pytorchkeras.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/Users/gokcen/.miniconda3/lib/python3.6/site-packages/statsmodels/compat/pandas.py:56: FutureWarning: The pandas.core.datetools module is deprecated and will be removed in a future version. Please use the pandas.tseries module instead.\n",
      "  from pandas.core import datetools\n",
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "import numpy as np\n",
    "import pandas as pd\n",
    "import plotnine as p9\n",
    "\n",
    "import torch\n",
    "from torch.autograd import Variable\n",
    "from torch.utils.data import TensorDataset, DataLoader, Dataset\n",
    "\n",
    "np.random.seed(555)\n",
    "torch.manual_seed(555)\n",
    "\n",
    "num_sample = 300\n",
    "num_feat = 10\n",
    "num_out = 10\n",
    "\n",
    "batch_size = 32\n",
    "lr = 1e-4\n",
    "epochs = 300\n",
    "\n",
    "X = np.random.normal(0, 0.5, (num_sample, num_feat)).astype(np.float64)\n",
    "W = np.random.normal(0, 0.5, (num_feat, num_out)).astype(np.float64)\n",
    "b = np.random.normal(0, 0.5, (1, num_out)).astype(np.float64)\n",
    "Y = np.dot(X, W) + b\n",
    "X -= X.mean(0)\n",
    "\n",
    "ds = TensorDataset(torch.from_numpy(X).double(), torch.from_numpy(Y).double())\n",
    "\n",
    "# train torch models with and without shuffling\n",
    "model1 = torch.nn.Linear(num_feat, num_out).double()\n",
    "opt1 = torch.optim.SGD(model1.parameters(), lr=lr)\n",
    "loss1 = torch.nn.MSELoss()\n",
    "\n",
    "train_hist_torch = []\n",
    "\n",
    "for epoch in range(epochs):\n",
    "\n",
    "    train_batch_losses = []\n",
    "    for x, y in DataLoader(ds, batch_size=batch_size, shuffle=True, drop_last=True):\n",
    "        x_var, y_var = Variable(x, requires_grad=False), Variable(y, requires_grad=False)\n",
    "        pred = model1(x_var)\n",
    "        l = loss1(pred, y_var)\n",
    "        train_batch_losses.append(l.data[0])\n",
    "\n",
    "        opt1.zero_grad()\n",
    "        l.backward()\n",
    "        opt1.step()\n",
    "\n",
    "    # save mean of all batch errors within the epoch\n",
    "    train_hist_torch.append(np.array(train_batch_losses).mean())\n",
    "\n",
    "\n",
    "from keras.models import Model\n",
    "from keras.layers import Input, Dense\n",
    "from keras.optimizers import SGD\n",
    "from keras import backend as K\n",
    "\n",
    "\n",
    "inputs = Input(shape=(num_feat,))\n",
    "predictions = Dense(num_out, activation='linear')(inputs)\n",
    "model = Model(inputs=inputs, outputs=predictions)\n",
    "\n",
    "opt = SGD(lr=lr)\n",
    "model.compile(optimizer=opt, loss='mse')\n",
    "losses = model.fit(X, Y,\n",
    "                   batch_size=batch_size,\n",
    "                   epochs=epochs, verbose=0, shuffle=True)\n",
    "        \n",
    "train_hist_keras = losses.history['loss']\n",
    "    \n",
    "(p9.ggplot(pd.DataFrame({'torch_train_torch': train_hist_torch,\n",
    "                         'torch_train_keras': train_hist_keras,\n",
    "                         'epochs': range(len(train_hist_torch))}),\n",
    "            p9.aes(x='epochs')) +\n",
    "  p9.geom_path(p9.aes(y='torch_train_torch', color='\"training loss (sgd, torch)\"')) +\n",
    "  p9.geom_path(p9.aes(y='torch_train_keras', color='\"training loss (sgd, keras)\"')) +\n",
    "  p9.labs(color='Loss', y=' ') +\n",
    "  p9.theme_minimal())"
   ]
  }
 ],
 "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
}