nips2016/mnist.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [],
   "source": [
    "%load_ext autoreload\n",
    "%autoreload 2\n",
    "\n",
    "import sys, os\n",
    "sys.path.insert(0, '..')\n",
    "from lib import models, graph, coarsening, utils\n",
    "\n",
    "import tensorflow as tf\n",
    "import numpy as np\n",
    "import time\n",
    "\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [],
   "source": [
    "flags = tf.app.flags\n",
    "FLAGS = flags.FLAGS\n",
    "\n",
    "# Graphs.\n",
    "flags.DEFINE_integer('number_edges', 8, 'Graph: minimum number of edges per vertex.')\n",
    "flags.DEFINE_string('metric', 'euclidean', 'Graph: similarity measure (between features).')\n",
    "# TODO: change cgcnn for combinatorial Laplacians.\n",
    "flags.DEFINE_bool('normalized_laplacian', True, 'Graph Laplacian: normalized.')\n",
    "flags.DEFINE_integer('coarsening_levels', 4, 'Number of coarsened graphs.')\n",
    "\n",
    "# Directories.\n",
    "flags.DEFINE_string('dir_data', os.path.join('..', 'data', 'mnist'), 'Directory to store data.')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "# Feature graph"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "3198 > 3136 edges\n",
      "Layer 0: M_0 = |V| = 960 nodes (176 added),|E| = 3198 edges\n",
      "Layer 1: M_1 = |V| = 480 nodes (67 added),|E| = 1444 edges\n",
      "Layer 2: M_2 = |V| = 240 nodes (23 added),|E| = 678 edges\n",
      "Layer 3: M_3 = |V| = 120 nodes (6 added),|E| = 322 edges\n",
      "Layer 4: M_4 = |V| = 60 nodes (0 added),|E| = 156 edges\n",
      "Execution time: 0.25s\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAA+MAAAEyCAYAAACReQFzAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nVmL/A4+Qd9qo4t6xAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f4040b1d6d8>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "def grid_graph(m, corners=False):\n",
    "    z = graph.grid(m)\n",
    "    dist, idx = graph.distance_sklearn_metrics(z, k=FLAGS.number_edges, metric=FLAGS.metric)\n",
    "    A = graph.adjacency(dist, idx)\n",
    "\n",
    "    # Connections are only vertical or horizontal on the grid.\n",
    "    # Corner vertices are connected to 2 neightbors only.\n",
    "    if corners:\n",
    "        import scipy.sparse\n",
    "        A = A.toarray()\n",
    "        A[A < A.max()/1.5] = 0\n",
    "        A = scipy.sparse.csr_matrix(A)\n",
    "        print('{} edges'.format(A.nnz))\n",
    "\n",
    "    print(\"{} > {} edges\".format(A.nnz//2, FLAGS.number_edges*m**2//2))\n",
    "    return A\n",
    "\n",
    "t_start = time.process_time()\n",
    "A = grid_graph(28, corners=False)\n",
    "A = graph.replace_random_edges(A, 0)\n",
    "graphs, perm = coarsening.coarsen(A, levels=FLAGS.coarsening_levels, self_connections=False)\n",
    "L = [graph.laplacian(A, normalized=True) for A in graphs]\n",
    "print('Execution time: {:.2f}s'.format(time.process_time() - t_start))\n",
    "graph.plot_spectrum(L)\n",
    "del A"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "# Data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Extracting ../data/mnist/train-images-idx3-ubyte.gz\n",
      "Extracting ../data/mnist/train-labels-idx1-ubyte.gz\n",
      "Extracting ../data/mnist/t10k-images-idx3-ubyte.gz\n",
      "Extracting ../data/mnist/t10k-labels-idx1-ubyte.gz\n",
      "(10000, 784)\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAP8AAAD8CAYAAAC4nHJkAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f402948e080>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAP8AAAD8CAYAAAC4nHJkAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f40293f2e80>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(10000, 28, 28)\n",
      "Execution time: 1.93s\n"
     ]
    }
   ],
   "source": [
    "from tensorflow.examples.tutorials.mnist import input_data\n",
