report.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "from main_train import *\n",
    "from model.SimpleANN import *\n",
    "import adversary"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "For this experiment was run two models to see advantage of maximizing lipschitz margin against adversary attacks (for this case was tested against gaussian noise)<br>\n",
    "Both models have the same structure -- simply 3 fully connected layers with ReLu as activation.<br>\n",
    "The first network was optimizing cross etropy loss function while the second model was optiziming cross entropy - margin_rate * margin loss.<br> The margin loss is defined next:<br>\n",
    "$$\\vec{y} = softmax(F(\\vec{x}))$$\n",
    "$$\\mathcal{L}_{lipschitz\\ margin} = \\frac{1}{N} \\sum_{\\vec{y} \\in \\mathcal{C}}(max(\\vec{y}) - max(\\vec{y}_{/{max(\\vec{y})})}) : \\mathcal{C} = \\{\\vec{y_i} : argmax(\\vec{y_i}) = argmax(\\vec{\\hat{y_i}})\\}$$\n",
    "In another words, munis sum of differences first 2 choices of network, taking to account only examples correctly predicted."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "n_dim = 784\n",
    "y_dim = 10\n",
    "trained_marginless = 'trained_models/margin=0/model.ckpt'\n",
    "trained_margin = 'trained_models/margin=1/model.ckpt'"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Run training for both models with equal hyperparameters<br>\n",
    "$\\lambda = 10^{-4}$<br>\n",
    "$batch\\_size = 64$<br>\n",
    "$steps = 10\\ 000$<br>\n",
    "For model including margin:<br>\n",
    "$margin\\_rate = 1.0$"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/artyiom/.local/lib/python3.6/site-packages/tensorflow/python/ops/gradients_impl.py:100: UserWarning: Converting sparse IndexedSlices to a dense Tensor of unknown shape. This may consume a large amount of memory.\n",
      "  \"Converting sparse IndexedSlices to a dense Tensor of unknown shape. \"\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "WARNING:tensorflow:From /home/artyiom/Desktop/Artyom/notebook/independent/ml/lipschitz_margin/main_train.py:23: read_data_sets (from tensorflow.contrib.learn.python.learn.datasets.mnist) is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "Please use alternatives such as official/mnist/dataset.py from tensorflow/models.\n",
      "WARNING:tensorflow:From /home/artyiom/.local/lib/python3.6/site-packages/tensorflow/contrib/learn/python/learn/datasets/mnist.py:260: maybe_download (from tensorflow.contrib.learn.python.learn.datasets.base) is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "Please write your own downloading logic.\n",
      "WARNING:tensorflow:From /home/artyiom/.local/lib/python3.6/site-packages/tensorflow/contrib/learn/python/learn/datasets/mnist.py:262: extract_images (from tensorflow.contrib.learn.python.learn.datasets.mnist) is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "Please use tf.data to implement this functionality.\n",
      "Extracting MNIST_data/train-images-idx3-ubyte.gz\n",
      "WARNING:tensorflow:From /home/artyiom/.local/lib/python3.6/site-packages/tensorflow/contrib/learn/python/learn/datasets/mnist.py:267: extract_labels (from tensorflow.contrib.learn.python.learn.datasets.mnist) is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "Please use tf.data to implement this functionality.\n",
      "Extracting MNIST_data/train-labels-idx1-ubyte.gz\n",
      "WARNING:tensorflow:From /home/artyiom/.local/lib/python3.6/site-packages/tensorflow/contrib/learn/python/learn/datasets/mnist.py:110: dense_to_one_hot (from tensorflow.contrib.learn.python.learn.datasets.mnist) is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "Please use tf.one_hot on tensors.\n",
      "Extracting MNIST_data/t10k-images-idx3-ubyte.gz\n",
