Meanfield learning.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "from __future__ import print_function\n",
    "import numpy as np\n",
    "import tensorflow as tf \n",
    "import mf\n",
    "import matplotlib.pyplot as plt\n",
    "import sudoku\n",
    "from tqdm import tqdm"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "%matplotlib inline\n",
    "np.set_printoptions(precision=3, suppress=True)\n",
    "os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 70,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "meanfield_iters = 50\n",
    "n_epoch = 50\n",
    "learning_rate = 0.01\n",
    "n_modes = 2\n",
    "n = 10\n",
    "p = 3\n",
    "n_samples = 1000\n",
    "batch_size = 50"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 71,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def grid_to_clip(grid):\n",
    "    n_samples,n,_,p = grid.shape\n",
    "    res = np.zeros((n_samples,2,n,n,p))\n",
    "    res[:,0,:,:,:] = -50\n",
    "    res[:,1,:,:,:] = np.where(grid == 1, -50, 50)\n",
    "    return res\n",
    "\n",
    "def to_prob(grid,p):\n",
    "    m,_ = grid.shape\n",
    "    prob_grid = np.zeros((m,m,p))\n",
    "    for x in range(m):\n",
    "        for y in range(m):\n",
    "            if grid[x,y] != p:\n",
    "                prob_grid[x,y,grid[x,y]] = 1\n",
    "    return prob_grid"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 111,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "dataset_X, dataset_Y = [],[]\n",
    "\n",
    "\n",
    "for _ in range(n_samples):\n",
    "    x,y = np.random.randint(0,n),np.random.randint(0,n)\n",
    "    sample = np.zeros((n,n),dtype=np.int32)\n",
    "    \n",
    "    if np.random.randint(0,2) == 0:\n",
    "        sample = sample+2\n",
    "        sample[(x+1)%n,(y+1)%n] = 0\n",
    "        sample[x,y] = 1\n",
    "    else:\n",
    "        sample[(x+1)%n,(y+1)%n] = 2\n",
    "        sample[x,y] = 1\n",
    "        \n",
    "    inpt = np.zeros((n,n),dtype=np.int32)+p\n",
    "    inpt[x,y] = 1\n",
    "    dataset_X.append(inpt)\n",
    "    dataset_Y.append(sample)\n",
    "\n",
    "dataset_X = [to_prob(x,p) for x in dataset_X]\n",
    "dataset_Y = [to_prob(x,p) for x in dataset_Y]\n",
    "\n",
    "inputs = grid_to_clip(np.array(dataset_X))\n",
    "labels = np.array(dataset_Y)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 112,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[[0 0 0 0 0 0 0 0 0 0]\n",
      " [0 0 0 0 0 0 0 0 0 0]\n",
      " [0 0 0 0 0 0 0 0 0 0]\n",
      " [0 0 0 0 0 0 0 0 0 0]\n",
      " [0 0 0 0 0 0 0 0 0 0]\n",
      " [0 0 0 0 0 0 0 0 0 0]\n",
      " [0 0 0 0 0 0 0 0 0 0]\n",
      " [0 0 0 0 0 0 0 0 0 0]\n",
      " [0 0 1 0 0 0 0 0 0 0]\n",
      " [0 0 0 0 0 0 0 0 0 0]]\n"
     ]
    }
   ],
   "source": [
    "print(sudoku.infer_grid(dataset_X[0]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 114,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(50, 10, 10, 3, 3)\n",
      "(10, 10, 3, 3)\n",
      "tf graph is built\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n",
      "epoch:   0%|          | 0/50 [00:00<?, ?it/s]\u001b[A\n",
      "epoch:   2%|▏         | 1/50 [00:07<06:27,  7.91s/it]\u001b[A\n",
      "\n",
      "epoch:   4%|▍         | 2/50 [00:14<05:54,  7.38s/it]\u001b[A\n",
      "epoch:   6%|▌         | 3/50 [00:20<05:29,  7.02s/it]\u001b[A\n",
      "epoch:   8%|▊         | 4/50 [00:26<05:09,  6.72s/it]\u001b[A\n",
      "epoch:  10%|█         | 5/50 [00:32<04:53,  6.51s/it]\u001b[A\n",
      "epoch:  12%|█▏        | 6/50 [00:38<04:41,  6.41s/it]\u001b[A\n",
      "epoch:  14%|█▍        | 7/50 [00:44<04:28,  6.25s/it]\u001b[A\n",
      "epoch:  16%|█▌        | 8/50 [00:50<04:20,  6.21s/it]\u001b[A\n",
      "epoch:  18%|█▊        | 9/50 [00:56<04:13,  6.18s/it]\u001b[A\n",
      "epoch:  20%|██        | 10/50 [01:02<04:06,  6.16s/it]\u001b[A\n",
      "epoch:  22%|██▏       | 11/50 [01:08<03:59,  6.13s/it]\u001b[A\n",
      "epoch:  24%|██▍       | 12/50 [01:14<03:52,  6.11s/it]\u001b[A\n",
      "epoch:  26%|██▌       | 13/50 [01:20<03:45,  6.10s/it]\u001b[A\n",
      "epoch:  28%|██▊       | 14/50 [01:26<03:38,  6.07s/it]\u001b[A\n",
      "epoch:  30%|███       | 15/50 [01:32<03:32,  6.08s/it]\u001b[A\n",
      "epoch:  32%|███▏      | 16/50 [01:39<03:27,  6.11s/it]\u001b[A\n",
      "epoch:  34%|███▍      | 17/50 [01:45<03:22,  6.14s/it]\u001b[A\n",
      "epoch:  36%|███▌      | 18/50 [01:51<03:17,  6.16s/it]\u001b[A\n",
      "epoch:  38%|███▊      | 19/50 [01:57<03:10,  6.16s/it]\u001b[A\n",
      "epoch:  40%|████      | 20/50 [02:03<03:04,  6.16s/it]\u001b[A\n",
      "epoch:  42%|████▏     | 21/50 [02:09<02:57,  6.13s/it]\u001b[A\n",
      "epoch:  44%|████▍     | 22/50 [02:15<02:51,  6.11s/it]\u001b[A\n",
      "epoch:  46%|████▌     | 23/50 [02:22<02:45,  6.11s/it]\u001b[A\n",
      "epoch:  48%|████▊     | 24/50 [02:28<02:39,  6.12s/it]\u001b[A\n",
      "epoch:  50%|█████     | 25/50 [02:34<02:31,  6.08s/it]\u001b[A\n",
      "epoch:  52%|█████▏    | 26/50 [02:40<02:26,  6.09s/it]\u001b[A\n",
      "epoch:  54%|█████▍    | 27/50 [02:46<02:21,  6.13s/it]\u001b[A\n",
      "epoch:  56%|█████▌    | 28/50 [02:52<02:14,  6.09s/it]\u001b[A\n",
      "epoch:  58%|█████▊    | 29/50 [02:58<02:08,  6.12s/it]\u001b[A\n",
      "epoch:  60%|██████    | 30/50 [03:04<02:01,  6.09s/it]\u001b[A\n",
      "epoch:  62%|██████▏   | 31/50 [03:11<01:57,  6.17s/it]\u001b[A\n",
      "epoch:  64%|██████▍   | 32/50 [03:17<01:49,  6.08s/it]\u001b[A\n",
      "epoch:  66%|██████▌   | 33/50 [03:23<01:43,  6.11s/it]\u001b[A\n",
      "epoch:  68%|██████▊   | 34/50 [03:29<01:38,  6.15s/it]\u001b[A\n",
      "epoch:  70%|███████   | 35/50 [03:35<01:32,  6.15s/it]\u001b[A\n",
      "epoch:  72%|███████▏  | 36/50 [03:41<01:26,  6.18s/it]\u001b[A\n",
      "epoch:  74%|███████▍  | 37/50 [03:48<01:20,  6.22s/it]\u001b[A\n",
      "epoch:  76%|███████▌  | 38/50 [03:54<01:14,  6.21s/it]\u001b[A\n",
