MOG_visualization.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# This is the demo of TAC-GAN on MOG (mixture of gaussian) "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Load the needed package"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Import and libraries"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "%matplotlib inline\n",
    "import os\n",
    "import torch\n",
    "import torch.nn as nn\n",
    "import numpy as np\n",
    "import torch.optim as optim\n",
    "import torch.nn.functional as F\n",
    "import matplotlib\n",
    "import matplotlib.pyplot as plt\n",
    "import pandas as pd\n",
    "from mmd_metric import polynomial_mmd"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Classes for cGAN"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "#######G network\n",
    "\n",
    "class G_guassian(nn.Module):\n",
    "\n",
    "    def __init__(self, nz,num_classes=2):\n",
    "        super(G_guassian, self).__init__()\n",
    "\n",
    "\n",
    "        self.embed = nn.Embedding(num_embeddings=num_classes, embedding_dim=nz)\n",
    "\n",
    "        self.decode = nn.Sequential(\n",
    "\n",
    "            nn.Linear(nz*2,10),\n",
    "            nn.Tanh(),\n",
    "            nn.Linear(10, 10),\n",
    "            nn.Tanh(),\n",
    "            nn.Linear(10, 10),\n",
    "            # nn.Tanh(),\n",
    "            nn.Linear(10, 1),\n",
    "        )\n",
    "\n",
    "\n",
    "        self.__initialize_weights()\n",
    "\n",
    "    def forward(self, z, label, output=None):\n",
    "\n",
    "        input = torch.cat([z,self.embed(label)],dim=1)\n",
    "        x = input.view(input.size(0), -1)\n",
    "        output = self.decode(x)\n",
    "\n",
    "        return output\n",
    "\n",
    "    def __initialize_weights(self):\n",
    "        for m in self.modules():\n",
    "            if isinstance(m, nn.Conv2d):\n",
    "                m.weight.data.normal_(0.0, 0.02)\n",
    "            elif isinstance(m, nn.BatchNorm2d):\n",
    "                m.weight.data.normal_(1.0, 0.02)\n",
    "                m.bias.data.fill_(0)\n",
    "                \n",
    "\n",
    "#####D network\n",
    "class D_guassian(nn.Module):\n",
    "\n",
    "    def __init__(self, num_classes=10,AC=True):\n",
    "        super(D_guassian, self).__init__()\n",
    "\n",
    "        self.AC = AC\n",
    "\n",
    "        self.encode = nn.Sequential(\n",
    "\n",
    "            nn.Linear(1, 10),\n",
    "            nn.Tanh(),\n",
    "            nn.Linear(10, 10),\n",
    "            nn.Tanh(),\n",
    "            nn.Linear(10, 10),\n",
    "            # nn.Tanh(),\n",
    "        )\n",
    "        self.gan_linear = nn.Linear(10, 1)\n",
    "        self.aux_linear = nn.Linear(10, num_classes)\n",
    "        self.mi_linear = nn.Linear(10, num_classes)\n",
    "\n",
    "        if not self.AC:\n",
    "            self.projection = nn.Embedding(num_embeddings=num_classes,embedding_dim=10)\n",
    "\n",
    "        self.sigmoid = nn.Sigmoid()\n",
    "        self.__initialize_weights()\n",
    "\n",
    "    def forward(self, input,y=None):\n",
    "\n",
    "        x = self.encode(input)\n",
    "        x = x.view(-1, 10)\n",
    "        c = self.aux_linear(x)\n",
    "        mi = self.mi_linear(x)\n",
    "\n",
    "        s = self.gan_linear(x)\n",
    "        if not self.AC:\n",
    "            s += (self.projection(y)*x).sum(dim=1,keepdim=True)\n",
    "        s = self.sigmoid(s)\n",
    "        return s.squeeze(1), c.squeeze(1), mi.squeeze(1)\n",
    "\n",
    "    def __initialize_weights(self):\n",
    "        for m in self.modules():\n",
    "            if isinstance(m, nn.Conv2d):\n",
    "                m.weight.data.normal_(0.0, 0.02)\n",
    "            elif isinstance(m, nn.BatchNorm2d):\n",
    "                m.weight.data.normal_(1.0, 0.02)\n",
    "                m.bias.data.fill_(0)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Helper functions to plot the results"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "def plot_density(flights,binwidth=0.1):\n",
    "    ax = plt.subplot(1,1,1)\n",
    "\n",
    "    # Draw the plot\n",
    "    ax.hist(flights, bins=int(180 / binwidth),\n",
