utils.data.DataLoader(\n",
    "        datasets.MNIST('../data', train=False, transform=transforms.Compose([\n",
    "                           transforms.ToTensor(),\n",
    "                           transforms.Normalize((0.1307,), (0.3081,))\n",
    "                       ])),\n",
    "batch_size=bsz, shuffle=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "    7     6     8     0\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXQAAAB4CAYAAADi1gmcAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f274d77b2e8>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# functions to show an image\n",
    "def imshow(img):\n",
    "    img = img / 2 + 0.5     # unnormalize\n",
    "    npimg = img.numpy()\n",
    "    plt.imshow(np.transpose(npimg, (1, 2, 0)))\n",
    "\n",
    "# get some random training images\n",
    "dataiter = iter(train_loader)\n",
    "images, labels = dataiter.next()\n",
    "\n",
    "# show images\n",
    "imshow(torchvision.utils.make_grid(images))\n",
    "# print labels\n",
    "classes = list(range(10))\n",
    "print(' '.join('%5s' % classes[labels[j]] for j in range(4)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 244,
   "metadata": {},
   "outputs": [],
   "source": [
    "EPS = 1e-10\n",
    "\n",
    "class Leaf(nn.Module):\n",
    "    def __init__(self, i_size, o_size, h_size=128):\n",
    "        super(Leaf, self).__init__()\n",
    "        self.i2h = nn.Linear(i_size, h_size)\n",
    "        self.h2o = nn.Linear(h_size, o_size)\n",
    "        self.soft = nn.LogSoftmax(1)\n",
    "        self.relu = nn.ReLU()\n",
    "        self.is_leaf = True\n",
    "\n",
    "    def forward(self, features):\n",
    "        out = self.i2h(features)\n",
    "        out = self.relu(out)\n",
    "        out = self.h2o(out)\n",
    "        return self.soft(out)\n",
    "\n",
    "    def accum_probs(self, features, path_prob):\n",
    "        return [[path_prob, self.forward(features)]]\n",
    "\n",
    "    def calc_regularization(self, features, path_prob):\n",
    "        return 0\n",
    "\n",
    "class Node(nn.Module):\n",
    "    def __init__(self, i_size, o_size):\n",
    "        super(Node, self).__init__()\n",
    "        self.o_size = o_size\n",
    "        self.i_size = i_size\n",
    "        self.i2o = nn.Linear(i_size, 1)\n",
    "        self.sigmoid = nn.Sigmoid()\n",
    "        self.is_leaf = False\n",
    "    \n",
    "    def build_tree(self, depth):\n",
    "        if depth - 1 < 0:\n",
    "            raise ValueError(\"Depth must be greater than zero.\")\n",
    "        if depth - 1 > 0:\n",
    "            self.left = Node(self.i_size, self.o_size)\n",
    "            self.right = Node(self.i_size, self.o_size)\n",
    "            self.left.build_tree(depth - 1)\n",
    "            self.right.build_tree(depth - 1)\n",
    "        else:\n",
    "            self.left = Leaf(self.i_size, self.o_size)\n",
    "            self.right = Leaf(self.i_size, self.o_size)\n",
    "\n",
    "    def forward(self, features):\n",
    "        pr = self.prob_left(features)\n",
    "        return pr*self.left(features) + (1 - pr)*self.right(features)\n",
    "\n",
    "    def prob_left(self, features):\n",
    "        return self.sigmoid(self.i2o(features))\n",
    "\n",
    "    def accum_probs(self, features, path_prob):\n",
    "        res = []\n",
    "        p_l = self.sigmoid(self.i2o(features)).squeeze()\n",
    "        res_l = self.left.accum_probs(features, p_l*path_prob)\n",
    "        res_r = self.right.accum_probs(features, (1 - p_l)*path_prob)\n",
    "        res.extend(res_l)\n",
    "        res.extend(res_r)\n",
    "        return res\n",
    "\n",
    "    def calc_regularization(self, features, path_prob):\n",
    "        p_l = self.prob_left(features).squeeze()\n",
    "        alpha = (path_prob*p_l).sum()/(path_prob.sum())\n",
    "        C_here = -0.5*torch.log(alpha + EPS) - 0.5*torch.log(1 - alpha + EPS)\n",
    "        C = self.left.calc_regularization(features, p_l*path_prob) + \\\n",
    "                        self.right.calc_regularization(features, (1 - p_l)*path_prob)\n",
    "        C = C + C_here\n",
    "        return C\n",
    "        \n",
    "\n",