      "epoch:  78%|███████▊  | 39/50 [04:00<01:07,  6.18s/it]\u001b[A\n",
      "epoch:  80%|████████  | 40/50 [04:06<01:01,  6.16s/it]\u001b[A\n",
      "epoch:  82%|████████▏ | 41/50 [04:12<00:55,  6.13s/it]\u001b[A\n",
      "epoch:  84%|████████▍ | 42/50 [04:18<00:49,  6.14s/it]\u001b[A\n",
      "epoch:  86%|████████▌ | 43/50 [04:24<00:42,  6.12s/it]\u001b[A\n",
      "epoch:  88%|████████▊ | 44/50 [04:30<00:36,  6.11s/it]\u001b[A\n",
      "epoch:  90%|█████████ | 45/50 [04:37<00:30,  6.16s/it]\u001b[A\n",
      "epoch:  92%|█████████▏| 46/50 [04:43<00:24,  6.19s/it]\u001b[A\n",
      "epoch:  94%|█████████▍| 47/50 [04:49<00:18,  6.16s/it]\u001b[A\n",
      "epoch:  96%|█████████▌| 48/50 [04:55<00:12,  6.22s/it]\u001b[A\n",
      "epoch:  98%|█████████▊| 49/50 [05:01<00:06,  6.15s/it]\u001b[A\n",
      "epoch: 100%|██████████| 50/50 [05:08<00:00,  6.16s/it]\u001b[A\n",
      "\u001b[A"
     ]
    }
   ],
   "source": [
    "tf_samples = tf.placeholder(tf.float32,[None, n, n, p])\n",
    "links = tf.Variable(tf.zeros([n,n,p,p]))\n",
    "links_sym = links+tf.transpose(links, [0, 1, 3, 2]) # pairwise weights are symmetric\n",
    "unary = tf.Variable(tf.zeros([n,n,p]))\n",
    "annealing = tf.Variable(tf.zeros([meanfield_iters]))\n",
    "\n",
    "mmmf = mf.BatchedMultiModalMeanField(n, n, p, batch_size, links_sym, unary, tf.exp(annealing), meanfield_iters)\n",
    "\n",
    "q_mf = tf.nn.softmax(-mmmf._theta_mf[-1]) #  sample, each mode, q values.\n",
    "# tf_sample: (s, n, n, p)\n",
    "# q_mf: (s*m, n, n, p)\n",
    "q_mf = tf.reshape(q_mf, (batch_size,-1,n,n,p))\n",
    "\n",
    "# q_mf: (s,m,n,n,p)\n",
    "\n",
    "number_of_modes = tf.shape(q_mf)[1]\n",
    "number_of_samples = tf.shape(tf_samples)[0]\n",
    "tf_samples_stack = tf.tile(tf.expand_dims(tf_samples, 1), [1, number_of_modes, 1, 1, 1]) # of shape s, m, n, n, p\n",
    "prod = tf_samples_stack*q_mf # broadcast magic -> s,m,n,n,p\n",
    "\n",
    "# => log_lh: (s, m)\n",
    "log_lh_samples = tf.reduce_sum(tf.log(tf.reduce_sum(prod,4)+0.0000001), (2,3))\n",
    "#log likelihood for each sample each mode.\n",
    "sample_mode = tf.argmax(log_lh_samples, axis=1)\n",
    "# shape is [number_of_samples]\n",
    "log_lh = tf.reduce_max(log_lh_samples, axis=1)\n",
    "\n",
    "# sum of\n",
    "energies = mmmf.get_modes_energy()\n",
    "modes_prob = mmmf.get_modes_probability()\n",
    "mean_energy = tf.reduce_sum(energies*tf.stop_gradient(modes_prob), axis=1)\n",
    "energy_sample_mode = tf.gather(tf.transpose(energies), sample_mode)\n",
    "\n",
    "to_maximize = tf.reduce_mean(log_lh - energy_sample_mode + mean_energy)\n",
    "mean_log_lh = tf.reduce_mean(log_lh)\n",
    "gradient_total = tf.gradients(to_maximize, [links, unary, annealing])\n",
    "gradient_norm = tf.reduce_sum(tf.abs(gradient_total[0])) + tf.reduce_sum(tf.abs(gradient_total[1])) + tf.reduce_sum(tf.abs(gradient_total[2]))\n",
    "\n",
    "update_rate = learning_rate*batch_size/float(n_samples)\n",
    "\n",
    "update_weights = tf.assign(links, tf.check_numerics(links + update_rate*gradient_total[0],\"weights_upd\"))\n",
    "update_unary = tf.assign(unary, tf.check_numerics(unary + update_rate*gradient_total[1],\"unary_upt\"))\n",
    "update_annealing = tf.assign(annealing, tf.check_numerics(annealing + update_rate*gradient_total[2],\"annealing_upd\"))\n",
    "\n",
    "print('tf graph is built')\n",
    "config = tf.ConfigProto()\n",
    "config.gpu_options.allow_growth = True\n",
    "\n",
    "sess = tf.Session(config=config)\n",
    "sess.run(tf.global_variables_initializer())\n",
    "\n",
    "evol = [[],[],[]]\n",
    "for i in tqdm(range(n_epoch),desc='epoch'):\n",
    "    for b in range(n_samples/batch_size):\n",
    "        #print(np.expand_dims(np.array(inputs[b*batch_size:(b+1)*batch_size]),axis=1).shape)\n",
    "        mmmf.reset_all(np.expand_dims(np.array(inputs[b*batch_size:(b+1)*batch_size]),axis=1))\n",
    "        for _ in range(n_modes): \n",
    "            mmmf.iteration(sess)\n",
    "\n",
    "        parameters = {\n",
    "                    mmmf._theta_clip: np.reshape(mmmf._modes,(-1,2,n,n,p)),\n",
    "                    mmmf._T: 1,\n",
    "                    tf_samples: labels[b*batch_size:(b+1)*batch_size]\n",
    "                }\n",
    "        loglh,loss,gradient,w,u,a = sess.run([mean_log_lh,to_maximize, gradient_norm, update_weights, update_unary, update_annealing], feed_dict=parameters)\n",
    "        evol[0].append(loglh)\n",
    "        evol[1].append(gradient)\n",
    "        evol[2].append(loss)\n",
    "        \n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 86,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[1, 2, 0, 0, 0, 2, 2, 2, 2, 2],\n",
       "       [2, 0, 0, 0, 0, 0, 2, 2, 2, 2],\n",
       "       [2, 2, 0, 0, 0, 0, 2, 2, 2, 2],\n",
       "       [2, 2, 2, 0, 0, 0, 0, 2, 2, 2],\n",
       "       [2, 2, 2, 2, 0, 0, 0, 0, 0, 2],\n",
       "       [2, 2, 2, 2, 2, 0, 0, 0, 0, 0],\n",
       "       [0, 2, 2, 2, 2, 2, 0, 0, 0, 0],\n",
       "       [0, 2, 2, 2, 2, 2, 2, 0, 0, 0],\n",
       "       [0, 0, 2, 2, 2, 2, 2, 2, 0, 0],\n",
       "       [0, 0, 0, 0, 2, 2, 2, 2, 2, 0]])"
      ]
     },
     "execution_count": 86,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    " parameters = {\n",
    "                    mmmf._theta_clip: np.reshape(mmmf._modes,(-1,2,n,n,p)),\n",
    "                    mmmf._T: 1./10.,\n",
    "                    tf_samples: labels[b*batch_size:(b+1)*batch_size]\n",
    "                }\n",
    "sudoku.infer_grid(sess.run(q_mf[0,1], feed_dict=parameters))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 87,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[<matplotlib.lines.Line2D at 0x7f289f025a90>]"
      ]
     },
