{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# HSIC Network - Models that train without back-prop\n",
    "In this notebook we explain how you can use HSIC Bottleneck paradigm for training a feed-forward neural network.\n",
    "For more information refer to the paper: [HSIC Bottleneck](https://arxiv.org/abs/1908.01580)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Importing glow modules"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import glow\n",
    "from glow.layers import Dense, Dropout, Conv2d, Flatten, HSICoutput\n",
    "from glow.datasets import mnist, cifar10\n",
    "from glow.models import IBSequential, Sequential, HSICSequential, Network\n",
    "from glow.information_bottleneck.estimator import HSIC\n",
    "from glow.information_bottleneck import Estimator\n",
    "import torch"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Load dataset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "# hyperparameter\n",
    "batch_size = 64\n",
    "num_workers = 3\n",
    "validation_split = 0.2\n",
    "num_epochs = 3\n",
    "\n",
    "# load the dataset\n",
    "train_loader, val_loader, test_loader = mnist.load_data(\n",
    "    batch_size=batch_size, num_workers=num_workers, validation_split=validation_split\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## HSIC Bottleneck-based Model\n",
    "These types of network uses a different paradigm of training a neural network which is described in the paper 'HSIC Bottleneck - Deep Learing without Back-Propagation'."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Running on CUDA enabled device !\n"
     ]
    }
   ],
   "source": [
    "model = HSICSequential(input_shape=(1, 28, 28), gpu=True)\n",
    "model.add(Conv2d(filters=16, kernel_size=3, stride=1, padding=1, activation='relu'))\n",
    "model.add(Flatten())\n",
    "model.add(Dense(500, activation='relu'), HSIC(kernel='gaussian', gpu=True, sigma=5), regularize_coeff=100)\n",
    "model.add(Dense(200, activation='relu'))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Compile and Pre-Training Phase\n",
    "Compile the model with HSIC IB-based loss objective and train the network for obtaining optimal intermediate representations (which is called pre-training phase)."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\r",
      "  0%|          | 0/750 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "\n",
      "Pre-Train-Epoch 1/3\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 750/750 [05:34<00:00,  2.20it/s]\n",
      "  0%|          | 0/750 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "\n",
      "Pre-Train-Epoch 2/3\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 750/750 [05:23<00:00,  2.03it/s]\n",
      "  0%|          | 0/750 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "\n",
      "Pre-Train-Epoch 3/3\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 750/750 [05:29<00:00,  2.35it/s]\n"
     ]
    }
   ],
   "source": [
    "model.compile(loss_criterion=HSIC(kernel='gaussian', gpu=True, sigma=10), optimizer='SGD', regularize_coeff=100)\n",
    "model.pre_training_loop(num_epochs, train_loader, val_loader)"
   ]
  },
  {
   "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": "3.6.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}