TF_vs_PyTorch.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "extraordinary-korean",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1.7.1\n"
     ]
    }
   ],
   "source": [
    "import torch\n",
    "from torch import nn\n",
    "\n",
    "print(torch.__version__)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "developing-remains",
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "data_path = os.path.expanduser('~/data')\n",
    "\n",
    "from torchvision.datasets import FashionMNIST\n",
    "from torchvision import transforms\n",
    "\n",
    "transform = transforms.ToTensor()\n",
    "\n",
    "train_set = FashionMNIST(data_path, train=True, download=True, transform=transform)\n",
    "test_set = FashionMNIST(data_path, train=False, download=True, transform=transform)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "running-pontiac",
   "metadata": {},
   "outputs": [],
   "source": [
    "from torch.utils.data import DataLoader\n",
    "\n",
    "batch_size = 2048\n",
    "\n",
    "train_loader = DataLoader(train_set, batch_size=batch_size, shuffle=True, pin_memory=True)\n",
    "test_loader = DataLoader(test_set, batch_size=batch_size, shuffle=False, pin_memory=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "supported-dakota",
   "metadata": {},
   "outputs": [],
   "source": [
    "model = nn.Sequential(\n",
    "    nn.Flatten(),\n",
    "    nn.Linear(28 * 28, 128),\n",
    "    nn.ReLU(),\n",
    "    nn.Linear(128, 10)\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "affiliated-beads",
   "metadata": {},
   "outputs": [],
   "source": [
    "from torch import optim\n",
    "\n",
    "optimizer = optim.Adam(model.parameters(), lr=0.001, betas=[0.9, 0.999], eps=1e-07)\n",
    "criterion = nn.CrossEntropyLoss()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "manufactured-bolivia",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "01 / 10 .............................. - loss: 1.2759 - accuracy: 63.2250%\n",
      "02 / 10 .............................. - loss: 0.6633 - accuracy: 77.4350%\n",
      "03 / 10 .............................. - loss: 0.5554 - accuracy: 81.3700%\n",
      "04 / 10 .............................. - loss: 0.5054 - accuracy: 82.9217%\n",
      "05 / 10 .............................. - loss: 0.4770 - accuracy: 83.6900%\n",
      "06 / 10 .............................. - loss: 0.4547 - accuracy: 84.4750%\n",
      "07 / 10 .............................. - loss: 0.4358 - accuracy: 85.1000%\n",
      "08 / 10 .............................. - loss: 0.4213 - accuracy: 85.5683%\n",
      "09 / 10 .............................. - loss: 0.4105 - accuracy: 85.8400%\n",
      "10 / 10 .............................. - loss: 0.3982 - accuracy: 86.2833%\n"
     ]
    }
   ],
   "source": [
    "model.train()\n",
    "\n",
    "epochs = 10\n",
    "\n",
    "for epoch in range(epochs):\n",
    "    print(f'{epoch+1:0>2} / {epochs}', end=' ')\n",
    "    running_loss = 0.0\n",
    "    running_accuracy = 0.0\n",
    "    for imgs, tgts in train_loader:\n",
    "        tgts_logits = model(imgs)\n",
    "        loss = criterion(tgts_logits, tgts)\n",
    "        loss.backward()\n",
    "        optimizer.step()\n",
    "        optimizer.zero_grad()\n",
    "        \n",
    "        predictions = tgts_logits.argmax(axis=1)\n",
    "        accuracy = (predictions == tgts).float().mean()\n",
    "        \n",
    "        running_loss += loss.item() * len(imgs)\n",
    "        running_accuracy += accuracy.item() * len(imgs)\n",
    "        \n",
    "        print('.', end='')\n",
    "\n",
    "    running_loss /= len(train_set)\n",
    "    running_accuracy /= len(train_set)\n",