    "def tree_loss(path_probs, y_true, C, gamma):\n",
    "    loss = 0\n",
    "    criterion = nn.NLLLoss(reduce=False)\n",
    "    for p, pred in path_probs:\n",
    "        loss += (p*criterion(pred, y_true)).mean()\n",
    "    return loss.mean() + gamma*C\n",
    "\n",
    "def tree_logloss(path_probs, y_true):\n",
    "    \"\"\"\n",
    "    Original loss from paper\n",
    "    \"\"\"\n",
    "    loss = 0\n",
    "    criterion = nn.NLLLoss()\n",
    "    for p, pred in path_probs:\n",
    "        loss -= (p.squeeze()*criterion(pred, y_true)).mean()\n",
    "    return -torch.log(loss.mean())\n",
    "\n",
    "def train(model, batches_train, batches_val, n_epoch=5, gamma=0.1,\n",
    "          criterion=tree_loss, val_every=500, print_every=100):\n",
    "    model.train()\n",
    "    optimizer = optim.Adam(model.parameters())\n",
    "    all_losses = np.zeros(print_every)\n",
    "    plot_train = []\n",
    "    plot_val = []\n",
    "    for epoch in range(n_epoch):\n",
    "        print('Epoch: {}'.format(epoch))\n",
    "        for i, batch in enumerate(batches_train):\n",
    "            optimizer.zero_grad()\n",
    "            features, targets = batch\n",
    "            features = Variable(features.view(-1, 28*28))\n",
    "            targets = Variable(targets)\n",
    "            prb = model.accum_probs(features, Variable(torch.Tensor([1]*targets.shape[0])))\n",
    "            C = model.calc_regularization(features, Variable(torch.Tensor([1]*targets.shape[0])))\n",
    "            loss = tree_loss(prb, targets, C, gamma=gamma)\n",
    "            loss.backward()\n",
    "            optimizer.step()\n",
    "            plot_train += [loss.data[0]]\n",
    "            all_losses[(i + 1)%print_every] = loss.data[0]\n",
    "            if (i + 1) % print_every == 0:\n",
    "                print(all_losses.max(), all_losses.mean())\n",
    "            if (i + 1) % val_every == 0:\n",
    "                plot_val += [validate(model, batches_val)]\n",
    "    return plot_train, plot_val\n",
    "\n",
    "def validate(model, batches_val):\n",
    "    model.eval()\n",
    "    y_pred = []\n",
    "    y_true = []\n",
    "    for batch in batches_val:\n",
    "        features, targets = batch\n",
    "        y_true += targets.tolist()\n",
    "        y_pred += model(Variable(features.view(-1, 28*28))).topk(1)[1].squeeze().data.tolist()\n",
    "    model.train()\n",
    "    return accuracy_score(y_true, y_pred)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 249,
   "metadata": {},
   "outputs": [],
   "source": [
    "net = Node(28*28, 10)\n",
    "net.build_tree(2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 250,
   "metadata": {},
   "outputs": [],
   "source": [
    "net.build_tree(2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 251,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.0883"
      ]
     },
     "execution_count": 251,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "validate(net, test_loader)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 252,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch: 0\n",
      "2.3208134174346924 0.9850304496288299\n",
      "1.124450922012329 0.4424265518411994\n",
      "1.487255573272705 0.39970444157719615\n",
      "1.1348315477371216 0.3702938113361597\n",
      "1.466813564300537 0.3933159306133166\n",
      "1.6960644721984863 0.3138192646577954\n",
      "1.217405080795288 0.30514644775539634\n",
      "1.1556371450424194 0.29529811756685376\n",
      "1.191265344619751 0.22567772466689348\n",
      "1.8193292617797852 0.27397711247205736\n",
      "1.23622465133667 0.2687176687456667\n",
      "1.2653847932815552 0.22984152415767312\n",
      "1.8631157875061035 0.19994493345264344\n",
      "1.286694884300232 0.2698298747185618\n",
      "2.304396152496338 0.21439555362798274\n",
      "1.6031297445297241 0.24639692983357236\n",
      "1.064756155014038 0.22799287303350865\n",
      "1.0995135307312012 0.2082734259031713\n",
      "1.0551551580429077 0.20649951677769424\n",
      "1.0215766429901123 0.25772748090326786\n",
      "0.7689920663833618 0.17644611961208284\n",
      "1.0986156463623047 0.18222827931400387\n",
