examples/training_md_gan_25_gaussions.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": true,
    "pycharm": {
     "name": "#%% md\n"
    }
   },
   "source": [
    "# Imports"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "outputs": [],
   "source": [
    "import torch\n",
    "from torch import optim\n",
    "from tqdm import tqdm\n",
    "# Imports for this repository\n",
    "from simplex_generator import simplex_params\n",
    "from md_gan_modules import LambdaNetwork, Discriminator, Generator\n",
    "from gmm_data_generator import load_db\n",
    "from numpy_dataset import NumpyDataset\n",
    "from md_gan_training import MDGANTraining,gaussian_likelihood_sum"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n",
     "is_executing": false
    }
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "# Parameters"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%% md\n"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "outputs": [],
   "source": [
    "PARAMS = {'batch_size': 500,\n",
    "          'zdim': 2,\n",
    "          'gamma':0.5,\n",
    "          'beta1':0.5,\n",
    "          'beta2':0.999,\n",
    "          'n_lr_steps':3,\n",
    "          'lambda_training_iterations':10001,\n",
    "          'epochs':2000,\n",
    "          'eta_lambda': 0.01,\n",
    "          'e_dim': 9,\n",
    "          'lr_d': 1e-3,\n",
    "          'lr_g': 1e-3,\n",
    "          'epsilon': 1e-8,  # for avoiding numerical instabilities\n",
    "          'samp_num_gen': 2500}"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n",
     "is_executing": false
    }
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "# Init Simplex and Working Device"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%% md\n"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "outputs": [
    {
     "name": "stdout",
     "text": [
      "Current Working Device is set to:cuda\n"
     ],
     "output_type": "stream"
    }
   ],
   "source": [
    "working_device = torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\") # Find working device\n",
    "simplex = simplex_params(PARAMS['e_dim'], working_device) # Create Simplex\n",
    "print(\"Current Working Device is set to:\" + str(working_device))"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n",
     "is_executing": false
    }
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "# Search for $\\lambda$"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%% md\n"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "outputs": [
    {
     "name": "stderr",
     "text": [
      "100%|██████████| 10001/10001 [00:10<00:00, 910.70it/s]\n"
     ],
     "output_type": "stream"
    },
    {
     "name": "stdout",
     "text": [
      "Lambda Loss:3.812762498855591\n",
      "Lambda Loss:3.812762498855591\n",
      "Lambda Loss:3.812762498855591\n",
      "Lambda Loss:3.812762498855591\n",
      "Lambda Loss:3.812762498855591\n",
      "Lambda Loss:3.812762498855591\n",
      "Lambda Loss:3.812762498855591\n",
      "Lambda Loss:3.812762498855591\n",
      "Lambda Loss:3.812762498855591\n",
      "Lambda Loss:3.812762498855591\n",
      "0.02208706922829151\n"
     ],
     "output_type": "stream"
    }
   ],
   "source": [
    "lambda_net = LambdaNetwork(PARAMS['e_dim']).to(working_device)\n",
    "lambda_training_data = torch.tensor([1.0], device=working_device, dtype=torch.float32, requires_grad=False)\n",
    "optimizer_lambda = optim.Adam(lambda_net.parameters(), lr=PARAMS['eta_lambda'])\n",
    "\n",
    "for i in tqdm(range(PARAMS['lambda_training_iterations'])):\n",
    "    optimizer_lambda.zero_grad()\n",
    "    e = lambda_net(lambda_training_data)\n",
    "    lambda_lk = gaussian_likelihood_sum(e, simplex)\n",
    "    lambda_loss = -torch.log(PARAMS['epsilon'] + lambda_lk).mean()\n",
    "    if i % 1000 == 0 and i > 0:\n",
    "        print(\"Lambda Loss:\" + str(lambda_loss.item()))\n",
    "        for group in optimizer_lambda.param_groups:\n",
    "            group['lr'] = group['lr'] * PARAMS['gamma']\n",
    "    lambda_loss.backward()\n",
    "    optimizer_lambda.step()\n",
    "e = lambda_net(lambda_training_data)\n",
    "lambda_value = gaussian_likelihood_sum(e, simplex).sum().item()\n",
