{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## ART BlackBox Classifier - Creating adversarial samples for remote classifiers"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "This demo consists of two parts. The first part deploys a trained model using IBM Watson Machine Learning service to create an API. The second connects to this endpoint and attacks the deployed model with the HopSkipJump attack using only the model predictions."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "from __future__ import absolute_import, division, print_function, unicode_literals\n",
    "\n",
    "import numpy as np\n",
    "from IPython.display import clear_output\n",
    "from matplotlib import pyplot as plt\n",
    "\n",
    "from art import config\n",
    "from art.estimators.classification import BlackBoxClassifier\n",
    "from art.attacks.evasion import HopSkipJump\n",
    "from art.utils import to_categorical\n",
    "from art.utils import load_dataset, get_file, compute_accuracy"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Deploy model and connect to API"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "To deploy a model and connect to the endpoint for model predictions, please follow the demo:\n",
    "[Use TensorFlow to predict handwritten digits](https://dataplatform.cloud.ibm.com/analytics/notebooks/v2/3bd3efb8-833d-460f-b07b-fee51dd0f1af/view?access_token=6bd0ff8d807861d09e0dab0cad28ce9685711078f612fcd92bb8cf8535d089c1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# generate wml client object following the tutorial above (part 1)\n",
    "client = 'TODO'\n",
    "\n",
    "# generate api endpoint following the tutorial above (part 6)\n",
    "scoring_url = 'TODO'"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Make predictions"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We show an example of a predict function below that connects to a wml endpoint and returns the predictions of the deployed model. A different function could be written for another remote classifier that outputs predictions."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# Read MNIST dataset\n",
    "(x_train, y_train), (x_test, y_test), min_, max_ = load_dataset(str('mnist'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# Sample predict function that reformats inputs, connects to wml scoring endpoint and \n",
    "# returns one-hot encoded predictions\n",
    "def predict(x):\n",
    "    x = np.array(x)\n",
    "    scoring_data = {'values': (np.reshape(x, (x.shape[0],784))).tolist()}\n",
    "    predictions = client.deployments.score(scoring_url, scoring_data)\n",
    "    return to_categorical(predictions['values'], nb_classes=10)\n",
    "    \n",
    "# Create blackbox object\n",
    "classifier = BlackBoxClassifier(predict, x_train[0].shape, 10, clip_values=(0, 255))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x1829ec52e8>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Prediction from API is: 5\n"
     ]
    }
   ],
   "source": [
    "# Select target image and show prediction\n",
    "target_image = x_train[0]\n",
    "plt.imshow(np.reshape(target_image.astype(np.float32), (28, 28)))\n",
    "plt.show(block=False)\n",
    "print('Prediction from API is: ' + str(np.argmax(classifier.predict(x_train[:1]), axis=1)[0]))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Generate black box HopSkipJump attack"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Adversarial image at step 0. L2 error 7.9944905692156425 and class label 3.\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x1825477748>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Adversarial image at step 10. L2 error 1.4959683890579352 and class label 3.\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x1829f85b70>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Adversarial image at step 20. L2 error 1.0073771113688348 and class label 3.\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x1851838b70>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Generate HopSkipJump attack against black box classifier\n",
    "attack = HopSkipJump(classifier=classifier, targeted=False, max_iter=0, max_eval=1000, init_eval=10)\n",
    "iter_step = 10\n",
    "x_adv = None\n",
    "for i in range(3):\n",
    "    x_adv = attack.generate(x=np.array([target_image]), x_adv_init=x_adv)\n",
    "    \n",
    "    print(\"Adversarial image at step %d.\" % (i * iter_step), \"L2 error\", \n",
    "          np.linalg.norm(np.reshape(x_adv[0] - target_image, [-1])),\n",
    "          \"and class label %d.\" % np.argmax(classifier.predict(x_adv)[0]))\n",
    "    plt.imshow(np.reshape(x_adv[0].astype(np.float32), (28, 28)))\n",
    "    plt.show(block=False)\n",
    "    \n",
    "    attack.max_iter = iter_step"
   ]
  }
 ],
 "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": "3.7.7"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}