    "import matplotlib.pyplot as plt\n",
    "mnist = input_data.read_data_sets(FLAGS.dir_data, one_hot=False)\n",
    "\n",
    "train_data = mnist.train.images.astype(np.float32)\n",
    "val_data = mnist.validation.images.astype(np.float32)\n",
    "test_data = mnist.test.images.astype(np.float32)\n",
    "\n",
    "#### TUTAJ TWORZĘ OBRÓCONY DATASET\n",
    "print(test_data.shape)\n",
    "test_transposed = test_data.reshape((10000,28,28))\n",
    "plt.imshow(test_transposed[0])\n",
    "plt.show()\n",
    "test_transposed = np.rot90(test_transposed, axes = (1,2))\n",
    "plt.imshow(test_transposed[0])\n",
    "plt.show()\n",
    "print(test_transposed.shape)\n",
    "test_transposed = test_transposed.reshape((10000,784))\n",
    "\n",
    "### TU KOŃCZE GO TWORZYĆ\n",
    "train_labels = mnist.train.labels\n",
    "val_labels = mnist.validation.labels\n",
    "test_labels = mnist.test.labels\n",
    "\n",
    "t_start = time.process_time()\n",
    "train_data = coarsening.perm_data(train_data, perm)\n",
    "val_data = coarsening.perm_data(val_data, perm)\n",
    "test_data = coarsening.perm_data(test_data, perm)\n",
    "test_transposed = coarsening.perm_data(test_transposed, perm)\n",
    "\n",
    "print('Execution time: {:.2f}s'.format(time.process_time() - t_start))\n",
    "del perm\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "# Neural networks"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "collapsed": true,
    "deletable": true,
    "editable": true
   },
   "outputs": [],
   "source": [
    "#model = fc1()\n",
    "#model = fc2(nhiddens=100)\n",
    "#model = cnn2(K=5, F=10)  # K=28 is equivalent to filtering with fgcnn.\n",
    "#model = fcnn2(F=10)\n",
    "#model = fgcnn2(L[0], F=10)\n",
    "#model = lgcnn2_2(L[0], F=10, K=10)\n",
    "#model = cgcnn2_3(L[0], F=10, K=5)\n",
    "#model = cgcnn2_4(L[0], F=10, K=5)\n",
    "#model = cgcnn2_5(L[0], F=10, K=5)\n",
    "\n",
    "if False:\n",
    "    K = 5  # 5 or 5^2\n",
    "    t_start = time.process_time()\n",
    "    mnist.test._images = graph.lanczos(L, mnist.test._images.T, K).T\n",
    "    mnist.train._images = graph.lanczos(L, mnist.train._images.T, K).T\n",
    "    model = lgcnn2_1(L, F=10, K=K)\n",
    "    print('Execution time: {:.2f}s'.format(time.process_time() - t_start))\n",
    "    ph_data = tf.placeholder(tf.float32, (FLAGS.batch_size, mnist.train.images.shape[1], K), 'data')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [],
   "source": [
    "common = {}\n",
    "common['dir_name']       = 'mnist/'\n",
    "common['num_epochs']     = 20\n",
    "common['batch_size']     = 100\n",
    "common['decay_steps']    = mnist.train.num_examples / common['batch_size']\n",
    "common['eval_frequency'] = 30 * common['num_epochs']\n",
    "common['brelu']          = 'b1relu'\n",
    "common['pool']           = 'mpool1'\n",
    "C = max(mnist.train.labels) + 1  # number of classes\n",
    "\n",
    "model_perf = utils.model_perf()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "NN architecture\n",
      "  input: M_0 = 960\n",
      "  layer 1: logits (softmax)\n",
      "    representation: M_1 = 10\n",
      "    weights: M_0 * M_1 = 960 * 10 = 9600\n",
      "    biases: M_1 = 10\n",
      "WARNING:tensorflow:From ../lib/models.py:277 in _weight_variable.: histogram_summary (from tensorflow.python.ops.logging_ops) is deprecated and will be removed after 2016-11-30.\n",
      "Instructions for updating:\n",
      "Please switch to tf.summary.histogram. Note that tf.summary.histogram uses the node name instead of the tag. This means that TensorFlow will automatically de-duplicate summary names based on their scope.\n",
      "WARNING:tensorflow:From ../lib/models.py:285 in _bias_variable.: histogram_summary (from tensorflow.python.ops.logging_ops) is deprecated and will be removed after 2016-11-30.\n",