    "print(lambda_value)"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n",
     "is_executing": false
    }
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "# Train Generator and Discriminator"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%% md\n"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "outputs": [
    {
     "name": "stderr",
     "text": [
      "100%|██████████| 2000/2000 [06:04<00:00,  5.49it/s]\n"
     ],
     "output_type": "stream"
    },
    {
     "name": "stdout",
     "text": [
      "Learning rate steps: [500, 1000, 1500]\n"
     ],
     "output_type": "stream"
    }
   ],
   "source": [
    "net_g = Generator(PARAMS['zdim']).to(working_device) # Init generator function\n",
    "net_d = Discriminator(PARAMS['e_dim']).to(working_device) # Init discriminator function\n",
    "optimizer_g = optim.Adam(net_g.parameters(), lr=PARAMS['lr_g'], betas=(PARAMS['beta1'], PARAMS['beta2']))\n",
    "optimizer_d = optim.Adam(net_d.parameters(), lr=PARAMS['lr_d'], betas=(PARAMS['beta1'], PARAMS['beta2']))\n",
    "\n",
    "epoch_step_size=PARAMS['epochs']/(PARAMS['n_lr_steps']+1) # calculate learning rate decay step size\n",
    "lr_steps=[int((i+1)*epoch_step_size) for i in range(PARAMS['n_lr_steps'])] \n",
    "print(\"Learning rate steps:\",lr_steps)\n",
    "lr_g = optim.lr_scheduler.MultiStepLR(optimizer_g, lr_steps, gamma=PARAMS['gamma'])\n",
    "lr_d = optim.lr_scheduler.MultiStepLR(optimizer_d, lr_steps, gamma=PARAMS['gamma'])\n",
    "\n",
    "training_data = load_db()\n",
    "train_dataset = NumpyDataset(training_data)\n",
    "train_loader = torch.utils.data.DataLoader(dataset=train_dataset,\n",
    "                                           batch_size=PARAMS['batch_size'],\n",
    "                                           shuffle=True)\n",
    "\n",
    "md_gan_training=MDGANTraining(net_d, net_g, optimizer_d,optimizer_g,PARAMS['batch_size'], PARAMS['zdim'], simplex,\n",
    "                     PARAMS['epsilon'],\n",
    "                     lambda_value,\n",
    "                     working_device)\n",
    "# Start Training Loop\n",
    "for epoch in tqdm(range(PARAMS['epochs'])):\n",
    "    for d in train_loader:\n",
    "        d = d.to(working_device)\n",
    "        md_gan_training.update_discriminator(d)\n",
    "        md_gan_training.update_generator()\n",
    "\n",
    "    lr_g.step(epoch) # update Generator learning rate\n",
    "    lr_d.step(epoch) # update Discriminator learning rate"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n",
     "is_executing": false
    }
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "# Plot Samples from the Generator"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%% md\n"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "outputs": [
    {
     "data": {
      "text/plain": "<Figure size 432x288 with 1 Axes>",
      "image/png": "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\n"
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "from matplotlib import pyplot as plt\n",
    "\n",
    "g_samples = net_g(torch.randn(PARAMS['batch_size'], PARAMS['zdim']).to(working_device))\n",
    "g_samples = g_samples.cpu().detach().numpy()\n",
    "\n",
    "plt.plot(training_data[:, 0], training_data[:, 1], 'o',label='real data')\n",
    "plt.plot(g_samples[:, 0], g_samples[:, 1], '^',label='MD-GAN')\n",
    "plt.legend()\n",
    "plt.grid()\n",
    "plt.xlabel(r'$x_o$')\n",
    "plt.ylabel(r'$x_1$')\n",
    "plt.savefig('images/gmm_25.png')\n",
    "plt.show() "
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n",
     "is_executing": false
    }
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "# Save Generator"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%% md\n"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "outputs": [],
   "source": [
    "torch.save(net_g.state_dict(), 'generator_25.pt')\n",
    " "
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n",
     "is_executing": false
    }
   }
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.6"
  },
  "pycharm": {
   "stem_cell": {
    "cell_type": "raw",
    "source": [],
    "metadata": {
     "collapsed": false
    }
   }
  }
 },
 "nbformat": 4,
 "nbformat_minor": 0
}