pytorch_autograd_examples.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Three methods of backpropagation with PyTorch\n",
    "\n",
    "#### 1.- Using loss function's `.backward()` call"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 124,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "19 tensor(1.1563, grad_fn=<SumBackward0>)\n",
      "tensor([-1.3625, -3.2816, -0.8294])\n",
      "tensor([1.3188, 2.5898, 2.9539], requires_grad=True)\n",
      "39 tensor(0.2308, grad_fn=<SumBackward0>)\n",
      "tensor([-0.9096, -0.6192, -0.0157])\n",
      "tensor([1.5452, 2.9226, 2.9991], requires_grad=True)\n",
      "59 tensor(0.0930, grad_fn=<SumBackward0>)\n",
      "tensor([-6.0725e-01, -1.1684e-01, -2.9755e-04])\n",
      "tensor([1.6964, 2.9854, 3.0000], requires_grad=True)\n",
      "79 tensor(0.0411, grad_fn=<SumBackward0>)\n",
      "tensor([-4.0541e-01, -2.2047e-02, -1.1444e-05])\n",
      "tensor([1.7973, 2.9972, 3.0000], requires_grad=True)\n",
      "99 tensor(0.0183, grad_fn=<SumBackward0>)\n",
      "tensor([-2.7065e-01, -4.1580e-03, -1.1444e-05])\n",
      "tensor([1.8647, 2.9995, 3.0000], requires_grad=True)\n",
      "119 tensor(0.0082, grad_fn=<SumBackward0>)\n",
      "tensor([-1.8069e-01, -7.8392e-04, -1.1444e-05])\n",
      "tensor([1.9097, 2.9999, 3.0000], requires_grad=True)\n",
      "139 tensor(0.0036, grad_fn=<SumBackward0>)\n",
      "tensor([-1.2063e-01, -1.4877e-04, -1.1444e-05])\n",
      "tensor([1.9397, 3.0000, 3.0000], requires_grad=True)\n",
      "159 tensor(0.0016, grad_fn=<SumBackward0>)\n",
      "tensor([-8.0533e-02, -2.8610e-05, -1.1444e-05])\n",
      "tensor([1.9597, 3.0000, 3.0000], requires_grad=True)\n",
      "179 tensor(0.0007, grad_fn=<SumBackward0>)\n",
      "tensor([-5.3765e-02, -1.1444e-05, -1.1444e-05])\n",
      "tensor([1.9731, 3.0000, 3.0000], requires_grad=True)\n",
      "199 tensor(0.0003, grad_fn=<SumBackward0>)\n",
      "tensor([-3.5894e-02, -1.1444e-05, -1.1444e-05])\n",
      "tensor([1.9821, 3.0000, 3.0000], requires_grad=True)\n"
     ]
    }
   ],
   "source": [
    "import torch\n",
    "\n",
    "learning_rate = 1e-2\n",
    "\n",
    "x = torch.tensor([1.,2.,3.])\n",
    "w = torch.tensor([1.,1.,1.], requires_grad=True)\n",
    "y = torch.tensor([2.,6.,9.])\n",
    "\n",
    "\n",
    "for i in range(200): \n",
    "    loss = (x.mul(w) - y).pow(2).sum()\n",
    "\n",
    "    loss.backward()\n",
    "    if i % 20 == 19:\n",
    "        print(i, loss)\n",
    "        print(w.grad)\n",
    "        print(w)\n",
    "    with torch.no_grad():\n",
    "        w -= learning_rate * w.grad\n",
    "\n",
    "    w.grad.zero_()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### 2.- Using `autograd.backward()` function\n",
    "\n",
    "*Method 1 is just a method that wraps this one*"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 126,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "19 tensor(1.1563, grad_fn=<SumBackward0>)\n",
      "tensor([-1.3625, -3.2816, -0.8294])\n",
      "tensor([1.3188, 2.5898, 2.9539], requires_grad=True)\n",
      "39 tensor(0.2308, grad_fn=<SumBackward0>)\n",
      "tensor([-0.9096, -0.6192, -0.0157])\n",
      "tensor([1.5452, 2.9226, 2.9991], requires_grad=True)\n",
      "59 tensor(0.0930, grad_fn=<SumBackward0>)\n",
      "tensor([-6.0725e-01, -1.1684e-01, -2.9755e-04])\n",
      "tensor([1.6964, 2.9854, 3.0000], requires_grad=True)\n",
      "79 tensor(0.0411, grad_fn=<SumBackward0>)\n",
      "tensor([-4.0541e-01, -2.2047e-02, -1.1444e-05])\n",
      "tensor([1.7973, 2.9972, 3.0000], requires_grad=True)\n",
      "99 tensor(0.0183, grad_fn=<SumBackward0>)\n",
      "tensor([-2.7065e-01, -4.1580e-03, -1.1444e-05])\n",
      "tensor([1.8647, 2.9995, 3.0000], requires_grad=True)\n",
      "119 tensor(0.0082, grad_fn=<SumBackward0>)\n",
      "tensor([-1.8069e-01, -7.8392e-04, -1.1444e-05])\n",
      "tensor([1.9097, 2.9999, 3.0000], requires_grad=True)\n",