     "execution_count": 87,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYMAAAD8CAYAAACVZ8iyAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz\nAAALEgAACxIB0t1+/AAAIABJREFUeJzt3XecVNXdx/HP2crSe5EiIFIVQRAUUBELCCbYkmCixp4n\nMT4mxhhsSewkmqLxMUo0GjVKYiwYu4gKiIgoKF2qVOlSpO7uef6Ye2fu3LKXNizc/b5fL17M/mZ2\n9t6Zued3zu/ce8ZYaxERkaotr7I3QEREKp+SgYiIKBmIiIiSgYiIoGQgIiIoGYiICEoGIiKCkoGI\niKBkICIiQEFlb8DuatiwoW3dunVlb4aIyCHlk08+WWutbRT3uEMmGbRu3ZopU6ZU9maIiBxSjDFf\n7s7jVCYSERElAxERUTIQERGUDEREBCUDERFByUBERFAyEBERlAxERHbLzBUb+XTJhsrejJxRMhCR\nQ8ausnLueX02G7fuCtw3etry0PjIcQv4YtVmAGav3MTXW3cC8MH8tXy57pvA4xes2cLyr7cF4kMe\nmMC5D00EUonhg/lrAVi9eTvznOePUlpWnt42ay1l5anvnt9ZWs5UJ8GUlpXz4YJ1FT5PLikZiMhB\nobzcMmP5RiDVYH6+7GsAysotT036kp2l5bzy+QoeeX8hv3tzDmXllvvenMuazTtYsGYL146axnX/\nnkZ5ueXWl2Ywc8VGysotd782h3P+7wMAzrx/PN95+EMAfvDoR5x873sAXPL4ZP709hcAnPqH9+k7\nYiwAF4ycxGVPfBzY1iEPTOAHj34EQN8RYzn9T+MA2LKjlGUbtgKphPHYhEUA3PziDI65/S1Ky8oZ\n8focjrjpNb7ZUcpt/53JOQ9N5ItVm2l38+tc8LdJTF2ygU+XbKD18FdZun7rfn+doygZHCBl5ZYV\nIb0NOTRZa7HWBuK7ysrZvD3YO90b5eXB5wfYUVoWGp+0cB3f7CgNxP/49hfp3qfXp0s2hDY2n3y5\ngZkrNgbii9d+w9yvgj3gyYvW88+PgiseLFyzhWcnLwHg6607ef+LNQC8M3sVPe54m+27svfjkXEL\nOesvE/jkyw2M+ngp337wA96ZvYpnJi/h1pdm8MTERezYVQ7A+i07eWnqch58dz6/Hj2D9d+kevuz\nVm7i7dmreGrSl1z2xMfsLE09/pudZYyfl/r781ZvYeXGzLE4ZtYq3pu7hvvfmZe1PXe+MosPF65j\n7JzVWfERr8/J+nlXWep9uuixjzjqN2/S73fvAqmEcccrs2g9/FX+NWUpAM9MXsIj4xYCcNOL0/nn\nR6nX5+H3FqSf75yHJqZHIJMWHriRgpJBjM+Wfs2EeWsD8X99vIT/fXZqIH7Pa7NpPfzVQENx75tz\n6TNiLKs3bc/ZtibZrrLy0PgnX64PNCrWWv741tx0L9Pr7tdmM2bWqkD8+3+bxK0vzQjEf/DoJM59\n6INA/OK/T6bNja8F4tc8M5Wjf/tWIP73CYtoPfzVwLau3rydu16dRaln/3aUljFh3lra3vRauhHf\nvquMjdt2MXnRejrc8gavfL6Cjdt20Xr4q7w3dzUrvt7GsJGT+NXzn2c9f1m55YF35nGO07h4nfvQ\nRE6+N9VwLduwlXVbdgBw3l8nMuSBCYHH97/vPQb+OdUD/mjhOkZPW866LTv47iMfcvOLqdfuqUlf\nct5fJ/La9JUM+MP73PjCdKy1XPrEx/zw75P5dMkGrn/uM9Z9s5MVX29j5cZt3P7fWUxdsoHR05an\nXpNN25mzclP6+V78dBkAm7eXUuokyDdmfsUvnvvMeb3K0739lRu386OnPgGgIC+PLr95I739Fz02\nOX37hHvGpm9f8WRmzTPvcfuo06sHmLViU/r2w+9nGm6v8Z524pMvw+cW3pu7Jn179LQV6duF+eFN\ncfWiA7d83CGzUF2uXfGPKdStXsh93zkmKz7UGV4uHjEkK/6r56cD8MAF3bPibtbfVWYpKjDp+Fuz\nvgJg0/ZdNK5dLR3ftH0Xj41fxDUD2lEQ8YHYV58u2UCLeiU0rlUtK/7Vxu2UFOZTp3phVnzhmi0A\ntG1UMys+b9Vm1n+zk95tG2TF5361mVkrN3JO9xZZ8RnLN/Li1OXcMqQTxpis+O2vzOLJy3pRrTA/\nK37WXyYw4Ven0KJe9XR89spNnHn/eB77YU9O7dQkHV+09hvO++uH/KB3K+465+h0fEdpOQ+Mnc9j\nExYx8/ZBLP96G/WqF1K9qICR4xYyctzC9PtZXm5ZsGYLExesY+KCddxx9lFAqpb7xMRFfDA/u2e2\nq6ycO16ZlXXg7ywtZ3tpGf83dj5vzPwqHV+89hvGz1vDms07eGDsfCBVRii3lo8WrufaUVPZtD3V\nkz+lQ2N2lJVz6ePZJYlPvtxAnjHpz6Fr7JzVNKhRDMAlj39My/olAMxfvYWdpeW0v+V1BnZpQmlZ\npnHbvquMtVt2sHT9Nm55KfX5LbepWrXbm/Wy1nLi799lZ2k5qzfvSMc/WriO742cFHj8u3NXpxOq\ntzFcs2UHU5ekSj7nepLSlh2lDPjD+wD8/YNMw7th6y6enJQaaXgbz6Xrt/KSkzC8Pl8WTPpAaN0/\nzo0vTA+ND35gfGh8dMj2ANz8YvjzRKlWGH7sFxfkYa3NOn5yRcnAMWZ2qrfoTQZhk1EAz3+yLBCz\n1vLbl2emf95VVk5RQeYNdoer+XnZb/o9r83m2clL6dSsFoOOapb529t2cdML07ltaBca1ixOxzd8\ns5Nj73ybpy7rTb8jG6bjO0rL6HDLGww/syOX92vDrrLydK/i3Icm0qJeCW///GRWbdpO64Y1ADj+\nnndoWLOI8TcMYOrSDfQ5IvV87gG64O7BfLRwHX3apeJuXXTRPYMZM3s1Azo2Jj/PpHuL53Rvwauf\nr+SUjo2oXlTA+Q9PZPuucq47vT1vzPiKvu0a0rRONW58YTrTl29k9spNrNy4nY5Na1GnpJAHnQZz\n/Ly1FObncVidajStU40nP1wMwLgv1lCjuID8PENhfh7/+jg1xJ6/egvzVm1m47Zd7CqzPOOUJiyp\nxt6t/3pt2VHKUb95MxDftrOMR8cvZNmGbemhPcDWnaVc/NhkdpaVZzU+23eV0WfE2HSZwlVaVk7/\n+94LPP83O0rpeeeYQLy4MJ/f/ndmIF6juIA/OrVsr5rFBVmft6XrUw1fSVE+ZzuJ482Z2SOgjre+\nQZh2N78eGg8b+QChiQAIJDJXr7veCY1/+8HgiAtS5ZMwL3l60l5rt+wIje+NUR8vjX+Qx7WjpoXG\n54SU04BAycn1jw/DFxa94skp9GvXkKev6L1H27U3qlwyuHbUVLq3rMslfdsAqUb7TU9vzjV1yYbA\n0Hrjtl0s27A1PTx1TVq4js6H1c56Q3eWlvP3CfPocXg9+rRrmE4GpWXl3PfmXPp3aETP1vVZtSn1\nQS7Iy+MPb82ltNzyq0EdeXrSl7w6fSUt61dn5oqNjJ+3lsk3ncrnyzZiLTw2YSGPTljIknVbGXt9\nf1Y7z/OPiYt5f+4aPly4jsUjhrB1Z6rnuWzDNn709CeM+2INC+8ezMZtqUS3dstOhr/wOaOnrWD8\nDaewylPGeujd+fzh7S94+vLezF6ZGSa//NkKrh01jV+f1ZmFa7ek49OWfs3Vz3zKsONa8tmyjWx3\n6rtrt+zgF899xlHNa/PVxh3pg3dHaTk/+eenHNGoBgvWZM7qqFFcEFqCq1FcwLCQhqhmcUE6UXkV\nFeRxzajg8wD84G/hDdoVT34cGA0A/GzUNKaEDP0feX9hIBEA6Vq539shJSpIJZUvVm0JxMfMWpWu\ntXsVF+QxN+IMllme90oOfRPmB8vUuVDlksHoaSsYPW1FOhn89b0FWT2vpeu38vs353JMizpZvzd+\n3houemwy153ePiv+yZcbGDZyEpf0aZ0VX7h2C39wnvdP3zsmPczeUVrOg+/O58F35/PYD3ume5kl\nRfn8xekZL12/NT3MrlWtIF2S6HV3podVq1ohL3+W6imd8af30w1J7WqFfOhMOv3x7S94wDMpNs5p\nVB4Zt5DfvZGZBJvu1NafmLg4ffYDZBqVCx/7KGvflqxLTTre/sqsrPgaZx9XbNyelTzcxnLG8uxG\nyq3pL92QPZyv7ikdeZWFTNhCKkmEKS7I49XPV4be91lEaSEsEQC8FdGI/2lMsNcOcOvoYC8f4M5X\nZ4fG3TNT/N6J6En+bfyi0LhbjhHZU1UuGbjWf7OTcV+sYeXG7AndX4+ewbtz1wTOCHFn9T9bmn2w\nLVmf6tE+MXFxVtzteQP8/F+ZkYS3F3n5PzITV94J0lc8DVitauFvUe2STNzbo/Q+3psIalUrYLNT\nn35uSvZQeK3TiLvJxRU2YgJ4cWp4nfSe11IN3ThfT9Y/knK5DaM7anJ5J/S8Hnl/YWjcv90ud9Ql\nIvGqTDKYuGAtP/PU9y55fDKfL9vI4KObZj1uh9MwVS/K7p26p48V+yZ6Jkb0JJ+bEpxXANJ1cb/b\n/jsrNO5PMq6nJ4WXIcJKGUA6EQAsXJt9oY07iblmc3bjGXFmY+D3Y+NrwuMicvAwYedKH4x69uxp\n9+VrL4c+OCG0NHBS+0aBnixAk9rF6lmKyEFh0T2D9/qMImPMJ9bannGPqzrXGUS8kGGJAFRiEJGD\nx47S8Ots9qcqkwzycn+arohUkuZ1S0Lj3zrmsND405eHn6r5xKXHhcZfuaYfDWoUBeK3D+3C/5x8\nRCBeoyj8JIiW9Uu46PjDA/HigrzANUuusb84OfKitP2pyiQD5QKRg8uPTm4bGo/quDWqVczPT2sf\net+EX50SGv9LRAPb78iG1A45OaNvu4Z0bForEK9TUsjgo5sF4o1rFYdeMHb5iW2507mA0atDk9qc\n6Lk+yNWoVjEt64UntLaNapJ/AHqzVSYZ5B2AK/hEDjW3ntU58j7/6dWQulL2P/9zQiDesGaw1+y6\naXDHQOyzX5/B8EHBOJB1Nbnr/mHdePNnJ3HlSW2y4j8/rT2Tbjw1UE8/rE41/uBbTQCgVnEB9w/r\nBgRPkOjUrDaF+XmBM/ha1CuhaZ1qoZXmY1vVC21bTnUuyPT71jHNQuPf793qgDT4FakyZxMpF8je\nurRvax7/YHHOnr8w36TPVtsdzepU4/TOTXhn9uqsJRfqlBTS78iGoddWXNKndeDMtA5NanF5vzZ0\nalaL7/8t+zqHcb88hdolBXS7/e2s+Jw7zgzdpim3nM6azTt4+P0FWdequMt+7Cwt5763Mtdk+JdA\ncc2760wK8/NoUKOIq5w1hgC6t6xH/RpFgbWdTmyfuqrd74JerTivR4tAfPDRzRjarTkApeWZOvwp\nHRrx+KW9nHjmvWjbqAZjf9EfgHLPyTZDux3G/cNSow5vIz766r50bVEHY0x6Ub+i/Dwm3XQqJYX5\nlBTl867n2pF//+gEuraoQ3FBHjM96x89c0VvPliwljM6Z5/tmEtVZmRwINb22FvuujJ+UT2Fod0O\no42zpITXwC5N6HNEg0D8jM5NOKNzk0D8qpPacs+5wV7YVSe15c/f6xYYLp/euUm6V+XVukF1Rl7U\nI3Rbn768NzV9F4XVKSnk+R/34eITgrXTcb88hUdCnmvsL07mX1cdH4i/+bOTGHPdSYH4Xy7ozt0h\nPcxnruzN8z8O9myf//EJfHFnsKG7tG9rfvOtLvRtF3xd5945iNu+3YUjGmW/F4vuGcyTl/XidN9r\nvnjEEMZcdxKX9m2dFZ9312Am/OoUrhnQLiv+4Y0DmHvnIF773xN9+3A8tw89ig+GD8iK3zKkE//3\n/WMD62gNP7Mjv/12FxbdMzgrfuHxrQDSy5C4nv9xH1o1qE7d6pnefu1qBYy/IVOKuaJfpofujiD8\npY4TPGtYHeFZ58rb837q8l5Zf9utjXdpnj0qaeCMPLy98GevPJ5jW9XDb8jRzbi0X5tA/PweLbj1\nW5mRkPudApf0aZ1Vr3fjF/RqyXM/ynxW3HWebh7cKZ0IvNt00fGHc0zLuum2Js85fru1qkv9GkWU\nOPMIbrxWtQJ6talPtcJ8jDFZx3ufdg355cCOHNOybmA/cqVKJIPWw19l8qL1e/W7p3duwoPfD9Yd\n7zz7KKbcclog/suBHVg8Ygg9Ds/+kHZoUovFI4bw+CXZE1TN65Yw/oYBPHtlsKFbcPdgJt98aiB+\n/7DuvH7tiYH4Ixf15J9X9KZJ7eKs+MiLezLy4p6c7+sp3TS4Exf0ahWoq940uBNnd2/Oq75G6M/f\n68bQbs2Z9uvTA9tzRpemTL01Oz7i3KPpd2TDrEYE4Oxuh9Hj8HrcPjS7ptrz8Hq0alCdgV2ye0MN\naxbRtlFNerdtEJjE69C0Fu0a16J/h0ZZ8W8dcxjf792K35/XNSve54iG9Di8PtN/e0ZWvMfh9Skq\nyGPxiCG0b5JpuH7zrS4A/POK4zn32Obp+JjrTqa4IJ8f9mnNd3q2TMf7tWuIMYaT2jfizKMy++Em\nxHaNa3Fqx0ySaOsk9Rb1qmfVpIcd15KmtatRXJCf9X4+e+XxoR2BMzo34bxjgz3hv13ck6tOTNXm\nvR2i967vz4WeiUx3wvOZK3pnfXbd23efezQt62cWD3SvBr9lSCdeurpvOu42aOd0b86znuTtNoDN\n65Yw/bcD03Hvqpyfe96TfM+2zrljUPoqc2+DeUJIxwfg9+d3DXRAAH7S/4isuDsCOKNzE2pVy4xU\n3Ea/b7uGNPCsC+Y+vjA/u5PmbpK/8+b+6H+8u28FvsdXdpkoZ8nAGPMvY8w0599iY8w0z303GmPm\nG2PmGmMGVvQ8B8rzP+7D4hFDAm/QXy7ozlldD+OZK7PPPjjxyIY0rFkcaKzdg9tfVz2tc2MATunY\nOCvewel9n3BEg6zG2m0AGteqxnMhNdpqhfl8eOOAQNwYw7vX9w/dR3/D6Io64yI/z9DVUzd2e211\nqxdlTf65F+LV8zXU7vP6460aBBszSF3z4erm6RF5E+XlJ2Z6fN/2bLe3znxc60xj9t3jMg21l/fg\n94u62M57SU4LTw/Y+5H543eP8cQzd/zjskwP2LtW4Ys/CTakACPO65puvHenARx5cc90g+vVt12D\n0HhhQV5WcnB3zb9yrlsa8R8X7nct5OeZrOcJ22bIvBYF/obReVxRfh61Pe+J9zXyrmy7O+2ld4HI\niuLu++mPl6Ub/fB4UUH2mUJ5MY27f3FK98dAvJKrFzlLBtba71lru1lruwHPAy8AGGM6A8OALsAg\n4CFjTPh5WAeQ2wP63NdjdD8Q/qG020toXKsavVrXT8fdD5YxhhGeEkybhpne5gs/6ZO+7W30fjmw\nQ/r2WV0zDV0Pz1D4d+dlnrNZnUyD1NbTW4xaAz2sUXDVi6jh1ilJxc/t3jz7oHEOpI5Na9Gxae10\n2F319dzuzbPWDLrxzMxk4aWedZy8+/PTUzJlEm+p6MgmmXLVT/pnHuMtWXlPLXzufzKvr9e954cn\nQ//IyL0Q019mchvGO84+KquBcl3Rr03W8uTu6929Vd2s3rZ70DeoUZRVO496eyp63+IU5IUf4oX+\nxt1t9H2Ntdtg+p/HTZj+BsxtAP2bnJ+XfX86ni6pEBr3251yr79RdhVFnJ7pb/TduQR/kihNJ4Pw\neH5gxBCRJCLilX36e87LRCb17n0XeNYJDQVGWWt3WGsXAfOBXlG/v781q1ONOXcMCpRrXNWLCrKW\novB+eL3D8FaeIbO7pIUxMMDT83fX5DcGzvOUGEo8DcnVngbQ+7duGtwpfdvbGHzvuFah2+0v6bg+\nvjlYyoJUDd7L7Rm+/NO+WXG3kRjStVlo/JzuzbPi6SvafR/so50acJfDamftT4/DM4nUGy+O6N15\n7ek80KCjwifjOjXLnhtxd8H7HkOmASzxJQL38f5G2z3o/SUL9332b39Uz3BfeoxRDaO/US5PN/r+\nZBCeJNz337/P7k/+bXZ/9jfybhKISip7I+pzEXWuftQIoDgQD08SbrzQl9EyI4OoEUMVKRN5nAis\nsta6q6Y1B7wrpS1zYjnhX26jZb3qVCvMD5RrvA53yhgPX3hs9nM5Tea953fN+gC5f+GHJ7TO+iC6\nn40ereplxd0DrqazNr/Le6Ds6QejJOIil0a1ikPj3slByAz7i31DYPeEC/8BYyN6hu5rEXVw+4/T\n3W2s9oeoxiDY682UQLxsVMnEzX8RveGoRt+/i1GNflQveXdEjSr85aDMvvkaOhteMsmMDAiP72ED\nGEgeOXj/o8tH2X8rPTfgHwE4cwlF+eGP92+zjUiw7uP8Cbayy0T7dGqpMWYMENbdutlaO9q5fQGZ\nUcGePv9VwFUArVqF94jjlPkKwA/5GniAd/y9ZOdd9E4epe5I/ed/06JqzOmDPuIACDQe+Qfuw+Cf\n1HI/uP54VPnAbST8++C+dv5j2f19Q/hr4ZeLZBD1nMFeb+r/qNco2Kt2XgvfvqXr54EkEd4ARm1f\nRKVnnwTf54i40xmISozBYyH8/U+XRiImUwPHQg4axj0dGfjLSm48kDDLwzsJbrnJn3jj5hgqyz4l\nA2tteA3CYYwpAM4FvOcKLge8s3otnFjY848ERkJqobq92cZSX0vd0N/AE11LDPYA4xrA8MbD/xan\nk4E/fgB7BsGeXngPMHIEENkwpv73N4xRveEDOWSOGoUE9pnwkUF55PtccQPoHxlENYBRveFc9Bh3\nezTk/O9PEm4DGBwZhL9GmYll/2TqniXGfeHfB9fuzg3sipgbcB8f6DBFJInMnEF4Wamy5LpMdBow\nx1rrXc/5ZWCYMabYGNMGOBKYHPrb+0HUF6l7RdVDg41Eiv/YjOoNRSWJdM+wEnsGUcP+wBkXVNwD\n9G9xOgH6Xwv3hu+OgnRjkP34XCTGqFpyoNGPKI1lGkx88dT/0SWz7MdH1cmj3v7c9JKjRoa7Vz6K\n2ufy8oo/84FR0h6WzPbF7s4lRJ9N5PT0fRvrPt7/Prllpd0dkeZFdBIOlFxfgTwMX4nIWjvTGPNv\nYBZQClxtrS0L++X9oXQ3ruwMDoFT/0fXScMf72/c3cf7n989sPwfngOZDKJ6vVEf9MgeYKAn6Y6e\nIl6jiJFBoPd8IBNjRB0matTjT4FRCTCql5wpH/r+XtQEcg5ei6iGMapjFDzLKPx9jj5GwhvGqJHh\ngewkB84mcjqQ/pMYohr3yHjEyKAs4lhzf4waweZaTpOBtfaSiPhdwF25/NuuXZ5Lzr2nLnr5G4O4\nnr5fpjSSLaqu7j5vVPmgMqQTYEF4Aow66KPjgb+QivuibmL0xw/kFePRI8PdS2jlEQ2jjXj/o+YM\nDoar5KPPMtq9YyQqnu49+88+ijoWDmCDGDU3EHX9QXDOIGrE4M4ZhMejEmPU6cC5lvgrkL0jgw4h\nqxFCSK/H+T+q1BEsB0T1kuJ6T9nPn4se4O5yt9V/YESVOtI9vd2sn9uI5BFVMjmQokZJ/vp2VG84\nap+jesmZ+ZPKawCjRHWM/KWxsoh9i5p7KIsoN0WdcnogE6M/6ZdGlImi5gYy8YjH+/bZXYfKH3f3\nubJGBlUqGURNIEVNmvr7qw2dK2lr+lY1jCoTuB9w/xe8u+91verRKz0eaG6ijLrS1H/Qux/kqIuU\n/BPIUaMnV2U2hFHlwKgTCAKJzvk/+sya8H2LGjFWpqj3M3LfIk+zDW8woyaoo/Y96nsK9id/Ix41\nMiiNmzPYx7j7YtStEX11fC4lftVS78qEUcOvYE3f7bllP274mZ3ofFht+rfPXgdnYJem3PvmXIZ2\ny17WoW+7hvy4/xFc7ls0q3Htavz6rM4MjLgIKmp5iGYhqzN6t9frF6e3Z9mGbYH4e9f358v1WwPx\npy7vzZyVmwIHxvk9WjB9+cas5RcArj+jA8UFeZzdLfsSEXfhsAG+6ziiJtNrFBVwasfGXNo3+zU6\nkKLmjKJOOd3d0ymjSmZRr0Vljgxd0eWg3RsNRc0xlUXU1d3ORthI4Jkre9Oucc1APNfSZxMF5hIq\nnhvwx888uhn3j5nHBb2yT4vv2boeg7o05YZBHbLidaoXcutZnUMXlTwQqkAyyIwMos7j9zemIy/u\nyajJSwILgpUU5YdeAdyucc3ASpGQ+nD8KmLN9stCVlWE1KJcYedDf3zzaaEXls24bWBob/uaU48M\nff7WDWvQOmShs/o1iujTLvilGxefcDgXHn944INep3oht4SshX90izrpZYi9uraoS7eWdQPr5+fl\nGR6LuBocUitK+r1yTb/0Mhm5EDXZGXVqcVRvOKrRL4toSCtzzsgVNX8STGip/3d3orgsIu7+vbBl\nqP1LwLhGXtQjfXW/X/2QbyPbU91b1WXqkq8DlYThZ3bk+uc+o23D7AR1XOt6/OeTZXRokl2Gbl63\nhOm3BZdeKy7I5+GIVX79HccDKfHJwHtqqb8e2qtNfSYvWh84iNs0rMGNnuUgDqSwNW8g+krisNUZ\n9ydjTOB0wDhhyaykKD9rdcvdEZZgAY5qHvzSFYB7zj06a835vVUeMR/i2t1yUOSIoTz8eQ7EvMmZ\nRzXljZlfRd4fLBOm/g8ktMizzFL/+48pdzmSC3plLxzYrE4J957ftcIVAfzO6BI+op5zx6DQ+E9P\naceKjcFRcpQnLu3FknVbA/vQv0NjptxyeuDx3+3ZkpPaN8paK+xQlPhk4J0z8Pd6Hr/kOFZv1hff\nJ4V/OO7670/7sWbL9kD8ouMP56lJXwbimRMIdq/Rjzy1NKIhdWvRTWpn94b35iSS8Teckl411uv3\n53Xlm52lgfhfLwzvkbr8+3zNgHYMf2F64GLN/h0a89asVYEyTlSSOKxuSWRy9y4Bvi+iOlLXD+wQ\nGr9lSCfGer5oxlWnpJCjQ77lLYox5pBPBFAVkoFnzsA/7KtRXECbHPespfKlDuzgwX3H2UdxR8j3\n1HZqWpsPF64LTCBmGnd88fAG0D1Nt3pxdiPVrnFN7j2/K6d1yq4N783IoGX98HJJ1NLdUf53QDse\nGDs/EB/WqxXDQpLsBb1aMvjopoE1rn7QuxXvzV3N9yMS88HkihPbcsWJ4d/DXBUlviX0rkbhP1VQ\nJMzDF/X3tIYCAAAN/klEQVRgzspNgRLcDYM68IvnPqOr79unos4mGnxUU+afeiRXnhisA4f1hiua\nM7hhUAe6Ns/dt15dd0YHrjsjvAcdxhgTSASQOjli9E/77c9NkwMk8cnAu1BdZZ2/K4eWOiWF9G4b\n/BKZ7q3qpb8P1+vH/Y9g2Yatgd54QX4e153efrf/rjGp74HwfyMdZH+Pg0guJD4ZeCcUo1YtFNkX\nDWsW88hFPff5eYwx/PF7we+YFjkQEt86eqYMAhPIIiKSkvxkYFUmEhGJU7WSgcpEIiKhEt86Zs0Z\naGQgIhIq+ckga84g8bsrIrJXEt86lnlGBgfDEsEiIgejxCcDm3VqqZKBiEiYxCcD71cgV9Y3CImI\nHOwS3zrq1FIRkXhVKhkcDF8eIiJyMKpSyUBERMIlPxmUxz9GRKSqS3wyKNPIQEQkVuKTgVUyEBGJ\nlfhkUKYykYhIrMQnA3cC+fozdv9LRkREqpoqkwzCvsdVRERSkp8MnK+9rOj7ZUVEqrrkJwNn/tj/\nZeUiIpJRBZJBKhuYxO+piMjeS3wT6SYDlYlERKIlPhm4p5aqTCQiEi3xycAdGWj1ahGRaIlvIt2z\niTQyEBGJlvxk4JxNpDkDEZFoiU8G7kJ1ygUiItESnwystRgDRtlARCRS4pNBubUqEYmIxMhZMjDG\ndDPGTDLGTDPGTDHG9PLcd6MxZr4xZq4xZmCutgFSp5Zq8lhEpGIFOXzu3wO3WWtfN8