    "            color='blue', edgecolor='black')\n",
    "\n",
    "    # Title and labels\n",
    "    ax.set_title('Histogram with Binwidth = %d' % binwidth, size=30)\n",
    "    ax.set_xlabel('Delay (min)', size=22)\n",
    "    ax.set_ylabel('Flights', size=22)\n",
    "\n",
    "    plt.tight_layout()\n",
    "    plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Runner"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "####TAC-GAN: set AC=True, MI=True, AC-GAN:set AC=True, MI=False, Projection-GAN: set AC=False, MI=False\n",
    "def train(data1,data2,data3,nz,G,D,optd,optg,AC=True,MI=True):\n",
    "    for _ in range(20):\n",
    "        for i in range(1000):\n",
    "\n",
    "            #####D step\n",
    "            for _ in range(1):\n",
    "                data = torch.cat(\n",
    "                    [data1[128 * i:128 * i + 128], data2[128 * i:128 * i + 128], data3[128 * i:128 * i + 128]],\n",
    "                    dim=0).unsqueeze(dim=1)\n",
    "                label = torch.cat([torch.ones(128).cuda().long()*0, torch.ones(128).cuda().long()*1, torch.ones(128).cuda().long()*2],dim=0)\n",
    "\n",
    "                ###D\n",
    "                d_real, c, _ = D(data)\n",
    "\n",
    "                z = torch.randn(256, nz).cuda()\n",
    "                fake_label = torch.LongTensor(256).random_(3).cuda()\n",
    "                fake_data = G(z, label=fake_label)\n",
    "                d_fake, _, mi = D(fake_data)\n",
    "\n",
    "\n",
    "                D_loss = F.binary_cross_entropy(d_real, torch.ones(384).cuda()) \\\n",
    "                         + F.binary_cross_entropy(d_fake, torch.zeros(256).cuda())\n",
    "                if AC:\n",
    "                    D_loss += F.cross_entropy(c, label)\n",
    "                if MI:\n",
    "                    D_loss += F.cross_entropy(mi, fake_label)\n",
    "\n",
    "                optd.zero_grad()\n",
    "                D_loss.backward()\n",
    "                optd.step()\n",
    "\n",
    "            #####G step\n",
    "            if i % 10 == 0:\n",
    "                z = torch.randn(256, nz).cuda()\n",
    "                fake_label = torch.LongTensor(256).random_(3).cuda()\n",
    "                fake_data = G(z, label=fake_label)\n",
    "                d_fake, c, mi = D(fake_data)\n",
    "\n",
    "                G_loss = F.binary_cross_entropy(d_fake, torch.ones(256).cuda())\n",
    "\n",
    "                if AC:\n",
    "                    G_loss += F.cross_entropy(c,fake_label)\n",
    "                if MI:\n",
    "                    G_loss -= F.cross_entropy(mi, fake_label)\n",
    "\n",
    "                optg.zero_grad()\n",
    "                G_loss.backward()\n",
    "                optg.step()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Visualize and evaluation"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Running the 1-D MOG toy experiments of different distance between different gaussian components. Cost about 10 min in 1080ti"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "\n",
    "######create folder to save the result\n",
    "distance = 3\n",
    "if os.path.exists(os.path.join('MOG','1D', str(distance) + '_1D')):\n",
    "    pass\n",
    "else:\n",
    "    os.makedirs(os.path.join('MOG','1D', str(distance) + '_1D'))\n",
    "save_path = os.path.join('MOG','1D', str(distance) + '_1D')\n",
    "\n",
    "######Generate MOG data of different mean and variance\n",
    "data1 = torch.randn(128000).cuda()\n",
    "data2 = torch.randn(128000).cuda() * 2 + distance\n",
    "data3 = torch.randn(128000).cuda() * 3 + distance * 2\n",
    "\n",
    "df1 = pd.DataFrame()\n",
    "df2 = pd.DataFrame()\n",
    "\n",
    "df1['score_{0}'.format(0)] = data1.cpu().numpy()\n",
    "df1['score_{0}'.format(1)] = data2.cpu().numpy()\n",
    "df1['score_{0}'.format(2)] = data3.cpu().numpy()\n",
    "r_data = torch.cat([data1, data2, data3], dim=0).cpu().numpy()\n",
    "df2['score_{0}'.format(2)] = r_data\n",
    "np.save(save_path+'/o_data',r_data)\n",
    "\n",
    "######Plot the distribution of original data\n",
    "fig, ax = plt.subplots(1, 1)\n",
    "for s in df1.columns:\n",
    "    df1[s].plot(kind='kde')\n",
    "\n",