      "Extracting MNIST_data/t10k-labels-idx1-ubyte.gz\n",
      "WARNING:tensorflow:From /home/artyiom/.local/lib/python3.6/site-packages/tensorflow/contrib/learn/python/learn/datasets/mnist.py:290: DataSet.__init__ (from tensorflow.contrib.learn.python.learn.datasets.mnist) is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "Please use alternatives such as official/mnist/dataset.py from tensorflow/models.\n",
      "t = 200, accuracy = 0.638672, loss = 0.745390\n",
      "t = 400, accuracy = 0.707031, loss = 0.455628\n",
      "t = 600, accuracy = 0.757812, loss = 0.143119\n",
      "t = 800, accuracy = 0.812500, loss = -0.005051\n",
      "t = 1000, accuracy = 0.835938, loss = -0.154094\n",
      "t = 1200, accuracy = 0.873047, loss = -0.240318\n",
      "t = 1400, accuracy = 0.851562, loss = -0.265323\n",
      "t = 1600, accuracy = 0.867188, loss = -0.275718\n",
      "t = 1800, accuracy = 0.871094, loss = -0.254630\n",
      "t = 2000, accuracy = 0.886719, loss = -0.366339\n",
      "t = 2200, accuracy = 0.890625, loss = -0.393514\n",
      "t = 2400, accuracy = 0.880859, loss = -0.364929\n",
      "t = 2600, accuracy = 0.908203, loss = -0.499407\n",
      "t = 2800, accuracy = 0.898438, loss = -0.429218\n",
      "t = 3000, accuracy = 0.896484, loss = -0.386158\n",
      "t = 3200, accuracy = 0.912109, loss = -0.501732\n",
      "t = 3400, accuracy = 0.923828, loss = -0.531738\n",
      "t = 3600, accuracy = 0.914062, loss = -0.509543\n",
      "t = 3800, accuracy = 0.929688, loss = -0.625197\n",
      "t = 4000, accuracy = 0.929688, loss = -0.572174\n",
      "t = 4200, accuracy = 0.939453, loss = -0.683048\n",
      "t = 4400, accuracy = 0.937500, loss = -0.649494\n",
      "t = 4600, accuracy = 0.943359, loss = -0.653604\n",
      "t = 4800, accuracy = 0.927734, loss = -0.586159\n",
      "t = 5000, accuracy = 0.925781, loss = -0.617655\n",
      "t = 5200, accuracy = 0.921875, loss = -0.577740\n",
      "t = 5400, accuracy = 0.943359, loss = -0.634732\n",
      "t = 5600, accuracy = 0.947266, loss = -0.698838\n",
      "t = 5800, accuracy = 0.945312, loss = -0.688558\n",
      "t = 6000, accuracy = 0.943359, loss = -0.654985\n",
      "t = 6200, accuracy = 0.947266, loss = -0.737660\n",
      "t = 6400, accuracy = 0.953125, loss = -0.735257\n",
      "t = 6600, accuracy = 0.962891, loss = -0.716830\n",
      "t = 6800, accuracy = 0.943359, loss = -0.657949\n",
      "t = 7000, accuracy = 0.962891, loss = -0.765028\n",
      "t = 7200, accuracy = 0.951172, loss = -0.738926\n",
      "t = 7400, accuracy = 0.951172, loss = -0.749412\n",
      "t = 7600, accuracy = 0.955078, loss = -0.777392\n",
      "t = 7800, accuracy = 0.978516, loss = -0.850462\n",
      "t = 8000, accuracy = 0.947266, loss = -0.717846\n",
      "t = 8200, accuracy = 0.951172, loss = -0.758928\n",
      "t = 8400, accuracy = 0.949219, loss = -0.743700\n",
      "t = 8600, accuracy = 0.962891, loss = -0.776927\n",
      "t = 8800, accuracy = 0.958984, loss = -0.796105\n",
      "t = 9000, accuracy = 0.972656, loss = -0.820073\n",
      "t = 9200, accuracy = 0.957031, loss = -0.778501\n",
      "t = 9400, accuracy = 0.953125, loss = -0.726253\n",
      "t = 9600, accuracy = 0.962891, loss = -0.829168\n",
      "t = 9800, accuracy = 0.957031, loss = -0.772785\n",
      "t = 10000, accuracy = 0.960938, loss = -0.798587\n",
      "Time of training: 141.062\n",
      "Extracting MNIST_data/train-images-idx3-ubyte.gz\n",
      "Extracting MNIST_data/train-labels-idx1-ubyte.gz\n",
      "Extracting MNIST_data/t10k-images-idx3-ubyte.gz\n",
      "Extracting MNIST_data/t10k-labels-idx1-ubyte.gz\n",
      "t = 200, accuracy = 0.292969, loss = 1.584225\n",
      "t = 400, accuracy = 0.488281, loss = 1.207786\n",