YMdn7ub4zp\nDAwDugCHAWOMMe2ttWU52xLlAhGRCuWyTGSB2s7tOsAK5/ZQYJS1doe1dhEwH+gV8vv7aSP05TYi\nInFyOTL4GfCmMeY+UkmnjxNvDkzyPG6ZE8sNq4GBiEicfUoGxpgxQNOQu24GTgV+bq193hjzXeAx\n4LQ9fP6rgKsAWrXa++8j0JSBiEjF9ikZWGsjG3djzJPAtc6PzwGPOreXAy09D23hxMKefyQwEqBn\nz557Ve9RkUhEJF4u5wxWACc7twcA85zbLwPDjDHFxpg2wJHA5BxuB0aFIhGRCuVyzuBK4H5jTAGw\nHafcY62daYz5NzALKAWuzuWZRNZqbCAiEidnycBaOwHoEXHfXcBdufrb2X9LcwYiInESfwWyiIjE\nS3wysOjUUhGROIlPBqBF6kRE4iQ+GWj+WEQkXuKTAahMJCISJ/HJQGsTiYjES34y0AyyiEisxCcD\nUC4QEYlTJZKBiIhUrEokA51aKiJSscQnA61NJCISL/nJAK1NJCISJ/HJADSBLCISJ/HJQFUiEZF4\niU8GoAlkEZE4iU8GugJZRCRe4pMBaM5ARCRO4pOB5gxEROIlPxmgU0tFROIkPhmkKBuIiFQk8clA\nZSIRkXiJTwagMpGISJwqkAw0NBARiZP4ZGCtZgxEROIkPhmAykQiInESnww0gSwiEi/xyQDAqFAk\nIlKhxCcDrU0kIhIv8ckANGcgIhIn8clAcwYiIvGSnwzQqaUiInESnwxAX24jIhIn8clAZSIRkXiJ\nTwYiIhIv8clAp5aKiMRLfDLA6tRSEZE4yU8GKBmIiMRJfDJQkUhEJF7ikwFobSIRkTg5SwbGmGOM\nMR8aY6YbY/5rjKntue9GY8x8Y8xcY8zAXG0DgNW5pSIisXI5MngUGG6tPRp4EfglgDGmMzAM6AIM\nAh4yxuTnaiMsmjMQEYmTy2TQHhjn3H4bOM+5PRQYZa3dYa1dBMwHeuVwO1QkEhGJkctkMJNUww/w\nHaClc7s5sNTzuGVOLCdUJRIRiVewL79sjBkDNA2562bgMuABY8ytwMvAzr14/quAqwBatWq1L9u5\n178rIlIV7FMysNaeFvOQMwCMMe2BIU5sOZlRAkALJxb2/COBkQA9e/bcqz6+BgYiIvFyeTZRY+f/\nPOAW4GHnrpeBYcaYYmNMG+BIYHKutgM0ZyAiEieXcwYXGGO+AOYAK4DHAay1M4F/A7OAN4CrrbVl\nudoInVoqIhJvn8pEFbHW3g/cH3HfXcBdufrbWX8LNDQQEYlRRa5AFhGRiiQ/GahKJCISK/nJAJ1a\nKiISJ/HJQF9uIyISL/nJwGrOQEQkTuKTAWihOhGROIlPBrrMQEQkXuKTAejLbURE4iQ+GWgCWUQk\nXuKTAWjOQEQkTuKTgeYMRETiJT8ZVPYGiIgcAhKfDEBXIIuIxEl8MlCZSEQkXuKTAegKZBGROFUg\nGWhoICISJ/HJwFqdWioiEifxyQCUDERE4iQ+GahIJCISL/HJALQ2kYhInMQnA6tzS0VEYiU+GYDm\nDERE4iQ+GWhcICISL/nJQF97KSISK/HJAFCdSEQkRuKTgcpEIiLxEp8MQGUiEZE4iU8GOrVURCRe\n4pMBaMpARCRO1UgGlb0BIiIHucQnA1WJRETiJT4ZgL72UkQkTuKTgdXJpSIisZKfDHQFsohIrMQn\nA9DZRCIicRKfDDSBLCISL/HJAPTlNiIicRKfDDSBLCISL/HJANAMsohIjH1KBsaY7xhjZhpjyo0x\nPX333WiMmW+MmWuMGeiJ9zDGTHfue8Dk+CIAzRmIiMTb15HBDOBcYJw3aIzpDAwDugCDgIeMMfnO\n3X8FrgSOdP4N2sdtqJBFAwMRkTj7lAystbOttXND7hoKjLLW7rDWLgLmA72MMc2A2tbaSTa1nOiT\nwNn7sg27Q6eWiohULFdzBs2BpZ6flzmx5s5tfzyUMeYqY8wUY8yUNWvW7N2WqEwkIhKrIO4Bxpgx\nQNOQu2621o7e/5uUYa0dCYwE6Nmz51436zq1VESkYrHJwFp72l4873KgpefnFk5suXPbH88Zq1kD\nEZFYuSoTvQwMM8YUG2PakJoonmytXQlsMsYc75xFdDGQ49GF5gxEROLs66ml5xhjlgEnAK8aY94E\nsNbOBP4NzALeAK621pY5v/YT4FFSk8oLgNf3ZRt2bztz/RdERA5tsWWiilhrXwRejLjvLuCukPgU\n4Kh9+bt7QvPHIiLxqsQVyJpAFhGpWOKTgdUlyCIisRKfDEBzBiIicRKfDDQuEBGJl/xkoGwgIhIr\n8ckAIMcLo4qIHPISnww0MBARiZf4ZABajEJEJE7yk4EmDUREYiU+GVh0aqmISJzEJwNQmUhEJE7i\nk4GqRCIi8RKfDECnloqIxEl8MrA6uVREJFbikwFozkBEJE7ik4HmDERE4lWJZKApAxGRiiU+GaQo\nG4iIVCTxyUBVIhGReIlPBqAykYhInMQnA33tpYhIvMQnA9CMgYhInKqRDJQNREQqlPhkoCqRiEi8\nxCcDAKNCkYhIhRKfDLQ2kYhIvOQnA12BLCISK/HJAJQMRETiFFT2BuTaSe0b0axOtcreDBGRg1ri\nk8GtZ3Wu7E0QETnoVYkykYiIVEzJQERElAxERETJQEREUDIQERGUDEREBCUDERFByUBERABzqHwT\nmDFmDfDlXv56Q2DtftycQ4H2uWrQPlcN+7LPh1trG8U96JBJBvvCGDPFWtuzsrfjQNI+Vw3a56rh\nQOyzykQiIqJkICIiVScZjKzsDagE2ueqQftcNeR8n6vEnIGIiFSsqowMRESkAolOBsaYQcaYucaY\n+caY4ZW9PfuLMaalMeZdY8wsY8xMY8y1Try+MeZtY8w85/96nt+50Xkd5hpjBlbe1u8bY0y+MWaq\nMeYV5+dE77Mxpq4x5j/GmDnGmNnGmBOqwD7/3PlczzDGPGuMqZa0fTbG/N0Ys9oYM8MT2+N9NMb0\nMMZMd+57wJh9+F5Ha20i/wH5wAKgLVAEfAZ0ruzt2k/71gw41rldC/gC6Az8HhjuxIcDv3Nud3b2\nvxho47wu+ZW9H3u579cBzwCvOD8nep+BfwBXOLeLgLpJ3megObAIKHF+/jdwSdL2GTgJOBaY4Ynt\n8T4Ck4HjAQO8Dpy5t9uU5JFBL2C+tXahtXYnMAoYWsnbtF9Ya1daaz91bm8GZpM6iIaSajxw/j/b\nuT0UGGWt3WGtXQTMJ/X6HFKMMS2AIcCjnnBi99kYU4dUo/EYgLV2p7X2axK8z44CoMQYUwBUB1aQ\nsH221o4D1vvCe7SPxphmQG1r7SSbygxPen5njyU5GTQHlnp+XubEEsUY0xroDnwENLHWrnTu+gpo\n4txOymvxZ+AGoNwTS/I+twHWAI87pbFHjTE1SPA+W2uXA/cBS4CVwEZr7VskeJ899nQfmzu3/fG9\nkuRkkHjGmJrA88DPrLWbvPc5PYXEnCpmjDkLWG2t/STqMUnbZ1I95GOBv1pruwPfkCofpCVtn506\n+VBSifAwoIYx5kLvY5K2z2EqYx+TnAyWAy09P7dwYolgjCkklQj+aa19wQmvcoaOOP+vduJJeC36\nAt82xiwmVfIbYIx5mmTv8zJgmbX2I+fn/5BKDkne59OARdbaNdbaXcALQB+Svc+uPd3H5c5tf3yv\nJDkZfAwcaYxpY4wpAoYBL1fyNu0XzhkDjwGzrbV/9Nz1MvBD5/YPgdGe+DBjTLExpg1wJKmJp0OG\ntfZGa20La21rUu/lWGvthSR7n78ClhpjOjihU4FZJHifSZWHjjfGVHc+56eSmhNL8j679mgfnZLS\nJmPM8c5rdbHnd/ZcZc+q53jGfjCpM20WADdX9vbsx/3qR2oI+Tkwzfk3GGgAvAPMA8YA9T2/c7Pz\nOsxlH844OBj+Af3JnE2U6H0GugFTnPf6JaBeFdjn24A5wAzgKVJn0SRqn4FnSc2J7CI1Arx8b/YR\n6Om8TguAB3EuJN6bf7oCWUREEl0mEhGR3aRkICIiSgYiIqJkICIiKBmIiAhKBiIigpKBiIigZCAi\nIsD/A54VSp+w+s3uAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f28c27a48d0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYEAAAD8CAYAAACRkhiPAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz\nAAALEgAACxIB0t1+/AAAIABJREFUeJztnXl8FeX1/z8nCYRNdkQEISi44IaSIrjQulUqrdh+rV/a\nqtS69FetWxcL1a+1Wpeqbd0qrVUrWiulSguyiiguIEvYd8IaskBCIASyJ/f8/rgzN3Nn5mRmkpvc\nMDnv1yuvzD13Zu7zzHI+z3OejZgZiqIoStskJdkJUBRFUZKHioCiKEobRkVAURSlDaMioCiK0oZR\nEVAURWnDqAgoiqK0YVQEFEVR2jAqAoqiKG0YFQFFUZQ2TFqyE+BF7969OSMjI9nJUBRFOa5YtWrV\nQWbu47VfqxeBjIwMZGVlJTsZiqIoxxVEtNfPfhoOUhRFacOoCCiKorRhVAQURVHaMCoCiqIobRgV\nAUVRlDaMioCiKEobRkVAURSlDRNaEXhzyW58sC4/2clQFEVp1YRWBP6xPAfzNhYkOxmKoiitmtCK\nAAFgTnYqFEVRWjfhFQFSEVAURfEivCIAAkNVQFEUpSHCKwJaE1AURfEktCIAQOsBiqIoHoRWBIhI\nawKKoigehFcEAGhdQFEUpWHCKwLaJqAoiuJJuEUg2YlQFEVp5YRXBEBgrQooiqI0SHhFQGsCiqIo\nnoRXBKBtAoqiKF6EVgRApDUBRVEUD0IrAtGagMqAoihKQ/gSASJ6gIg2EdFGInqXiDoQUU8iWkhE\n2cb/Hpb9JxPRDiLaRkTXWOwjiGiD8d2LRETNkanobzXXmRVFUcKDpwgQUX8A9wLIZOZzAKQCmABg\nEoBFzDwUwCLjM4homPH92QDGAniFiFKN000BcAeAocbf2ITmxppuaJuAoiiKF37DQWkAOhJRGoBO\nAPIBjAcw1fh+KoDrje3xAKYxcxUz7wawA8BIIuoHoCszL+NonOYtyzEJh0hnEVUURfHCUwSYOQ/A\ncwByABQAOMLMHwLoy8zm0l37AfQ1tvsD2Gc5Ra5h629s2+0OiOhOIsoioqyioqIA2bGcA1oTUBRF\n8cJPOKgHoqX7wQBOBtCZiG6y7mOU7BPmcpn5VWbOZObMPn36NOocOm2EoiiKN37CQVcB2M3MRcxc\nA2AGgIsBHDBCPDD+Fxr75wE4xXL8AMOWZ2zb7c2CLiqjKIrijR8RyAEwiog6Gb15rgSwBcAsABON\nfSYCmGlszwIwgYjSiWgwog3AK4zQUSkRjTLOc4vlmMSjNQFFURRP0rx2YOblRPQegNUAagGsAfAq\ngC4AphPRbQD2ArjR2H8TEU0HsNnY/25mrjNOdxeANwF0BDDP+GsWCDpthKIoiheeIgAAzPwbAL+x\nmasQrRW47f8EgCdc7FkAzgmYxkZBBHCkJX5JURTl+CXEI4a1TUBRFMWL8IqAtgkoiqJ4Em4RSHYi\nFEVRWjnhFQFdVEZRFMWT8IqA1gQURVE8Ca0IANomoCiK4kVoRYB0URlFURRPwisCgFYFFEVRPAiv\nCGibgKIoiifhFQFoRUBRFMWL8IqALiqjKIriSXhFAFoTUBRF8SK8IqDTRiiKongSWhEAtIuooiiK\nF6EVgWhNQGVAURSlIcIrAslOgKIoynFAeEVA2wQURVE8Ca8I6KIyiqIonoRXBLQmoCiK4km