    "for s in df2.columns:\n",
    "    df2[s].plot(style='--',kind='kde')\n",
    "plt.xlim((-4, 9 + distance * 2))\n",
    "ax.legend([\"Class_0\", \"Class_1\",\"Class_2\",\"Marginal\"])\n",
    "plt.title('Original')\n",
    "fig.savefig(save_path + '/original.eps')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Build Network"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "####TAC-GAN: set AC=True, MI=True, AC-GAN:set AC=True, MI=False, Projection-GAN: set AC=False, MI=False\n",
    "##Default TAC-GAN\n",
    "AC = True\n",
    "MI = True\n",
    "if AC:\n",
    "    if MI:\n",
    "        name = 'TAC'\n",
    "    else:\n",
    "        name = 'AC'\n",
    "else:\n",
    "    name = 'Projection'\n",
    "            \n",
    "        \n",
    "#####Bulding network\n",
    "nz = 2\n",
    "G = G_guassian(nz=nz, num_classes=3).cuda()\n",
    "\n",
    "D = D_guassian(num_classes=3,AC=AC).cuda()\n",
    "\n",
    "optg = optim.Adam(G.parameters(), lr=0.002,\n",
    "                      betas=(0.5, 0.999))\n",
    "optd = optim.Adam(D.parameters(), lr=0.002,\n",
    "                      betas=(0.5, 0.999))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training finished\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "#####Train network\n",
    "train(data1,data2,data3,nz,G,D,optd,optg,AC=AC,MI=MI)\n",
    "print('Training finished')\n",
    "\n",
    "#####Sample data from Generator\n",
    "z = torch.randn(10000, nz).cuda()\n",
    "label = torch.zeros(10000).long().cuda()  # torch.LongTensor(10000).random_(2).cuda()#\n",
    "data1_g = G(z=z, label=label).squeeze().cpu().detach()\n",
    "\n",
    "z = torch.randn(10000, nz).cuda()\n",
    "label = torch.ones(10000).long().cuda()  # torch.LongTensor(10000).random_(2).cuda()#\n",
    "data2_g = G(z=z, label=label).squeeze().cpu().detach()\n",
    "\n",
    "z = torch.randn(10000, nz).cuda()\n",
    "label = torch.ones(10000).long().cuda() + 1  # torch.LongTensor(10000).random_(2).cuda()#\n",
    "data3_g = G(z=z, label=label).squeeze().cpu().detach()\n",
    "\n",
    "df1 = pd.DataFrame()\n",
    "df2 = pd.DataFrame()\n",
    "\n",
    "df1['score_{0}'.format(0)] = data1_g.numpy()\n",
    "df1['score_{0}'.format(1)] = data2_g.numpy()\n",
    "df1['score_{0}'.format(2)] = data3_g.numpy()\n",
    "g_data = torch.cat([data1_g, data2_g, data3_g], dim=0).numpy()\n",
    "df2['score_{0}'.format(2)] = g_data\n",
    "\n",
    "#####Plot generated conditional distribution\n",
    "fig, ax = plt.subplots(1, 1)\n",
    "for s in df1.columns:\n",
    "    df1[s].plot(kind='kde')\n",
    "\n",
    "for s in df2.columns:\n",
    "    df2[s].plot(style='--',kind='kde')\n",
    "\n",
    "plt.xlim((-4, 9 + distance * 2))\n",
    "ax.legend([\"Class_0\", \"Class_1\", \"Class_2\", \"Marginal\"])\n",
    "plt.title(name)\n",
    "fig.savefig(save_path + '/'+name+'.eps')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Evaluate the MMD distance between true data and generated data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "the distance of class 0 is :  0.008205491237123663\n",
      "the distance of class 1 is :  1.5177659917671917\n",
      "the distance of class 2 is :  0.781029228938678\n",
      "the distance of marginal is :  -0.11618767918389494\n"
     ]
    }
   ],
   "source": [
    "\n",
    "mean0_0,var0_0 = polynomial_mmd(np.expand_dims(data1_g.numpy(), axis=1), np.expand_dims(data1.cpu().numpy(),axis=1))\n",
    "mean0_1, var0_1 = polynomial_mmd(np.expand_dims(data2_g.numpy(), axis=1),\n",
    "                                     np.expand_dims(data2.cpu().numpy(), axis=1))\n",
    "mean0_2, var0_2 = polynomial_mmd(np.expand_dims(data3_g.numpy(), axis=1),\n",
    "                                     np.expand_dims(data3.cpu().numpy(), axis=1))\n",
    "mean0, var0 = polynomial_mmd(np.expand_dims(g_data, axis=1), np.expand_dims(r_data, axis=1))\n",
    "print('the distance of class 0 is : ', mean0_0)\n",
    "print('the distance of class 1 is : ', mean0_1)\n",
    "print('the distance of class 2 is : ', mean0_2)\n",
    "print('the distance of marginal is : ', mean0)"
   ]
  },
  {
   "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": "3.7.4"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}