      "1.1006221771240234 0.16866874655243008\n",
      "0.8370348215103149 0.17149998996872454\n",
      "1.397258996963501 0.16925717217149214\n",
      "1.1604899168014526 0.20883429812733084\n",
      "0.8361403346061707 0.17829439171589911\n",
      "1.0696121454238892 0.15313711863942445\n",
      "0.8889804482460022 0.1469799780426547\n",
      "0.7911593317985535 0.11512288965983317\n",
      "1.3839755058288574 0.16378633158979938\n",
      "0.8525345325469971 0.1707738986320328\n",
      "1.0671029090881348 0.17591319049708545\n",
      "0.9335495233535767 0.16334078460931778\n",
      "0.7481372356414795 0.172690078986343\n",
      "1.143003225326538 0.13502014294732362\n",
      "0.6795305013656616 0.1392819927376695\n",
      "1.4145958423614502 0.15149038768431639\n",
      "0.991976797580719 0.1343320899049286\n",
      "1.0135029554367065 0.1405631335428916\n",
      "1.165534496307373 0.179615306679043\n",
      "1.126476764678955 0.14920794221223332\n",
      "1.3399792909622192 0.20933078387752174\n",
      "0.7174442410469055 0.13387076538288967\n",
      "0.9662591218948364 0.1236598418536596\n",
      "1.1802544593811035 0.16196205306681805\n",
      "1.2636271715164185 0.13309095038857777\n",
      "0.5890356302261353 0.10510191385401413\n",
      "0.9154562950134277 0.15658470588503406\n",
      "1.0138267278671265 0.13859826664207503\n",
      "0.8046207427978516 0.12664143761619925\n",
      "1.191956877708435 0.14940872391103766\n",
      "1.6261845827102661 0.1756876912410371\n",
      "1.0684453248977661 0.12140343839768321\n",
      "2.283851146697998 0.11533851633837912\n",
      "0.8713347315788269 0.14415901546075474\n",
      "0.9744457006454468 0.12392199873924255\n",
      "0.7241734862327576 0.11982066660420969\n",
      "1.3630993366241455 0.13718239458132303\n",
      "1.0163742303848267 0.12103703456756193\n"
     ]
    }
   ],
   "source": [
    "training, testing = train(net, train_loader, test_loader, gamma=0, n_epoch=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 253,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.9634"
      ]
     },
     "execution_count": 253,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "validate(net, test_loader)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 254,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[<matplotlib.lines.Line2D at 0x7f2712d93f98>]"
      ]
     },
     "execution_count": 254,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXcAAAD8CAYAAACMwORRAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f2712d55e48>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(training)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 255,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[<matplotlib.lines.Line2D at 0x7f2711bb7668>]"
      ]
     },
     "execution_count": 255,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAX0AAAD8CAYAAACb4nSYAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nhb6ISAJR6IuIJBCFvohIAlHoi4gkkP8PzuQWlh3Lt6cAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f2712d23ef0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(testing)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 256,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXQAAAB4CAYAAADi1gmcAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nEEIEUeiCIAghgih0QRCEEOH/AW5jAgkAZU6cAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f2711bff4a8>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "for data in test_loader:\n",
    "    images, labels = data\n",
    "    features = Variable(images.view(-1, 28*28))\n",
    "    break\n",
    "imshow(torchvision.utils.make_grid(images))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 257,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Variable containing:\n",
       "    7\n",
       "    8\n",
       "    1\n",
       "    3\n",
       "    2\n",
       "    0\n",
       "    6\n",
       "    7\n",
       "    6\n",
       "    2\n",
       "[torch.LongTensor of size 10x1]"
      ]
     },
     "execution_count": 257,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "net(features).topk(1)[1]"
   ]
  },
  {
   "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": "3.6.2"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}