      "t = 600, accuracy = 0.697266, loss = 0.866318\n",
      "t = 800, accuracy = 0.710938, loss = 0.890208\n",
      "t = 1000, accuracy = 0.726562, loss = 0.837466\n",
      "t = 1200, accuracy = 0.750000, loss = 0.738783\n",
      "t = 1400, accuracy = 0.802734, loss = 0.606250\n",
      "t = 1600, accuracy = 0.792969, loss = 0.705499\n",
      "t = 1800, accuracy = 0.806641, loss = 0.604901\n",
      "t = 2000, accuracy = 0.835938, loss = 0.553375\n",
      "t = 2200, accuracy = 0.851562, loss = 0.472567\n",
      "t = 2400, accuracy = 0.867188, loss = 0.429112\n",
      "t = 2600, accuracy = 0.884766, loss = 0.384463\n",
      "t = 2800, accuracy = 0.894531, loss = 0.369352\n",
      "t = 3000, accuracy = 0.910156, loss = 0.278552\n",
      "t = 3200, accuracy = 0.894531, loss = 0.351131\n",
      "t = 3400, accuracy = 0.916016, loss = 0.266925\n",
      "t = 3600, accuracy = 0.925781, loss = 0.275025\n",
      "t = 3800, accuracy = 0.933594, loss = 0.224382\n",
      "t = 4000, accuracy = 0.943359, loss = 0.216474\n",
      "t = 4200, accuracy = 0.935547, loss = 0.220935\n",
      "t = 4400, accuracy = 0.939453, loss = 0.174492\n",
      "t = 4600, accuracy = 0.960938, loss = 0.159795\n",
      "t = 4800, accuracy = 0.927734, loss = 0.238280\n",
      "t = 5000, accuracy = 0.939453, loss = 0.217402\n",
      "t = 5200, accuracy = 0.941406, loss = 0.277139\n",
      "t = 5400, accuracy = 0.933594, loss = 0.243811\n",
      "t = 5600, accuracy = 0.949219, loss = 0.164599\n",
      "t = 5800, accuracy = 0.943359, loss = 0.179968\n",
      "t = 6000, accuracy = 0.957031, loss = 0.138097\n",
      "t = 6200, accuracy = 0.953125, loss = 0.141990\n",
      "t = 6400, accuracy = 0.943359, loss = 0.177666\n",
      "t = 6600, accuracy = 0.933594, loss = 0.211562\n",
      "t = 6800, accuracy = 0.941406, loss = 0.216890\n",
      "t = 7000, accuracy = 0.935547, loss = 0.201629\n",
      "t = 7200, accuracy = 0.960938, loss = 0.137131\n",
      "t = 7400, accuracy = 0.958984, loss = 0.137099\n",
      "t = 7600, accuracy = 0.957031, loss = 0.167829\n",
      "t = 7800, accuracy = 0.966797, loss = 0.138541\n",
      "t = 8000, accuracy = 0.947266, loss = 0.181392\n",
      "t = 8200, accuracy = 0.958984, loss = 0.142004\n",
      "t = 8400, accuracy = 0.958984, loss = 0.110111\n",
      "t = 8600, accuracy = 0.947266, loss = 0.172718\n",
      "t = 8800, accuracy = 0.957031, loss = 0.197608\n",
      "t = 9000, accuracy = 0.949219, loss = 0.140449\n",
      "t = 9200, accuracy = 0.974609, loss = 0.094949\n",
      "t = 9400, accuracy = 0.962891, loss = 0.116764\n",
      "t = 9600, accuracy = 0.974609, loss = 0.100090\n",
      "t = 9800, accuracy = 0.962891, loss = 0.111772\n",
      "t = 10000, accuracy = 0.958984, loss = 0.129683\n",
      "Time of training: 137.043\n"
     ]
    }
   ],
   "source": [
    "margin_statistics, marginless_statistics = train_models()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Load models trained models"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/artyiom/.local/lib/python3.6/site-packages/tensorflow/python/ops/gradients_impl.py:100: UserWarning: Converting sparse IndexedSlices to a dense Tensor of unknown shape. This may consume a large amount of memory.\n",
      "  \"Converting sparse IndexedSlices to a dense Tensor of unknown shape. \"\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "INFO:tensorflow:Restoring parameters from trained_models/margin=0/model.ckpt\n",
      "INFO:tensorflow:Restoring parameters from trained_models/margin=1/model.ckpt\n"
     ]
    }
   ],
   "source": [
    "with tf.Graph().as_default():\n",
    "    model_marginless = SimpleANN(n_dim=n_dim, y_dim=y_dim, margin_rate=0.0, source=trained_marginless)\n",