4RSDZ\niVAURWnlhFcEdFEZRVEUT0IrAtCagKIoiiehFYHoVNLJToWiKErrJrwioIvKKIqieBJeEYCOGFYU\nRfEivCKgbQKKoiiehFcEoOMEFEVRvAivCOiiMoqiKJ6EVwSgNQFFURQvQisC0GkjFEVRPAmtCJBO\nJq0oiuJJeEVAF5VRFEXxJLwiAO0iqiiK4kV4RUDbBBRFUTwJrwjoojKKoiiehFcEtCagKIriiS8R\nIKLuRPQeEW0loi1ENJqIehLRQiLKNv73sOw/mYh2ENE2IrrGYh9BRBuM714kombrwqPTRiiKonjj\ntybwAoD5zHwmgPMBbAEwCcAiZh4KYJHxGUQ0DMAEAGcDGAvgFSJKNc4zBcAdAIYaf2MTlA8XSGsC\niqIoHniKABF1AzAGwOsAwMzVzFwCYDyAqcZuUwFcb2yPBzCNmauYeTeAHQBGElE/AF2ZeRlH+26+\nZTkm4ZAuKKAoiuKJn5rAYABFAP5ORGuI6DUi6gygLzMXGPvsB9DX2O4PYJ/l+FzD1t/YttubBZ02\nQlEUxRs/IpAG4EIAU5j5AgBlMEI/JkbJPmEul4juJKIsIsoqKipq5Dm0HqAoiuKFHxHIBZDLzMuN\nz+8hKgoHjBAPjP+Fxvd5AE6xHD/AsOUZ23a7A2Z+lZkzmTmzT58+fvMShy40ryiK4o2nCDDzfgD7\niOgMw3QlgM0AZgGYaNgmAphpbM8CMIGI0oloMKINwCuM0FEpEY0yegXdYjkm4WhNQFEUxZs0n/vd\nA+AdImoPYBeAWxEVkOlEdBuAvQBuBABm3kRE0xEViloAdzNznXGeuwC8CaAjgHnGX7OgbQKKoije\n+BIBZl4LINPlqyuF/Z8A8ISLPQvAOUES2FiINBykKIriRWhHDAMaDlIURfEitCJAOo2ooiiKJ+EV\nAZBqgKIoigfhFQFdVEZRFMWT8IoANBqkKIriRXhFQKeSVhRF8STEIqCLyiiKongRXhGA1gQURVG8\nCK0IQKeNUBRF8SS0IkCqAoqiKJ6EVwQI2iagKIriQXhFANomoCiK4kV4RUCjQYqiKJ6EVwR0URlF\nURRPwisCWhNQFEXxJLwiAG0TUBRF8SK0IhCdS1pRFEVpiNCKgCkB2i6gKIoiE14RMFRANUBRFEUm\nvCJg1AVUAxRFUWTCKwKxmoDKgKIoikR4RcD4rxKgKIoiE14R0DYBRVEUT0IsAmabgKqAoiiKRGhF\nwERrAoqiKDKhFQEdK6YoiuJNeEXA7CKqNQFFURSR8IqA2TCsbQKKoigi4RUB47/WBBRFUWTCKwKx\nmoCiKIoiEV4RiLUJqAwoiqJIhFcEtCagKIriSWhFwEQrAoqiKDKhFQHSqoCiKIon4RUB4/8Nf1ma\n1HQoiqK0ZsIrAoYKZBceS25CFEVRWjG+RYCIUoloDRHNNj73JKKFRJRt/O9h2XcyEe0gom1EdI3F\nPoKINhjfvUjUfJM76KwRiqIo3gSpCdwHYIvl8yQAi5h5KIBFxmcQ0TAAEwCcDWAsgFeIKNU4ZgqA\nOwAMNf7GNin1DdCM+qIoihIafIkAEQ0AMA7AaxbzeABTje2pAK632KcxcxUz7wawA8BIIuoHoCsz\nL+No5/23LMckHNUARVEUb/zWBJ4H8CCAiMXWl5kLjO39APoa2/0B7LPsl2vY+hvbdnuzoBqgKIri\njacIENE3ARQy8yppH6Nkn7DOmER0JxFlEVFWUVFRY0+SqOQoiqKEFj81gUsAXEdEewBMA3AFEf0D\nwAEjxAPjf6Gxfx6AUyzHDzBseca23e6AmV9l5kxmzuzTp0+A7NSjEqAoiuKNpwgw82RmHsDMGYg2\n+H7MzDcBmAVgorHbRAAzje1ZACYQUToRDUa0AXiFEToqJaJRRq+gWyzHJBytCCiKoniT1oRjnwYw\nnYhuA7AXwI0AwMybiGg6gM0AagHczcx1xjF3AXgTQEcA84y/ZoG0LqAoiuJJIBFg5sUAFhvbxQCu\nFPZ7AsATLvYsAOcETWRj0JqAoiiKN+EdMZzsBCiKohwHhFcEVAUURVE8Ca8IaF1AURTFk9CKgGqA\noiiKN6EVAdUARVEUb8IrApZGAV1nWFEUxZ3wioBlO6IaoCiK4kp4RcCiAhGtCSiKoriiIqAoitKG\nCa8IwNomkMSEKIqitGLCKwJaE1AURfEktCJgRRuGFUVR3AmtCFi7iGpNQFEUxZ3wioBlmyPiboqi\nKG2a8IqARQU4cStfKoqihIrwigCs4aAkJkRRFKUVE14R0N5BiqIonoRXBCzbKgKKoijuhFcErG0C\nqgGKoiiuhFYEAO0iqiiK4kVoRSC+TSB56VAUJZyszjmMjElzsP9IZbKT0iTCKwKW7YiqgNLKWbar\nGIfKqpOdjCaxKf9IspPgi5cWZWP8y180+Txvf7kXALBkx8EmnyuZhFcESCeQU44PmBkTXl2GH7y2\nPNlJaTQfrMvHuBe/wJz1BclOiid/WLgd63L9C9bRyhp8uGm/w57I1QvfXZGD/JKKBJ7RP+EVAcu2\ntgkozUF1bQRTl+5BXRNrmubhWwpKHd/9+O0s/GPZ3iadvyXILjwGANhh/G8N3PTaclzy9Me+9y8+\nVoWlO52l+p9NX4c7316FvcVlrsf5vfvl1bX49itLsDk//j6XlFdj8owNuPn15BQCwisCOk5AaWb+\n8ulO/GbWJvw7a1+TztOQiCzYdAAP/3djk86fSHYVHcMPXluG8uraOLv5uiVjdP6uomPYd6jcYf9i\nx0HkBShdT3h1Gb7/N6cjzimOnru8ui7+C6EqUFMXwZ8/2YHKmvj9V+09jDU5JXhy7pY4u3n7kxUO\nbCMikLx0KMc/NXURrN1X4rCXlNcAAI5VxTvEY1W1mPjGCt8OKJmFlFnr8gOVQJ+cuxVLdhTji2z/\ncfCCIxXNus73FX/4FJc980mTz2PWZuxtiKYvkbJgz9u0FTl4dsE2TFm8031/QSiT9RSEVwSgC80r\nieH387bi+j8vwbb9R33tP3dDAT7dXoQ/Ldzua/9kPp73vrsGnwdw6EHZUlCK0U99jKlL9zTbbySa\nOp83xPQx9r3NGoOztmTsbzsgVotK0nMQWhGA1gSUgJSUV+Pz7CKHfaPR66W4rCrOTgFbBg+XVSNj\n0hy89eWeOHtrDFd+uGk/Rj25CFW1da7f21MsXYs9B6Nx9C93Fccfz4yXP852DeNIfLK10LcQNwXp\nfthL8FQfA/OFV40iWYRWBLRhWAnK7VOzcPPrKxzhnRhiOMCf3QwPvbsivg3Bb8nTD//O2ucQmcbw\n2w82Y39pJYqOBhM+v1nJPVyB5z7cjjveyvKdplvfXIlrnv/M9/6NJWKbet7saSiV4O0Edfamf0pW\nxCItKb/aAuiiMkpQth+IljJr6+K9AAmvu9QYGrTrYGPWu3hl8Q707pKOGzNPibP/8r31AIBbRmcE\nOl8kwkhJcabcr8AFxXwnHY2trQC7KHvdT+f9dw8TSc+LGalIlpcKrwhYtlUDlETg9zGKlRyl8IGN\nxhRSnpm/DQAcIiBRF2FEmNEu1b3yX8eMlCb0fBeF0qvm0ArX+vB7P6S8iXkWagjJbrMMbzhIJ5BT\nBB77YDMyJs1JdjJitERN9XuvLsPQh+aJ3zd1rIOJdBZnKEVWhxW7D+HgsSrx++YmeO8gf3Ypz7Gf\n04bhxBK/qIyqgFLPG0t2N/i9X4cYtGHYxF7yS2SbgMSKPYca/N536Veyi11c3I9oqAZw41+/xHde\nWeorPU1Bmk7Gfv9jIiCEfYJi/9VYm0CjztZ0wisCOliszfPQfzZg3Iuf+97fDOM4YsIJagyVnEZr\neDyD1wSaN9E5AXoNNRbpPtsvhdS108R/mND9gGT7pzbRJqBdRNsm7yzPadRx9t4hJv6dvXmAz9+T\nuiS2oHOQ3pFEhYn80rJ5dv8tuz1w7N/Ab4cB8+eS1TYQ2poA4toEVAXCTG1dxNGjpykErQlI+O1L\n3xocsBQH9z9wKoqUZ2cIxLA7SsW+fi4hOLqCGv+l6y6Lte08QpfS2P6O3kGtPBxERKcQ0SdEtJmI\nNhHRfYa9JxEtJKJs438PyzGTiWgHEW0jomss9hFEtMH47kWixr5e3uhC822HkU8uwlee+CjwcfbC\nQSwcIDwwzt4+QldA4almyfGJTsf9PM2B5OwlcWjy7wV0tM2B75qAZJdqCIHTEfCABOOnJlAL4OfM\nPAzAKAB3E9EwAJMALGLmoQAWGZ9hfDcBwNkAxgJ4hYhSjXNNAXAHgKHG39gE5iUObRM4folEOFDt\n7VBZNQ4b8/gE+h2fJXCvcQL+fy+Y42tRhyg1kvos/Qb+PeEELVn7kYXPZjCciZQ0v12BpZzVDxYT\ndmhmPEWAmQuYebWxfRTAFgD9AYwHMNXYbSqA643t8QCmMXMVM+8GsAPASCLqB6ArMy/j6Bv+luWY\nhKMjhlsPczcUOOZRaYhTfz0X9/9rrcO+q+gYCksTt4qT5HCaq7eO7OyD7d8cSHkO2lMq6ChZv/bm\nwD5IT+oYYOK8FsGKAZKz51g46DhoEyCiDAAXAFgOoC8zmytI7AfQ19juD8A6Lj7XsPU3tu32ZqGh\nRWUqqusCzVmiNJ6NeUdw1zurA0+HPHNtvsN2xR8+xcgnFyUqaXK1XwoHBXxHHV1BfXZJrE9fsN9r\nClK3SKmR3I4U2ZV6jraKPPsN+wh2E3lCOHbdT2ofSRa+RYCIugB4H8D9zBy3KoJRsk9YVojoTiLK\nIqKsoiLnhF7+zlG/bb95t01dmZCpZ8MMM6O61n9ja01dBM9/tN1R4j9aGf2cd9g5rXJFdV1Sl/4M\nWhPwGxoJ2gAshb5atFTsKJ1G/zsdpXs7SOw4n25AEpcWbQz32Q5Sb4//HLR9RJojqNWHgwCAiNoh\nKgDvMPMMw3zACPHA+F9o2PMAWMeyDzBseca23e6AmV9l5kxmzuzTp4/fvMSn2bJtv6dLdxabv9Oo\nczcHFdV1ri/Aoi0HXNdu/WBdvuuSd4nijSV7cPrD8xwjNwuPVuLed9c4nP17q3Lx/EfZeGFRdpxd\ncgpVtXU465H5+N2cLa7ftwS1QUvmzg7kAPzPDx8Rqv2iOCSuw5MnQRtq/ZaKg8bB5VpY4t9Vvz2i\nRLuZNuH8fkv8pri05t5BBOB1AFuY+Y+Wr2YBmGhsTwQw02KfQETpRDQY0QbgFUboqJSIRhnnvMVy\nTMLxM4Gc/aYcPFaF6SubtkpUYznrkfm4b9oah/22qVkY96JzUex73l2DO99e5Wp3GyC171A5lros\niL11fylesjluAJixOhq5KyiJj8E/O38bZq3Lxwfr4sM15ipKlT4nBKusiT759lW5WrSfeMASn+/4\nuDQ9QEBH2xIjib1+q6k9l+pPawuNNVFcgiDXtNz3l7qOir2DHHb3LqJSV9Bkt1n6qQlcAuBmAFcQ\n0Vrj71oATwO4moiyAVxlfAYzbwIwHcBmAPMB3M3Mpme4C8BriDYW7wQgT2bSRPp0SY9tSw+B/aW+\n6x+r8eD765F72F97wb5D5Tjv0QWxOdNNIhHGK4t34Iitx0ptXQS3vLECK3a7D+GfnYBFuj9Yl49N\n+c61ai975hN832Uh82//eSn+sHC7o5+9ecnsVV7zitkdXf3+7gc4qs71J4ojmb1DvBoG/b6sUs2g\nTij9toruklKp2G+tSDpvQHEJGorzlwZ3uxTqCppmu1XuIux+/2OfW2s4iJm/YGZi5vOYebjxN5eZ\ni5n5SmYeysxXMfMhyzFPMPNpzHwGM8+z2LOY+Rzju59yMxb7+nXvENv2G3/cb/Q8qa2Lt6/dV4Jz\nf7MAxbbQyH/W5KG0shbvrcqNsy/ZeRDPzN+G/5sZ3xh64GgVPtte