      "Instructions for updating:\n",
      "Please switch to tf.summary.histogram. Note that tf.summary.histogram uses the node name instead of the tag. This means that TensorFlow will automatically de-duplicate summary names based on their scope.\n",
      "WARNING:tensorflow:From ../lib/models.py:217 in loss.: scalar_summary (from tensorflow.python.ops.logging_ops) is deprecated and will be removed after 2016-11-30.\n",
      "Instructions for updating:\n",
      "Please switch to tf.summary.scalar. Note that tf.summary.scalar uses the node name instead of the tag. This means that TensorFlow will automatically de-duplicate summary names based on the scope they are created in. Also, passing a tensor or list of tags to a scalar summary op is no longer supported.\n",
      "WARNING:tensorflow:From ../lib/models.py:218 in loss.: scalar_summary (from tensorflow.python.ops.logging_ops) is deprecated and will be removed after 2016-11-30.\n",
      "Instructions for updating:\n",
      "Please switch to tf.summary.scalar. Note that tf.summary.scalar uses the node name instead of the tag. This means that TensorFlow will automatically de-duplicate summary names based on the scope they are created in. Also, passing a tensor or list of tags to a scalar summary op is no longer supported.\n",
      "WARNING:tensorflow:From ../lib/models.py:219 in loss.: scalar_summary (from tensorflow.python.ops.logging_ops) is deprecated and will be removed after 2016-11-30.\n",
      "Instructions for updating:\n",
      "Please switch to tf.summary.scalar. Note that tf.summary.scalar uses the node name instead of the tag. This means that TensorFlow will automatically de-duplicate summary names based on the scope they are created in. Also, passing a tensor or list of tags to a scalar summary op is no longer supported.\n",
      "WARNING:tensorflow:From ../lib/models.py:223 in loss.: scalar_summary (from tensorflow.python.ops.logging_ops) is deprecated and will be removed after 2016-11-30.\n",
      "Instructions for updating:\n",
      "Please switch to tf.summary.scalar. Note that tf.summary.scalar uses the node name instead of the tag. This means that TensorFlow will automatically de-duplicate summary names based on the scope they are created in. Also, passing a tensor or list of tags to a scalar summary op is no longer supported.\n",
      "WARNING:tensorflow:From ../lib/models.py:224 in loss.: scalar_summary (from tensorflow.python.ops.logging_ops) is deprecated and will be removed after 2016-11-30.\n",
      "Instructions for updating:\n",
      "Please switch to tf.summary.scalar. Note that tf.summary.scalar uses the node name instead of the tag. This means that TensorFlow will automatically de-duplicate summary names based on the scope they are created in. Also, passing a tensor or list of tags to a scalar summary op is no longer supported.\n",
      "WARNING:tensorflow:From ../lib/models.py:225 in loss.: scalar_summary (from tensorflow.python.ops.logging_ops) is deprecated and will be removed after 2016-11-30.\n",
      "Instructions for updating:\n",
      "Please switch to tf.summary.scalar. Note that tf.summary.scalar uses the node name instead of the tag. This means that TensorFlow will automatically de-duplicate summary names based on the scope they are created in. Also, passing a tensor or list of tags to a scalar summary op is no longer supported.\n",
      "WARNING:tensorflow:From ../lib/models.py:238 in training.: scalar_summary (from tensorflow.python.ops.logging_ops) is deprecated and will be removed after 2016-11-30.\n",
      "Instructions for updating:\n",
      "Please switch to tf.summary.scalar. Note that tf.summary.scalar uses the node name instead of the tag. This means that TensorFlow will automatically de-duplicate summary names based on the scope they are created in. Also, passing a tensor or list of tags to a scalar summary op is no longer supported.\n",