    "    print(f' - loss: {running_loss:.4f} - accuracy: {running_accuracy:.4%}')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "reliable-tattoo",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "..... - loss: 0.4345 - accuracy: 84.8000%\n"
     ]
    }
   ],
   "source": [
    "model.eval()\n",
    "\n",
    "running_loss = 0.0\n",
    "running_accuracy = 0.0\n",
    "for imgs, tgts in test_loader:\n",
    "    tgts_logits = model(imgs)\n",
    "    loss = criterion(tgts_logits, tgts)\n",
    "\n",
    "    predictions = tgts_logits.argmax(axis=1)\n",
    "    accuracy = (predictions == tgts).float().mean()\n",
    "\n",
    "    running_loss += loss.item() * len(imgs)\n",
    "    running_accuracy += accuracy.item() * len(imgs)\n",
    "\n",
    "    print('.', end='')\n",
    "\n",
    "running_loss /= len(test_set)\n",
    "running_accuracy /= len(test_set)\n",
    "print(f' - loss: {running_loss:.4f} - accuracy: {running_accuracy:.4%}')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "compatible-jesus",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Tensorflow\n",
    "import tensorflow as tf\n",
    "\n",
    "model_tf = tf.keras.Sequential([\n",
    "    tf.keras.layers.Flatten(input_shape=(28, 28)),\n",
    "    tf.keras.layers.Dense(128, activation='relu'),\n",
    "    tf.keras.layers.Dense(10)\n",
    "])\n",
    "\n",
    "model_tf.compile(optimizer='adam',\n",
    "                 loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),\n",
    "                 metrics=['accuracy'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "coupled-mapping",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/10\n",
      "1875/1875 [==============================] - 2s 1ms/step - loss: 0.4929 - accuracy: 0.8268\n",
      "Epoch 2/10\n",
      "1875/1875 [==============================] - 2s 1ms/step - loss: 0.3727 - accuracy: 0.8655\n",
      "Epoch 3/10\n",
      "1875/1875 [==============================] - 2s 1ms/step - loss: 0.3349 - accuracy: 0.8787\n",
      "Epoch 4/10\n",
      "1875/1875 [==============================] - 2s 1ms/step - loss: 0.3106 - accuracy: 0.8857\n",
      "Epoch 5/10\n",
      "1875/1875 [==============================] - 2s 1ms/step - loss: 0.2916 - accuracy: 0.8922\n",
      "Epoch 6/10\n",
      "1875/1875 [==============================] - 2s 1ms/step - loss: 0.2772 - accuracy: 0.8976\n",
      "Epoch 7/10\n",
      "1875/1875 [==============================] - 2s 1ms/step - loss: 0.2659 - accuracy: 0.9018\n",
      "Epoch 8/10\n",
      "1875/1875 [==============================] - 2s 1ms/step - loss: 0.2557 - accuracy: 0.9046\n",
      "Epoch 9/10\n",
      "1875/1875 [==============================] - 2s 1ms/step - loss: 0.2464 - accuracy: 0.9087\n",
      "Epoch 10/10\n",
      "1875/1875 [==============================] - 2s 1ms/step - loss: 0.2375 - accuracy: 0.9117\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<tensorflow.python.keras.callbacks.History at 0x7fa8482441d0>"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "imgs = train_set.data.numpy() / 255.0\n",
    "lbls = train_set.targets.numpy()\n",
    "\n",
    "model_tf.fit(imgs, lbls, epochs=10)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "compliant-thinking",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "313/313 - 0s - loss: 0.3318 - accuracy: 0.8857\n",
      "\n",
      "Test accuracy: 0.885699987411499\n"
     ]
    }
   ],
   "source": [
    "imgs = test_set.data.numpy() / 255.0\n",
    "lbls = test_set.targets.numpy()\n",
    "\n",
    "test_loss, test_acc = model_tf.evaluate(imgs,  lbls, verbose=2)\n",
    "print('\\nTest accuracy:', test_acc)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "subject-custom",
   "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": "3.7.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}