      "139 tensor(0.0036, grad_fn=<SumBackward0>)\n",
      "tensor([-1.2063e-01, -1.4877e-04, -1.1444e-05])\n",
      "tensor([1.9397, 3.0000, 3.0000], requires_grad=True)\n",
      "159 tensor(0.0016, grad_fn=<SumBackward0>)\n",
      "tensor([-8.0533e-02, -2.8610e-05, -1.1444e-05])\n",
      "tensor([1.9597, 3.0000, 3.0000], requires_grad=True)\n",
      "179 tensor(0.0007, grad_fn=<SumBackward0>)\n",
      "tensor([-5.3765e-02, -1.1444e-05, -1.1444e-05])\n",
      "tensor([1.9731, 3.0000, 3.0000], requires_grad=True)\n",
      "199 tensor(0.0003, grad_fn=<SumBackward0>)\n",
      "tensor([-3.5894e-02, -1.1444e-05, -1.1444e-05])\n",
      "tensor([1.9821, 3.0000, 3.0000], requires_grad=True)\n"
     ]
    }
   ],
   "source": [
    "x = torch.tensor([1.,2.,3.])\n",
    "w = torch.tensor([1.,1.,1.], requires_grad=True)\n",
    "y = torch.tensor([2.,6.,9.])\n",
    "\n",
    "for i in range(200):\n",
    "    loss = (x.mul(w) - y).pow(2).sum()\n",
    "    \n",
    "    torch.autograd.backward(loss)\n",
    "    if i % 20 == 19:\n",
    "        print(i, loss)\n",
    "        print(w.grad)\n",
    "        print(w)\n",
    "    with torch.no_grad():\n",
    "        w -= learning_rate * w.grad\n",
    "        w.grad.zero_()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### 3.- Using `autograd.grad()` function\n",
    "\n",
    "*Method 2 uses this one iterating through each variable that needs a gradient*"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 125,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "19 tensor(1.1563, grad_fn=<SumBackward0>)\n",
      "tensor([-1.3625, -3.2816, -0.8294])\n",
      "tensor([1.3188, 2.5898, 2.9539], requires_grad=True)\n",
      "39 tensor(0.2308, grad_fn=<SumBackward0>)\n",
      "tensor([-0.9096, -0.6192, -0.0157])\n",
      "tensor([1.5452, 2.9226, 2.9991], requires_grad=True)\n",
      "59 tensor(0.0930, grad_fn=<SumBackward0>)\n",
      "tensor([-6.0725e-01, -1.1684e-01, -2.9755e-04])\n",
      "tensor([1.6964, 2.9854, 3.0000], requires_grad=True)\n",
      "79 tensor(0.0411, grad_fn=<SumBackward0>)\n",
      "tensor([-4.0541e-01, -2.2047e-02, -1.1444e-05])\n",
      "tensor([1.7973, 2.9972, 3.0000], requires_grad=True)\n",
      "99 tensor(0.0183, grad_fn=<SumBackward0>)\n",
      "tensor([-2.7065e-01, -4.1580e-03, -1.1444e-05])\n",
      "tensor([1.8647, 2.9995, 3.0000], requires_grad=True)\n",
      "119 tensor(0.0082, grad_fn=<SumBackward0>)\n",
      "tensor([-1.8069e-01, -7.8392e-04, -1.1444e-05])\n",
      "tensor([1.9097, 2.9999, 3.0000], requires_grad=True)\n",
      "139 tensor(0.0036, grad_fn=<SumBackward0>)\n",
      "tensor([-1.2063e-01, -1.4877e-04, -1.1444e-05])\n",
      "tensor([1.9397, 3.0000, 3.0000], requires_grad=True)\n",
      "159 tensor(0.0016, grad_fn=<SumBackward0>)\n",
      "tensor([-8.0533e-02, -2.8610e-05, -1.1444e-05])\n",
      "tensor([1.9597, 3.0000, 3.0000], requires_grad=True)\n",
      "179 tensor(0.0007, grad_fn=<SumBackward0>)\n",
      "tensor([-5.3765e-02, -1.1444e-05, -1.1444e-05])\n",
      "tensor([1.9731, 3.0000, 3.0000], requires_grad=True)\n",
      "199 tensor(0.0003, grad_fn=<SumBackward0>)\n",
      "tensor([-3.5894e-02, -1.1444e-05, -1.1444e-05])\n",
      "tensor([1.9821, 3.0000, 3.0000], requires_grad=True)\n"
     ]
    }
   ],
   "source": [
    "import torch\n",
    "\n",
    "learning_rate = 1e-2\n",
    "\n",
    "x = torch.tensor([1.,2.,3.])\n",
    "w = torch.tensor([1.,1.,1.], requires_grad=True)\n",
    "y = torch.tensor([2.,6.,9.])\n",
    "\n",
    "for i in range(200):\n",
    "    loss = (x.mul(w) - y).pow(2).sum()\n",
    "    \n",
    "    w.grad, = torch.autograd.grad(loss, w)\n",
    "    if i % 20 == 19:\n",
    "        print(i, loss)\n",
    "        print(w.grad)\n",
    "        print(w)\n",
    "    with torch.no_grad():\n",
    "        w -= learning_rate * w.grad"
   ]
  },
  {
   "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": "3.8.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}