      "WARNING:tensorflow:From ../lib/models.py:252 in training.: histogram_summary (from tensorflow.python.ops.logging_ops) is deprecated and will be removed after 2016-11-30.\n",
      "Instructions for updating:\n",
      "Please switch to tf.summary.histogram. Note that tf.summary.histogram uses the node name instead of the tag. This means that TensorFlow will automatically de-duplicate summary names based on their scope.\n",
      "WARNING:tensorflow:From ../lib/models.py:252 in training.: histogram_summary (from tensorflow.python.ops.logging_ops) is deprecated and will be removed after 2016-11-30.\n",
      "Instructions for updating:\n",
      "Please switch to tf.summary.histogram. Note that tf.summary.histogram uses the node name instead of the tag. This means that TensorFlow will automatically de-duplicate summary names based on their scope.\n",
      "WARNING:tensorflow:From ../lib/models.py:168 in build_graph.: initialize_all_variables (from tensorflow.python.ops.variables) is deprecated and will be removed after 2017-03-02.\n",
      "Instructions for updating:\n",
      "Use `tf.global_variables_initializer` instead.\n",
      "WARNING:tensorflow:From ../lib/models.py:171 in build_graph.: merge_all_summaries (from tensorflow.python.ops.logging_ops) is deprecated and will be removed after 2016-11-30.\n",
      "Instructions for updating:\n",
      "Please switch to tf.summary.merge_all.\n",
      "WARNING:tensorflow:From /usr/local/lib/python3.5/dist-packages/tensorflow/python/ops/logging_ops.py:264 in merge_all_summaries.: merge_summary (from tensorflow.python.ops.logging_ops) is deprecated and will be removed after 2016-11-30.\n",
      "Instructions for updating:\n",
      "Please switch to tf.summary.merge.\n",
      "WARNING:tensorflow:From ../lib/models.py:87 in fit.: SummaryWriter.__init__ (from tensorflow.python.training.summary_io) is deprecated and will be removed after 2016-11-30.\n",
      "Instructions for updating:\n",
      "Please switch to tf.summary.FileWriter. The interface and behavior is the same; this is just a rename.\n",
      "step 600 / 11000 (epoch 1.09 / 20):\n",
      "  learning_rate = 1.90e-02, loss_average = 3.78e-01\n",
      "  validation accuracy: 90.86 (4543 / 5000), f1 (weighted): 90.84, loss: 3.60e-01\n",
      "  time: 2s (wall 1s)\n",
      "step 1200 / 11000 (epoch 2.18 / 20):\n",
      "  learning_rate = 1.80e-02, loss_average = 3.65e-01\n",
      "  validation accuracy: 91.52 (4576 / 5000), f1 (weighted): 91.47, loss: 3.36e-01\n",
      "  time: 4s (wall 2s)\n",
      "step 1800 / 11000 (epoch 3.27 / 20):\n",
      "  learning_rate = 1.71e-02, loss_average = 3.65e-01\n",
      "  validation accuracy: 92.26 (4613 / 5000), f1 (weighted): 92.23, loss: 3.25e-01\n",
      "  time: 6s (wall 3s)\n",
      "step 2400 / 11000 (epoch 4.36 / 20):\n",
      "  learning_rate = 1.63e-02, loss_average = 3.35e-01\n",
      "  validation accuracy: 92.02 (4601 / 5000), f1 (weighted): 91.99, loss: 3.20e-01\n",
      "  time: 8s (wall 4s)\n",
      "step 3000 / 11000 (epoch 5.45 / 20):\n",
      "  learning_rate = 1.55e-02, loss_average = 3.18e-01\n",
      "  validation accuracy: 92.40 (4620 / 5000), f1 (weighted): 92.37, loss: 3.16e-01\n",
      "  time: 10s (wall 5s)\n",
      "step 3600 / 11000 (epoch 6.55 / 20):\n",
      "  learning_rate = 1.47e-02, loss_average = 3.29e-01\n",
      "  validation accuracy: 92.20 (4610 / 5000), f1 (weighted): 92.18, loss: 3.17e-01\n",
      "  time: 12s (wall 6s)\n",
      "step 4200 / 11000 (epoch 7.64 / 20):\n",
      "  learning_rate = 1.40e-02, loss_average = 3.10e-01\n",
      "  validation accuracy: 92.32 (4616 / 5000), f1 (weighted): 92.29, loss: 3.13e-01\n",
      "  time: 15s (wall 7s)\n",
      "step 4800 / 11000 (epoch 8.73 / 20):\n",
      "  learning_rate = 1.33e-02, loss_average = 3.35e-01\n",
