two_moons_ensemble.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import torch\n",
    "import torch.utils.data\n",
    "from torch import nn\n",
    "from torch.nn import functional as F\n",
    "\n",
    "from ignite.engine import Events, Engine\n",
    "from ignite.metrics import Accuracy, Loss\n",
    "\n",
    "import numpy as np\n",
    "import sklearn.datasets\n",
    "\n",
    "import matplotlib.pyplot as plt\n",
    "import seaborn as sns\n",
    "\n",
    "sns.set()\n",
    "\n",
    "torch.manual_seed(1)\n",
    "np.random.seed(1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "class Model(nn.Module):\n",
    "    def __init__(self, features):\n",
    "        super().__init__()\n",
    "        \n",
    "        self.fc1 = nn.Linear(2, features)\n",
    "        self.fc2 = nn.Linear(features, features)\n",
    "        self.fc3 = nn.Linear(features, features)\n",
    "        self.fc4 = nn.Linear(features, 2)\n",
    "\n",
    "    def forward(self, x):\n",
    "        x = F.relu(self.fc1(x))\n",
    "        x = F.relu(self.fc2(x))\n",
    "        x = F.relu(self.fc3(x))\n",
    "        x = self.fc4(x)\n",
    "        \n",
    "        return F.log_softmax(x, dim=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "noise = 0.1\n",
    "\n",
    "X_train, y_train = sklearn.datasets.make_moons(n_samples=1000, noise=noise)\n",
    "X_test, y_test = sklearn.datasets.make_moons(n_samples=200, noise=noise)\n",
    "\n",
    "num_classes = 2\n",
    "batch_size = 64\n",
    "\n",
    "def train_model(max_epochs):\n",
    "    model = Model(20)\n",
    "\n",
    "    optimizer = torch.optim.SGD(model.parameters(), lr=0.01, momentum=0.9, weight_decay=1e-4)\n",
    "\n",
    "    def step(engine, batch):\n",
    "        model.train()\n",
    "        optimizer.zero_grad()\n",
    "\n",
    "        x, y = batch\n",
    "\n",
    "        y_pred = model(x)\n",
    "        loss =  F.nll_loss(y_pred, y)\n",
    "\n",
    "        loss.backward()\n",
    "        optimizer.step()\n",
    "\n",
    "        return loss.item()\n",
    "\n",
    "    def eval_step(engine, batch):\n",
    "        model.eval()\n",
    "\n",
    "        x, y = batch\n",
    "        y_pred = model(x)\n",
    "\n",
    "        return y_pred, y\n",
    "\n",
    "\n",
    "    trainer = Engine(step)\n",
    "    evaluator = Engine(eval_step)\n",
    "\n",
    "    metric = Accuracy()\n",
    "    metric.attach(evaluator, \"accuracy\")\n",
    "\n",
    "    metric = Loss(F.nll_loss)\n",
    "    metric.attach(evaluator, \"nll\")\n",
    "\n",
    "    ds_train = torch.utils.data.TensorDataset(torch.from_numpy(X_train).float(), torch.from_numpy(y_train))\n",
    "    dl_train = torch.utils.data.DataLoader(ds_train, batch_size=batch_size, shuffle=True, drop_last=True)\n",
    "\n",
    "    ds_test = torch.utils.data.TensorDataset(torch.from_numpy(X_test).float(), torch.from_numpy(y_test))\n",
    "    dl_test = torch.utils.data.DataLoader(ds_test, batch_size=200, shuffle=False)\n",
    "\n",
    "    @trainer.on(Events.EPOCH_COMPLETED)\n",
    "    def log_results(trainer):\n",
    "        evaluator.run(dl_test)\n",
    "        metrics = evaluator.state.metrics\n",
    "\n",
    "        print(f\"Test Results - Epoch: {trainer.state.epoch} Acc: {metrics['accuracy']:.4f} NLL: {metrics['nll']:.2f}\")\n",
    "    \n",
    "    trainer.run(dl_train, max_epochs=max_epochs)\n",
    "    \n",
    "    return model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "scrolled": false
   },
   "outputs": [],
   "source": [
    "ensemble = 5\n",
    "models = [train_model(50) for _ in range(ensemble)]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "domain = 3\n",
    "x = np.linspace(-domain+0.5, domain+0.5, 100)\n",
    "y = np.linspace(-domain, domain, 100)\n",
    "\n",
    "xx, yy = np.meshgrid(x, y)\n",
    "\n",
    "X = np.column_stack([xx.flatten(), yy.flatten()])\n",
    "\n",
    "X_vis, y_vis = sklearn.datasets.make_moons(n_samples=500, noise=noise)\n",
    "mask = y_vis.astype(np.bool)\n",
    "\n",
    "for model in models:\n",
    "    model.eval()\n",
    "\n",
    "with torch.no_grad():\n",
    "    predictions = torch.stack([model(torch.from_numpy(X).float()) for model in models])\n",
    "\n",
    "    mean_prediction = torch.mean(predictions.exp(), dim=0)\n",
    "    confidence = torch.sum(mean_prediction * torch.log(mean_prediction), dim=1)\n",
    "\n",
    "z = confidence.reshape(xx.shape)\n",
    "\n",
    "plt.figure()\n",
    "plt.contourf(x, y, z, cmap='cividis')\n",
    "\n",
    "plt.scatter(X_vis[mask,0], X_vis[mask,1])\n",
    "plt.scatter(X_vis[~mask,0], X_vis[~mask,1])\n",
    "\n",
    "plt.figure()\n",
    "plt.contourf(x, y, z, cmap='cividis')"
   ]
  }
 ],
 "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": 2
}