MLPScratch.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "from mxnet import autograd, nd, gluon\n",
    "from mxnet.gluon.data.vision import transforms"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "def get_dataloader(batch_size):\n",
    "    transformer = transforms.Compose([\n",
    "        transforms.ToTensor()\n",
    "    ])\n",
    "    train = gluon.data.vision.datasets.FashionMNIST(train=True)\n",
    "    train = train.transform_first(transformer)\n",
    "    train_iter = gluon.data.DataLoader(train, batch_size=batch_size, shuffle=True, num_workers=4)\n",
    "    test = gluon.data.vision.datasets.FashionMNIST(train=False)\n",
    "    test = test.transform_first(transformer)\n",
    "    test_iter = gluon.data.DataLoader(test, batch_size=batch_size, shuffle=False, num_workers=4)\n",
    "    return train_iter, test_iter"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "def show_images(X, nrows, ncols):\n",
    "    fig, axes = plt.subplots(nrows, ncols)\n",
    "    axes = axes.flatten()\n",
    "    for img, ax in zip(X, axes):\n",
    "        ax.imshow(img)\n",
    "        ax.get_xaxis().set_visible(False)\n",
    "        ax.get_yaxis().set_visible(False)\n",
    "    plt.show()\n",
    "    return axes"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [],
   "source": [
    "def accuracy(net, data_iter):\n",
    "    acc = 0\n",
    "    size = 0\n",
    "    for X, y in data_iter:\n",
    "        res = net(X)\n",
    "        y_hat = softmax(res)\n",
    "        acc += (y_hat.argmax(axis=1) == y.astype('float32')).sum().asscalar()\n",
    "        size += len(y)\n",
    "    return acc / size"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [],
   "source": [
    "def softmax(X):\n",
    "    X_exp = X.exp()\n",
    "    normalization_constant = X_exp.sum(axis=1, keepdims=True)\n",
    "    return X_exp / normalization_constant"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "def net(X):\n",
    "    X = X.reshape((-1, num_inputs))\n",
    "    H = nd.dot(X, W1) + b1\n",
    "    H = H.relu()\n",
    "    O = nd.dot(H, W2) + b2\n",
    "    return O"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [],
   "source": [
    "batch_size = 256\n",
    "train_iter, test_iter = get_dataloader(batch_size)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "num_inputs = 28 * 28\n",
    "num_hidden = 256\n",
    "num_output = 10"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [],
   "source": [
    "W1 = nd.random.normal(scale=0.01, shape=(num_inputs, num_hidden))\n",
    "b1 = nd.zeros(num_hidden)\n",
    "W2 = nd.random.normal(scale=0.01, shape=(num_hidden, num_output))\n",
    "b2 = nd.zeros(num_output)\n",
    "params = [W1, b1, W2, b2]\n",
    "for param in params:\n",
    "    param.attach_grad()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 0, acc 0.867200\n",
      "Epoch 1, acc 0.863100\n",
      "Epoch 2, acc 0.871100\n",
      "Epoch 3, acc 0.871200\n",
      "Epoch 4, acc 0.873800\n",
      "Epoch 5, acc 0.875400\n",
      "Epoch 6, acc 0.877300\n",
      "Epoch 7, acc 0.877300\n",
      "Epoch 8, acc 0.878700\n",
      "Epoch 9, acc 0.875600\n",
      "Epoch 10, acc 0.881900\n",
      "Epoch 11, acc 0.879400\n",
      "Epoch 12, acc 0.883200\n",
      "Epoch 13, acc 0.876400\n",
      "Epoch 14, acc 0.878000\n",
      "Epoch 15, acc 0.885300\n",
      "Epoch 16, acc 0.884300\n",
      "Epoch 17, acc 0.886200\n",
      "Epoch 18, acc 0.883500\n",
      "Epoch 19, acc 0.884100\n"
     ]
    }
   ],
   "source": [
    "loss = gluon.loss.SoftmaxCrossEntropyLoss()\n",
    "epochs = 20\n",
    "lr = 0.1\n",
    "for epoch in range(epochs):\n",
    "    for X, y in train_iter:\n",
    "        with autograd.record():\n",
    "            res = net(X)\n",
    "            l = loss(res, y)\n",
    "        l.backward()\n",
    "        for param in params:\n",
    "            param[:] = param - (lr/batch_size) * param.grad\n",
    "    epoch_acc = accuracy(net, test_iter)\n",
    "    print(\"Epoch %d, acc %f\" % (epoch, epoch_acc))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 10 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(256, 28, 28)\n",
      "(256, 10)\n"
     ]
    }
   ],
   "source": [
    "for X, y in train_iter:\n",
    "    X = X.squeeze(axis=1)\n",
    "    show_images(X.asnumpy(), 2, 5)\n",
    "    res = net(X)\n",
    "    print(X.shape)\n",
    "    print(res.shape)\n",
    "    break"
   ]
  }
 ],
 "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": "3.6.7"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}