plain numpy, torch.autograd and torch.nn implementation of a simple nn

1_numpy_net.ipynb

2_torch_autograd_net.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import torch\n",
    "from torch.autograd import Variable"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "# N is batch size; D_in is input dimension;\n",
    "# H is hidden dimension; D_out is output dimension.\n",
    "N, D_in, H, D_out = 6, 3, 10, 1\n",
    "\n",
    "w_1 = Variable(torch.randn(D_in, H),  requires_grad=True)\n",
    "h_1 = Variable(torch.randn(1, H),     requires_grad=True)\n",
    "w_2 = Variable(torch.randn(H, D_out), requires_grad=True)\n",
    "h_2 = Variable(torch.randn(1, D_out), requires_grad=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "learning_rate = 1e-6\n",
    "# X: N x D_in\n",
    "# y: N x D_out\n",
    "def train(X, y):\n",
    "    global w_1, h_1, w_2, h_2, learning_rate\n",
    "    \n",
    "    X, y = Variable(torch.Tensor(X)), Variable(torch.Tensor(y))\n",
    "    \n",
    "    # forward\n",
    "    # we have to use `repeat` since\n",
    "    # torch doesn't support broadcasting\n",
    "    m = X.mm(w_1) + h_1.repeat(N, 1)      # m:      N x H\n",
    "    m_relu = m.clamp(min=0)               # m_relu: N x H\n",
    "    p = m_relu.mm(w_2) + h_2.repeat(N, 1) # p:      N x D_out\n",
    "    \n",
    "    # loss = Sum((y - p)^2)\n",
    "    loss = (p - y).pow(2).sum()\n",
    "    \n",
    "    loss.backward()\n",
    "    \n",
    "    w_1.data -= w_1.grad.data * learning_rate\n",
    "    h_1.data -= h_1.grad.data * learning_rate\n",
    "    w_2.data -= w_2.grad.data * learning_rate\n",
    "    h_2.data -= h_2.grad.data * learning_rate\n",
    "    \n",
    "    w_1.grad.data.zero_()\n",
    "    h_1.grad.data.zero_()\n",
    "    w_2.grad.data.zero_()\n",
    "    h_2.grad.data.zero_()\n",
    "    \n",
    "    return (p, loss)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0 106.6131820678711\n",
      "1000 64.248291015625\n",
      "2000 41.778751373291016\n",
      "3000 28.251008987426758\n",
      "4000 19.758804321289062\n",
      "5000 14.260133743286133\n",
      "6000 10.607431411743164\n",
      "7000 8.12516975402832\n",
      "8000 6.402351379394531\n",
      "9000 5.181938171386719\n",
      "10000 4.300069808959961\n",
      "11000 3.650324583053589\n",
      "12000 3.162468671798706\n",
      "13000 2.789226531982422\n",
      "14000 2.498776435852051\n",
      "15000 2.2687385082244873\n",
      "16000 2.0838499069213867\n",
      "17000 1.9329262971878052\n",
      "18000 1.8078796863555908\n",
      "19000 1.7029905319213867\n",
      "20000 1.6138317584991455\n",
      "21000 1.5370678901672363\n",
      "22000 1.470285415649414\n",
      "23000 1.411571741104126\n",
      "24000 1.3593971729278564\n",
      "25000 1.3125015497207642\n",
      "26000 1.2700220346450806\n",
      "27000 1.2311971187591553\n",
      "28000 1.1954683065414429\n",
      "29000 1.1623502969741821\n",
      "30000 1.131489872932434\n",
      "31000 1.1025365591049194\n",
      "32000 1.0751968622207642\n",
      "33000 1.0493370294570923\n",
      "34000 1.0245527029037476\n",
      "35000 1.0009534358978271\n",
      "36000 0.9783092141151428\n",
      "37000 0.9566856026649475\n",
      "38000 0.9378811120986938\n",
      "39000 0.9202557802200317\n",
      "40000 0.9032085537910461\n",
      "41000 0.886813759803772\n",
      "42000 0.8709046244621277\n",
      "43000 0.8553593158721924\n",
      "44000 0.8402742147445679\n",
      "45000 0.8256545066833496\n",
      "46000 0.8113911747932434\n",
      "47000 0.7974827885627747\n",
      "48000 0.7839452028274536\n",
      "49000 0.7708430886268616\n"
     ]
    }
   ],
   "source": [
    "for i in range(50000):\n",
    "    data = [\n",
    "        (1, [1,2,3]),\n",
    "        (1, [1,2,4]),\n",
    "        (1, [2,3,4]),\n",
    "        (0, [1,4,2]),\n",
    "        (0, [2,4,3]),\n",
    "        (0, [5,3,1])\n",
    "    ]\n",
    "    res, loss = train([x[1] for x in data], [[x[0]] for x in data])\n",
    "    if i % 1000 == 0:\n",
    "        print(i, loss.data[0])"
   ]
  }
 ],
 "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": "3.6.1"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}

3_torch_nn_net.ipynb

See http://pytorch.org/tutorials/beginner/pytorch_with_examples.html.