    "with tf.Graph().as_default():\n",
    "    model_margin = SimpleANN(n_dim=n_dim, y_dim=y_dim, margin_rate=1.0, source=trained_margin)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.legend.Legend at 0x7fa77419fdd8>"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fa7517dd8d0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.ylabel(\"Accuracy, %\")\n",
    "plt.xlabel(\"iteration, k\")\n",
    "plt.xscale('log')\n",
    "plt.plot(np.array(marginless_statistics[2]) / 1000, np.array(marginless_statistics[0]) * 100, label=\"w/out margin\")\n",
    "plt.plot(np.array(margin_statistics[2]) / 1000, np.array(margin_statistics[0]) * 100, label=\"w/ margin\")\n",
    "plt.legend(bbox_to_anchor=(1.05, 1), loc=2, borderaxespad=0.)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.legend.Legend at 0x7fa774073da0>"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fa77404fd30>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.ylabel(\"loss\")\n",
    "plt.xlabel(\"iteration, k\")\n",
    "plt.xscale('log')\n",
    "plt.plot(np.array(marginless_statistics[2]) / 1000, np.array(marginless_statistics[1]), label=\"w/out margin\")\n",
    "plt.plot(np.array(margin_statistics[2]) / 1000, np.array(margin_statistics[1]), label=\"w/ margin\")\n",
    "plt.legend(bbox_to_anchor=(1.05, 1), loc=2, borderaxespad=0.)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Extracting MNIST_data/train-images-idx3-ubyte.gz\n",
      "Extracting MNIST_data/train-labels-idx1-ubyte.gz\n",
      "Extracting MNIST_data/t10k-images-idx3-ubyte.gz\n",
      "Extracting MNIST_data/t10k-labels-idx1-ubyte.gz\n"
     ]
    }
   ],
   "source": [
    "dataset = read_data_sets('MNIST_data', one_hot=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Example of trained models with small test set"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'accuracy': 0.96484375, 'loss': 0.13771212}"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "adversary.test(model_marginless, dataset=dataset, batch_size=1024, noise=0.0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'accuracy': 0.966796875, 'loss': -0.7701862}"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "adversary.test(model_margin, dataset=dataset, batch_size=1024, noise=0.0)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Run pairwise test with different noise level (from 0 to 1, complete random image).\n",
    "Plot below shows accuracy of the model on images with given noise."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fa7755b4208>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "results = adversary.pair_test(model_margin, model_marginless, dataset, 2**10)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Depends on hyperparameters, the model with margin shows from no improvement against gaussian noise to up 40% improvements. Was noticed examples where with noise of 40% simple model performed with 40% accuracy while model trained with margin still performed with accuracy >95%.<br>\n",
    "Result is very unstable vary from up to 40% improvement to not at all."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.legend.Legend at 0x7fa75152f7f0>"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fa753779048>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# plot difference graph\n",
    "diff = (np.array(results[0]) - np.array(results[1])) / np.array(results[0])\n",
    "plt.xlabel(\"gaussian noise\")\n",
    "plt.ylabel(\"Margin - marginless. accuracy %\")\n",
    "plt.plot(np.arange(0.0, 1.0, 0.01), diff, label=\"margin accuracy dominance\")\n",
    "plt.legend(bbox_to_anchor=(1.05, 1), loc=2, borderaxespad=0.)"
   ]
  }
 ],
 "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": "3.6.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}