5CjxJzMsVWGU4KXQiB0WnHfM\n7FM0aoybYn9f/KYjEfidO6beHv9ZLvG7/544eEgqkTbDcyEWiHzaJXGQB4U1rRTudR4/BI79S4WD\nJnYd9srbcdE76HiiXWoKzuh7AgD5Iai13RXzJqTYPNlfFu/E0apaLNsVX4I3z2ufht0MdZTZFicx\nX8BkOj4J+wMeNEVBhcwUWnvNoTlqAkEdX1Pj4F4lft+OshnaBAL3UJIKUD7vd9DR0M0xLiFoTcvp\n7M393c8virvPkcHHQzjouOVP/zscQP1iIXbsN9t84J2lWXdnL4VAzJtqt0v722seycAuRBwTOFse\n4C5kJn7nlTcF2GlP/LXwPyjM3V5/HiEmbKO5RacpBC2By07Z/rxI53UXxKaWzoPge2BabAK5YPdf\nel6ceXb/QppCo6UItQiYN+m5D90X/PbbO0SMj0uOUqghSOepaY4iX0DEB1moE8gCB1/2WE3AZzqa\nQtBSbtAwkXO/6H/73vUxYfcSoiMdzSICAYXIZ2hMincnapR0U54LeTBe/GepxE8BxUFKqVQ4qB8s\nlhxCLQJ252zHURMwboZdHCIxR2Yv8Zu/E39ejtmFOLi9MbQFawLSg+gIjRm7OV4+j6qvw6nHQmC2\n2k/sWrjbE4nf3jcs2KWYcOz8wnPk+D0xJizYxbYF7+cl6NgDqZFUiv3b95fyFjT81ByDyJoaDpLs\nUq8hrx5VznBQ/OeXFmW7dgtvLkIuAg1/7wyBRP87HwJ3BxdrE0hxFwep9GtHqgk0x0Aqv6ER06k7\nrpHx39HZR8hznZC3GqEmIF2jplyLwOMBxJKjED5qYsk+cNdRH9ciaCw+aG8dSfj8holMu7NWJO3f\nhJpAwPCeo6eUV68hIezjdPbxtaUj5TWIRDju2jEz/rBwO657eYmUnYQTahHwmqTUXgI3b0aNrVhT\nX+L35+zrG4xtNYE6s0eMvzaB5oiP2/PmlQbJafgdLel1XqdoJP5aBC7ZiW0F8cebezkcYkBH61d0\nTPw0ygZu6A2cZruzZ1d7rJHc7+8FFD4/4uC/AViYKE6YQkOqIVqdfV2E8aeF21FaWROXjoPHqnD+\nYx/ihUXZcQ3GZuGoJbtKh1oEvGoCfieKivXq8dkmEBH2rxUcnywCiQ+N+HWI5qWxp02Maxr/neEd\n9zyLYtTEeLwbwbuC+nO+5nntu5uXzG8pt6m9kRJxzsKj7mv6+n1HzP3+uzY/Lt9eQua3d1jQnjlu\nv+X3WGlksFcjecGRCsd5524owAuLsvH0vK1x16KwNHq9F2zaH7e/+c57+a5EEnIRqL+Sy3dFp4qo\nsIxotYcqzHshhUDsxJy9zS61CdTGagLxSOGg5qgJSOd0dpeN4rfraCwc5Pv3YrIRZ5fCR0EF8ebX\n6wfGSWmQHNR+YaZSqUHX0bbg4fikMIHJuytysHV/aeCQjp99JEd2z7vxY1fMR3dH4bF4u3G/fjdn\nS2yUeCTCKDfeq6KjVfjSmJbF/M4NqZE8aInfeu3khtf67fj3Pxp+sddKIxFGVW0dcorLHfYjFTV4\nc8nuuN+qi0Sd/einPsaM1bl4a9keAEBpZU3sGlXVRCzTQ7CloEhxNYeaWveCZXPSZkTg0Q82AwCm\nLN4Rs326PX4aBfPGOqt30f9WZ8LMsc9SN0q7mlfXRVz3F0MgzdBg7DfcYl4Lyflad6+LMAqPGo7T\nd5uAe4knaE0g/mWs3/48u/7eTlm8M7Y9b0P9qOwjFfEjus3bIs14+sjMTbFJ9dbkHMbfPt8NIDpK\nm5mxdMdB5JVU4JkF26JpM477ZGshKmvq4tJaWVOHjElz8OaS3Y7BhpNnbMDY5z/Hws0HXNNhdeT5\nJc6J+ez7WGmox1FdhFFaWWOs1Bbd7/mPnFOKmOw/Er3nT83bgr8v2ROz55ZUYOIbK7DnYBkWbIrm\nYeWeQw7Haf3d0349F+8s34v5G+vnwzIdaDTd9ftbG02t+Xxl8U5kHzjq6BI+d0MBthSU4rPtRTjr\nkfkx++6DZfj59HU47ddz4/a/9c2V+NV76zHm2U9QVlUbE778I5V46D8b8OgHm5G197AlbYysPdHP\nP5u+DvsORe/JnPUF2JgXTeuKPcX4+b/XAQA25pVi5Z7omKOS8mrcPjUrdq4pn0af1ZYUgdCuMQzE\nhyDMzRc/rheBx2dvxm2XDgYA5B4ux8Fj1QCiN8bk0+1F+Gx7dCZTq5N5YVF27MG3NgxX1tTFXl5r\nVfdQWTX+nznXjyVdO4uO4et/+syR9sqaOlc7EH2oTapq65CeFl2zZ4xlZtT9RypxUrfoqGmrE8wv\nqUDPzu0BAHuL66e7WJtTgjNP6goA+O+aPOwsKnPk+dkFW2NTPC/bVYwbRkTnA3xiTr0TsLZ3bM4v\nxQP/Wue4FoB1sFi9raS8GmOfd18YvsYiiG8u2Y0fXhK9b0stpc7rXv4Cc+69zHHsG0t245FvDQMA\n/OSd1TH7D15bjj1PjwMAjP/zktj9t/LlzmJ8ur1+Jts1OYfx6AebsaWgfmqO3MMVeHz2FryxZHfc\nsXPWF+Cqs3LxwL/W4XsjB2JHYdQ57SoqwyrDiZiFExOrE3zWEBMg6mi/MqgnTurWATPX1s+7eOnv\nP8ZbP7oIqSmEQb06xezHKmvxwbp85JdUoFfn+ilUnpyzBc9PuACAs+T80H82YNrKfTirX1fkWcTl\nWFUtPt9ehMKjVZhjefbMo/9pW8t5xupcLNt1CF97bnHMdrSyFjNW5+F/jGfGdN75Ryrx76x9qIsw\nHvpPvPj+4t/rsP9IJcad1w8b8uod/7gXv8Dup64FM3DGw/VO/dkF22LXbK7lOfit7RqbuK1ZYfLJ\ntug9t86k++KibIw6tScAYPG2Qry5dA+AqC/4zoXus+KbU7iYwmBiThFTcCS+1vnh5uj9b0ENCLcI\nWJ1z4dHKWEjIjUt/X+9A73x7Vcw5THxjRcz+q/fX45qzT0JqCsWVfN74YjduvTgDKSmEx2ZvxtwN\n0Rs5f+N+PPddRmoK4c+f7ECpMa3yrqIylJRXo3un9lhtKVEAwJaCUpzVryt2Fh2Lm8Ezv6QCJ3fv\niEiEcZfFkX3rpS/w4QNfBQDkWNZIGPXUolgefj9/a8z+zZe+iNm/+uzimH3SjA2YMHIggPiX47ap\nWdj15LVISSH8+ZN6MXlvVS5+8rXTcFqfLlhgmc3UWrp+/qP68RmfbS/C8l3FuOjUXgDqp5g+UFqF\nNTmHccHAHo4R2W9/uQc3j85AZU1d3PKRj36wOSYCP7DMh7QpvxSllTWuXW6nZ+3Dqj2HHfbF2wox\nY3Ue1u0ribMfPFaFqtoIvve3ZXH2R2ZuwjaXwYdSl753V+wz/sc7Suu1sfLch9tc7X/9dBf+il0O\ne4SBm153zgl18dMfu57nv2vz8dC4YThQWokf2yYg/LdRI7EKHAA8+N662DNtpb4xON4u1WD3Fpdh\n4eYDGNizE+61hJ/eX53ruv/qvYeRf6QyrtRtMnjy3NiMAG5c6zKJYkMUHq2MhbSA+jDW8x9tx5cW\nv2EWRqwZGXgUAAAWqElEQVTvAgDsLXafbyxojcysIae2YKNAyMNB9dsHj1Xjf19dJu9swy2+eLSy\nFruKojFSawk5r6QCC7dES/87LTHUipo6vPXlHsf+AGKlHrv95tejomOPonzjhehDXW1rUN1+ID5m\n2xzYf9OkxJggz5qHd1fkYK3hUO0P+ixj5tGlOw/i7n/WC9njsze77m9OM13k0mh5uKzaEc4BgAsf\nW4gLHl/osD/43nr8K8s5Q+wP/74yli4rmb/7CJe4OFI3AQAQE3g7QXs7BVnDobF8/U+f4psvfRFX\n2gfktO4qKnO1L95WhIxJc2LzT8XOIzi4LfuP4o63snDN8/E13BpBNLyCodK9aAwjn4hfrMjMw9Qv\n98bZpQ4Nq1yECpAbtNfklLjazRqDhoMSRFMuZF2EkZbqPF5qMDUbnOyOzIyb2vcvrTQcqOOFcY/F\nmw6vqgWchB05jsyu32/MO4Lhp3R3bVQuKa/GEtuU1oyo83NrkH/542x0Tnc+pm6OHkjePEz20rOJ\nm1ABshPIaYEV74KuxywJ09OWGqYVKW/2WXVNisvceybZQyUtSbkwJXpQkbbXMP3Skr2DQi0CTRHT\n2gjDCLXHERtQ5hhtajpv9940zh4k0f92n1VvD1ZSbM4VumrqGAdcesxEGPg8u8jh7I9U1OD387ei\nuCw+xn64rBrDH3M67wgDpz88D53bx1/w6tqIOOXH8YK9d83xyK6D7jWBoA5xxR73KdTt8fLWzNb9\niat9NIR9AGpzEmoR8FMT+Dy7yLX0KJUoTWcv9auXBg7Z7eZ57Pbismq8uCjb9fe/yD6IY1XOsMPC\nzQdcF3Kf9P76uHimyYrdh1zbR5buOOhYGQwA/vbZLrz8yQ6H/bt/+dJhA+IbNK0s2lroajdLS2U+\nF6RRlLCj4aAE4edCmjF4O2tyDqNjO2dVYMrinbhldIarCBQfq3KU+F/7Yjc6tEtFvq1qu2RHMTIm\nzUG3ju0cv/HHhe6lX7cGQAC4460sV/s0YZW0G//q7rzdFp0B4CoAjaEl4t2KEgYOlVVj5to8jDu3\nH9JSm7fpllrTOrtuZGZmclaWu5PzoqTcPfygKErbJaNXJ+wRevO0NrY+PhYdXAqjfiCiVcyc6bVf\nqHsHec0dpCjK8c+tl2S42s8b0M3VfvWwvgn53bP6dU3IeRqiXTPXAoCQi0BLtrAripIc0t16cAC4\nZEhvV3v7tMS4vbNOch+ncPkZfVztZwr7SxC1zHiBUIuA1gSU44HeXdIxYlAP1+++erq7Q2mLDBNK\n3umCU28vlKLbpwYLr9xzxRBXu1sXcgBITXH/XalU369bB2RYRnvHzt9CpdhQi4DWBJTm5H8uHOBq\nD+q42wnOhAgY0KNj4HQlg3Hn9Wv235CEUirZp7cTREDY/84xp7rapdK41GAr3U9JNNJSyXVgXJog\nJokm5CKgKqA0H9LLLtklJOfQrgEncGqfzoF+o7lpJzjKxsTNJSctXifBLtYEhPP37tLe/XeFvEl5\nFkVDPE+w/CaaUIuAaoCSCE7q2sHVLlXvpRLc4N6dXZ/JdikprtOUNOQEvj3cfcKyE1xGVzcG6TxS\nLUcKgbQP6Mh6d2mP4ad0d9iJgFThhZautxgmChg+kkr8ck0g2HMh1hA0HNR0tCbQtrnqrGC9QCZ/\n40xX+82jB7naxZKp4GROPCHd1UE0xglIv/ErIQ8ST33nXFf7A1ef7mofemIX9/SIpfRgLiYtJcWx\nJgVgCKVwjFgTkMJEoji4txVI9yHofZP3DyYyiSbUIqAS0Lb5mtBL45kbznO19+6S7moXq/FSSVDY\nXwxzSKXotIYcXzAHd/FpvQLtL4lMasB4d9CeOA3FzSXEWkhAEZDsUngnaBgnqDhIz1GiCbUIpKZQ\n4G5ZSniQBtlIdtkBSc47MSVEyeFK4kAkn0vK27XnujfcSvtLDkh0fAF7xAw5sYtryEnaPzXFvjK3\n5beF6y11HQ3aViCHgxJTspefI60JNBkiwvz7xyQ7GUqScJv2A2ig9Ct2KQxWE2gwhuxStJccblqq\nu+Nrl5IitndJeeucHswhynkLJnCSY+3Zqb3rucRSdwMOUfztgM5eunZBa4JyHhJTyEg0oRYBpW3w\n0LVnudo7tnd/vMXSb+AGwGDVeLnh0b2LYO7hCle7tL4DAKQLeevYzr2hV+pGKffECXaNRJFJcz9/\nTnG5q8BV1ciTC0qCKDYAi11K3a9d4N4+AccPSDE/aTrrRNMmRUAaZq60bk4TukX27ebee6eDEA4Q\nS3yBY7nBHKLkTNyWtWwsQWsCkiCKghXQIUoOV3SIAmXVdYF7+wXtBSSW1BN0/6XCgbQIz36X6dub\ngzYpAt+5wH2Qj9K6ue58926RkuPr0D6Y45Pi3XKXv2BtAuLSPAmcxFFyZJ2EaxG0ZC8PnApWKpby\nHA2BSce4msX9g9YEgsbyg4uGIAIRFvPWErRJEUjU3CFK05BeljFCX3QpRCqdR2oT6JCgEIjchuBu\nr7OvGWrAkB1ccOcgXQv3cFDQ0a1Br4VklxZNksYCNAQL8hq8ATigUw/YSC6JlbSsZ0vNeNAmvOFn\nv7wcSyddEfssOYFkcskQ9y583zjnJFf7L685w9Xet6t7N8eg3JjpXlt69FvDXO0/F/qVXz/8ZFf7\n6X274OTu7lMidBFCF5J/kMaDSCV+qddI4MZQcboHd7u0lq7nYrpuvxFw/6A1gYbaL9yQagjSPZMc\nX4RZPCaoPsi9hqRwTcDaT9ASvyB80rVoKVqfN2wGBvbqFOdw+gojQBcIPYl+Izi+h8e5N0ie0de9\nW+qQE7ugV2f3oemSIxvoMrEUIM8p887tF7nan/vu+a72G0a4O/tvnOPepfCkbu6/Ky2HJz3edRH5\nZQ+McB5J7O3LWJpIzkFyoF2EUbWJ6q0CJG7Ue5cOwdIadIoDqQQv1WQkv9cYfyiVsKV3SppALmhI\nK2jNQVoCVhZEV3PCaRMiYEcqIUqOVRp52qOTu0O//MwTXe19u6a7vtSSMDSEPEOqu328UCJ/9Lqz\nXe3dOjlXPAPk+VW6Ck6mU3t3e+f0NPEllV7qoLPCSo6ph3C9+5zgXouSBpFJ91/smtrAhGZye4Hw\nhYj7AdIz313Ig2SXSsU1DYS6XO1CxqTQTiLpGLAQIDViB61dSTUBuUbpak44bUoEfvH1010XlLjv\nyqHo1D4VnW0lu9n3XIofXpyB/rawxciMngCcTsOcVdIe2vn+RQMBALdfdiqsTtoUnTGn9xEdnOgQ\nXa3ygyOVTqRS8Sk93Gsg3QVx6OqyTCYgh3a6pKe55uGE9LTEDfUWziO9pD0FcZBEwP68mARukA4Y\nkmkMoiAK91MKK7qtZQ0AJ3friH/c5qyFyj20Ul2f+Q7tUhsIByXmwThBKLBIwifVBKRnXhIZ6VpI\nYiIVJhJNmxKBn14xFH+7xbna2gNXn47Nj4112M/p3w2PXne2I9Tx0vcvwL1XDnUsWnHHmMFYOukK\nXDY0vmFzZEZP7Hl6HC4/48S4B/zac/vh/Z+MxtP/c26cvxp9alREvnlev7j9zRf2hPQ08UWxv+zv\n3H4RFtw/xvECmcJmt5uCZS/xm6Evezhozr2X4uphfR21pUnGHDZ2uykio0/tFeekzbWWR53WK+5a\nmOmQphEGEqcZkjOWHGUfQRy6C85BmpTtRLEGEryGKCFPV+1ud1v7GgBO6OBuv+bsk3Dp0N6OefGD\ndk1tyPElqheVVCCSarPS/ZGukRSG7CRci45CbTm0IkBEY4loGxHtIKJJLf37JvPuuwwvTBiOV35w\nYZx9kBCDtz4gJ56Qjp9dfTpSUyhuDnICxdoepIZb+6pDIwb1RHpaKkYO7hmzpaYQVj50Ff5443DH\ngzbn3kux6BdfxZjT++Dc/vXL5/3z9osw66eXYFCvTnjgqvpG2kuG9MYZxtQZj42vD/3Mve+yWGP5\nCxOGx+y/G38Otj4+1uEcbr/sVOx+6lpHHDyjV2f87ZZMR6k4c1APZD/xDVx0aq+4h/nnXz8Ds++5\nFHdfPiTOeV97bj9M//HouLQAQP8enfDJL76Gt340Ms6RtU9NwRVnnohffP10R2+M5757Pl75wYXo\n3Tk9roH7tVsyMcV2v4Ho7J5/uWkEAOCUnvUiN374yXjjh5lIS03BT752Wtwxlw7pjYG9OmHVw1dh\niG1StWvP7efaIP7TK4bg7sudC5Q8/Z3zcJJLyfudO0Y5bDEChgPTUlOw6bfXxNlONsZXuBWMiAhr\n/u9qh/268092OKfzBnSLlX6vOTu+I4MUDpRKy34dnzSR3YUDnTOQAvHzSFnDkFZBsHYFtRZerAUf\na23xhPT6dzPTUkix5mHk4J6xab87tU+Ntf/cPGoQfnTJYOM8aTjJuBeXDOkVm9SvsWsLByUx8876\nhIhSAfwZwNUAcgGsJKJZzLy5JdMBROc5d5vr/NNfXo4j5TUoqYgfwLPioatQWVOH0oraOAf562vP\nws2jBuGfK3Jwet/6B/Puy4fgxsxT8NyCbRhr6eHzu+vPxfXD++Om15fHNcr+5KunYVi/rrj1zZUY\nc3rvWKjpR5cMRnl1HV5clI0endvj7JPrHf8H91yKjElzAAAXW2ol9101FH/6aLsjb7eMzsAjMzcB\niJZiTIEZP7w/7pu2FkC0gbdDSvThe/XmEbjz7VWx4818L5l0BS55+mPDVn/+cef1w5z1BbF9zZL1\n7HsvxT+X52DO+gJcc3ZfnHhC9IGfctMIPDN/K3IOlePHY05FRu/oyzJmaB/MXl+AEzqk4fHxZ2Ow\nYb95VAZyDpVj+4FjuPvyIbFpjSMRxt2Xn4ZdRWX41vknx82T88wN52PEoB7o3SUdV1pe7Pd/Mhq5\nhyvQvVP7uOmRP3/wCnyeXYS9xeW4aVT97KG/GnsmvveVgdiyvzTO0fXqko4Zd12MTXmlGNCjI7p3\naofUFMLzEy7A7284D0t3FuPWv68EEC1F33fVUNx75RCs3VeCb7+yNHovOrXDMzecj68PO4B+3Tvg\nv2vysGrvYQzu3RlXnHki3l2Rgye/fS5yD5fjlcU7cdOogRh79kn4y+KdmHztmXhq7lbklVTEFQrS\n01LwtTP6YMGmAzFb5/Q0fP7g5Xjuw22YuTYft1ycASC65u7yX1+Jz7YX4ZfvrY8VInp0bo/Z91yK\nj7YcwPMfZeO0Pp2RkkJY88jVOFZVi5LyapRX12FQr/pBfA+OPRN3jDkVmb/7CADiBPLDB8bg63/6\nDEB8Lecft12Em15fDiA+xPq768/BdzMH4IyH58etn3DvlUNx/5VDceqv58aFM8cPPxkvTLgg9k6Y\nDD2xC968dWTMbq3Zb318LIY8NA92XpuYicc+2Iw3luyOK/is/r+rcd+0NZi5Nj+ugPbeTy7GS4uy\n8YeF2zGwZ31ap/94NGauzcN909ZiWL9umHffZcjacwj/+5WBqKqtQ2VtHe6/aihSifDp9iJ858IB\nOFZVi8kzNuD/2QoezQYzt9gfgNEAFlg+TwYwuaFjRowYwW2JA6UVHIlEHPa3vtzDuYfLHfZBv5rN\ng34127c9Y9JsPv2huQ77pPfX8/iXv3Cm50gFr8k57LDvLDzKf/10h8O+/0gF/2HBVq6rc+bBL5FI\nhA8erWz08a2N6Stz+NNthQ57SXk1l5RVex5fWVPb4Pf5JeU8bcVeZmYurajmMx6ey4u3FXJFdS3f\nP20Nz1qbF7f//iMV/N2/LOXiY1WOcy3JLuJa2707cKSCB/1qNv9m5kbPtJps21/Kc9fnMzPzyx9n\n84jHFzIz87p9h/ln/1rLxypr+LPthXzDlCVcWxfhOevz+fJnP+EDpRW8Ma+Ex734GR+rrGFm5gUb\nC/hAaQXvO1TGV/9xMR84UsHMzKv3HuL9Ryq4+FgVj3ryI96cf4SZmTfnH+G9B8u4qqaOvztlKWft\nKWZm5r0Hy3hX0TFmZp74xnJesLGAmZlzD5fzhtwSZmaePGM9v79qXywf5ru4o/Ao7ztUxszMFdW1\nvKUg+lsrdxfzZ9uj97a2LhJLw87Co7wkuyh2jrUu71BzAyCLffhl4hYcqkZENwAYy8y3G59vBnAR\nM/9UOiYzM5OzsrJaKonHHbPX5+Osfl1xWp/46vGM1bno160jRtumEK405mBpqaqmEg72HCxD/x4d\nA68PoCQPIlrFzM5Yn40WDQf5hYjuBHAnAAwcODDJqWndfPM8966f3xHWv1XnrzQGM1SnhI+WlvU8\nAKdYPg8wbHEw86vMnMnMmX36BFu0W1EURfFPS4vASgBDiWgwEbUHMAHArBZOg6IoimLQouEgZq4l\nop8CWAAgFcAbzLypJdOgKIqi1NPibQLMPBfA3Jb+XUVRFMWJNvUriqK0YVQEFEVR2jAqAoqiKG0Y\nFQFFUZQ2TIuOGG4MRFQEYG8jD+8N4GACk3M8oHluG2ie2wZNyfMgZvYcaNXqRaApEFGWn2HTYULz\n3DbQPLcNWiLPGg5SFEVpw6gIKIqitGHCLgKvJjsBSUDz3DbQPLcNmj3PoW4TUBRFURom7DUBRVEU\npQFCKQKtZR3jRENEpxDRJ0S0mYg2EdF9hr0nES0komzjfw/LMZON67CNiK6Rz966IaJUIlpDRLON\nz6HOMxF1J6L3iGgrEW0hotFtIM8PGM/1RiJ6l4g6hC3PRPQGERUS0UaLLXAeiWgEEW0wvnuR7IuC\nB8HP8mPH0x+is5PuBHAqgPYA1gEYlux0JShv/QBcaGyfAGA7gGEAngEwybBPAvB7Y3uYkf90AION\n65Ka7Hw0Mu8/A/BPALONz6HOM4CpAG43ttsD6B7mPAPoD2A3gI7G5+kAfhi2PAMYA+BCABsttsB5\nBLACwCgABGAegG80Nk1hrAmMBLCDmXcxczWAaQDGJzlNCYGZC5h5tbF9FMAWRF+e8Yg6DRj/rze2\nxwOYxsxVzLwbwA5Er89xBRENADAOwGsWc2jzTETdEHUWrwMAM1czcwlCnGeDNAAdiSgNQCcA+QhZ\nnpn5MwCHbOZAeSSifgC6MvMyjirCW5ZjAhNGEegPYJ/lc65hCxVElAHgAgDLAfRl5gLjq/0A+hrb\nYbkWzwN4EEDEYgtzngcDKALwdyME9hoRdUaI88zMeQCeA5ADoADAEWb+ECHOs4WgeexvbNvtjSKM\nIhB6iKgLgPcB3M/MpdbvjJJBaLp8EdE3ARQy8yppn7DlGdES8YUApjDzBQDKEA0TxAhbno04+HhE\nBfBkAJ2J6CbrPmHLsxvJyGMYRcDXOsbHK0TUDlEBeIeZZxjmA0YVEcb/QsMehmtxCYDriGgPoqG9\nK4joHwh3nnMB5DLzcuPze4iKQpjzfBWA3cxcxMw1AGYAuBjhzrNJ0DzmGdt2e6MIowiEdh1jowfA\n6wC2MPMfLV/NAjDR2J4IYKbFPoGI0oloMIChiDYoHTcw82RmHsDMGYjey4+Z+SaEO8/7AewjojMM\n05UANiPEeUY0DDSKiDoZz/mViLZ5hTnPJoHyaISOSololHGtbrEcE5xkt5Y3Uwv8tYj2nNkJ4KFk\npyeB+boU0ariegBrjb9rAfQCsAhANoCPAPS0HPOQcR22oQk9CFrDH4Cvob53UKjzDGA4gCzjXv8X\nQI82kOffAtgKYCOAtxHtFROqPAN4F9E2jxpEa3y3NSaPADKN67QTwMswBv425k9HDCuKorRhwhgO\nUhRFUXyiIqAoitKGURFQFEVpw6gIKIqitGFUBBRFUdowKgKKoihtGBUBRVGUNoyKgKIoShvm/wOi\nCcY8T14RdwAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f28c27a4590>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(evol[0])\n",
    "plt.figure()\n",
    "plt.plot(evol[1])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 2",
   "language": "python",
   "name": "python2"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.13"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}