      "  validation accuracy: 92.42 (4621 / 5000), f1 (weighted): 92.39, loss: 3.12e-01\n",
      "  time: 18s (wall 8s)\n",
      "step 5400 / 11000 (epoch 9.82 / 20):\n",
      "  learning_rate = 1.26e-02, loss_average = 3.29e-01\n",
      "  validation accuracy: 92.48 (4624 / 5000), f1 (weighted): 92.46, loss: 3.14e-01\n",
      "  time: 20s (wall 9s)\n",
      "step 6000 / 11000 (epoch 10.91 / 20):\n",
      "  learning_rate = 1.20e-02, loss_average = 3.36e-01\n",
      "  validation accuracy: 92.62 (4631 / 5000), f1 (weighted): 92.60, loss: 3.12e-01\n",
      "  time: 22s (wall 10s)\n",
      "step 6600 / 11000 (epoch 12.00 / 20):\n",
      "  learning_rate = 1.14e-02, loss_average = 3.33e-01\n",
      "  validation accuracy: 92.54 (4627 / 5000), f1 (weighted): 92.51, loss: 3.11e-01\n",
      "  time: 23s (wall 11s)\n",
      "step 7200 / 11000 (epoch 13.09 / 20):\n",
      "  learning_rate = 1.03e-02, loss_average = 3.16e-01\n",
      "  validation accuracy: 92.58 (4629 / 5000), f1 (weighted): 92.56, loss: 3.10e-01\n",
      "  time: 25s (wall 12s)\n",
      "step 7800 / 11000 (epoch 14.18 / 20):\n",
      "  learning_rate = 9.75e-03, loss_average = 3.41e-01\n",
      "  validation accuracy: 92.58 (4629 / 5000), f1 (weighted): 92.56, loss: 3.10e-01\n",
      "  time: 28s (wall 13s)\n",
      "step 8400 / 11000 (epoch 15.27 / 20):\n",
      "  learning_rate = 9.27e-03, loss_average = 3.38e-01\n",
      "  validation accuracy: 92.50 (4625 / 5000), f1 (weighted): 92.47, loss: 3.10e-01\n",
      "  time: 30s (wall 14s)\n",
      "step 9000 / 11000 (epoch 16.36 / 20):\n",
      "  learning_rate = 8.80e-03, loss_average = 3.32e-01\n",
      "  validation accuracy: 92.62 (4631 / 5000), f1 (weighted): 92.60, loss: 3.09e-01\n",
      "  time: 33s (wall 15s)\n",
      "step 9600 / 11000 (epoch 17.45 / 20):\n",
      "  learning_rate = 8.36e-03, loss_average = 3.53e-01\n",
      "  validation accuracy: 92.54 (4627 / 5000), f1 (weighted): 92.52, loss: 3.09e-01\n",
      "  time: 36s (wall 16s)\n",
      "step 10200 / 11000 (epoch 18.55 / 20):\n",
      "  learning_rate = 7.94e-03, loss_average = 3.26e-01\n",
      "  validation accuracy: 92.50 (4625 / 5000), f1 (weighted): 92.48, loss: 3.10e-01\n",
      "  time: 37s (wall 17s)\n",
      "step 10800 / 11000 (epoch 19.64 / 20):\n",
      "  learning_rate = 7.55e-03, loss_average = 3.35e-01\n",
      "  validation accuracy: 92.72 (4636 / 5000), f1 (weighted): 92.71, loss: 3.08e-01\n",
      "  time: 39s (wall 18s)\n",
      "step 11000 / 11000 (epoch 20.00 / 20):\n",
      "  learning_rate = 7.55e-03, loss_average = 3.52e-01\n",
      "  validation accuracy: 92.62 (4631 / 5000), f1 (weighted): 92.60, loss: 3.09e-01\n",
      "  time: 40s (wall 18s)\n",
      "validation accuracy: peak = 92.72, mean = 92.58\n",
      "train accuracy: 92.46 (50851 / 55000), f1 (weighted): 92.44, loss: 3.13e-01\n",
      "time: 1s (wall 1s)\n",
      "test  accuracy: 15.44 (1544 / 10000), f1 (weighted): 12.62, loss: 5.96e+00\n",
      "time: 0s (wall 0s)\n"
     ]
    }
   ],
   "source": [
    "if True:\n",
    "    name = 'softmax'\n",
    "    params = common.copy()\n",
    "    params['dir_name'] += name\n",
    "    params['regularization'] = 5e-4\n",
    "    params['dropout']        = 1\n",
    "    params['learning_rate']  = 0.02\n",
    "    params['decay_rate']     = 0.95\n",
    "    params['momentum']       = 0.9\n",
    "    params['F']              = []\n",
    "    params['K']              = []\n",
    "    params['p']              = []\n",
    "    params['M']              = [C]\n",
    "    model_perf.test(models.cgcnn(L, **params), name, params,\n",
    "                    train_data, train_labels, val_data, val_labels, test_transposed, test_labels)"
   ]
  }
 ],
 "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": "3.5.2"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 0
}