ch07-卷积神经网络.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 简单的卷积网络实现\n",
    "\n",
    "https://github.com/oreilly-japan/deep-learning-from-scratch/blob/master/ch07/simple_convnet.py"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "# coding: utf-8\n",
    "import sys, os\n",
    "sys.path.append(os.pardir)  # 为导入父目录中的文件而进行的设置\n",
    "import pickle\n",
    "import numpy as np\n",
    "from collections import OrderedDict\n",
    "from common.layers import *\n",
    "from common.gradient import numerical_gradient\n",
    "\n",
    "\n",
    "class SimpleConvNet:\n",
    "    \"\"\"简单的ConvNet\n",
    "    conv - relu - pool - affine - relu - affine - softmax\n",
    "    \n",
    "    Parameters\n",
    "    ----------\n",
    "    input_size : 输入的大小（MNIST数据的输入大小为784）\n",
    "    hidden_size_list : 隐藏层神经元数量列表（e.g. [100, 100, 100]）\n",
    "    output_size : 输出的大小（MNIST数据的输出大小为10）\n",
    "    activation : 'relu' or 'sigmoid'\n",
    "    weight_init_std : 设置权重的标准差（e.g. 0.01）\n",
    "        使用'relu'或'he'时，指定为「He的初始值」\n",
    "        使用'sigmoid'或'xavier'时，指定为「Xavier的初始值」\n",
    "    \"\"\"\n",
    "    def __init__(self, input_dim=(1, 28, 28), \n",
    "                 conv_param={'filter_num':30, 'filter_size':5, 'pad':0, 'stride':1},\n",
    "                 hidden_size=100, output_size=10, weight_init_std=0.01):\n",
    "        filter_num = conv_param['filter_num']\n",
    "        filter_size = conv_param['filter_size']\n",
    "        filter_pad = conv_param['pad']\n",
    "        filter_stride = conv_param['stride']\n",
    "        input_size = input_dim[1]\n",
    "        conv_output_size = (input_size - filter_size + 2*filter_pad) / filter_stride + 1\n",
    "        pool_output_size = int(filter_num * (conv_output_size/2) * (conv_output_size/2))\n",
    "\n",
    "        # 权重初始化\n",
    "        self.params = {}\n",
    "        self.params['W1'] = weight_init_std * \\\n",
    "                            np.random.randn(filter_num, input_dim[0], filter_size, filter_size)\n",
    "        self.params['b1'] = np.zeros(filter_num)\n",
    "        self.params['W2'] = weight_init_std * \\\n",
    "                            np.random.randn(pool_output_size, hidden_size)\n",
    "        self.params['b2'] = np.zeros(hidden_size)\n",
    "        self.params['W3'] = weight_init_std * \\\n",
    "                            np.random.randn(hidden_size, output_size)\n",
    "        self.params['b3'] = np.zeros(output_size)\n",
    "\n",
    "        # 生成层\n",
    "        self.layers = OrderedDict()\n",
    "        self.layers['Conv1'] = Convolution(self.params['W1'], self.params['b1'],\n",
    "                                           conv_param['stride'], conv_param['pad'])\n",
    "        self.layers['Relu1'] = Relu()\n",
    "        self.layers['Pool1'] = Pooling(pool_h=2, pool_w=2, stride=2)\n",
    "        self.layers['Affine1'] = Affine(self.params['W2'], self.params['b2'])\n",
    "        self.layers['Relu2'] = Relu()\n",
    "        self.layers['Affine2'] = Affine(self.params['W3'], self.params['b3'])\n",
    "\n",
    "        self.last_layer = SoftmaxWithLoss()\n",
    "\n",
    "    def predict(self, x):\n",
    "        for layer in self.layers.values():\n",
    "            x = layer.forward(x)\n",
    "\n",
    "        return x\n",
    "\n",
    "    def loss(self, x, t):\n",
    "        \"\"\"求损失函数\n",
    "           参数x为输入数据、t为监督数据的标签\n",
    "        \"\"\"\n",
    "        y = self.predict(x)\n",
    "        return self.last_layer.forward(y, t)\n",
    "\n",
    "    def accuracy(self, x, t, batch_size=100):\n",
    "        if t.ndim != 1 : t = np.argmax(t, axis=1)\n",
    "        \n",
    "        acc = 0.0\n",
    "        \n",
    "        for i in range(int(x.shape[0] / batch_size)):\n",
    "            tx = x[i*batch_size:(i+1)*batch_size]\n",
    "            tt = t[i*batch_size:(i+1)*batch_size]\n",
    "            y = self.predict(tx)\n",
    "            y = np.argmax(y, axis=1)\n",
    "            acc += np.sum(y == tt) \n",
    "        \n",
    "        return acc / x.shape[0]\n",
    "\n",
    "    def numerical_gradient(self, x, t):\n",
    "        \"\"\"求梯度（数值微分）\n",
    "        Parameters\n",
    "        ----------\n",
    "        x : 输入数据\n",
    "        t : 监督数据的标签\n",
    "        Returns\n",
    "        -------\n",
    "        保存有各层梯度的字典变量\n",
    "            grads['W1']、grads['W2']、...为各层的权重\n",
    "            grads['b1']、grads['b2']、...为各层的偏置\n",
    "        \"\"\"\n",
    "        loss_w = lambda w: self.loss(x, t)\n",
    "\n",
    "        grads = {}\n",
    "        for idx in (1, 2, 3):\n",
    "            grads['W' + str(idx)] = numerical_gradient(loss_w, self.params['W' + str(idx)])\n",
    "            grads['b' + str(idx)] = numerical_gradient(loss_w, self.params['b' + str(idx)])\n",
    "\n",
    "        return grads\n",
    "\n",
    "    def gradient(self, x, t):\n",
    "        \"\"\"求梯度（误差反向传播法）\n",
    "        Parameters\n",
    "        ----------\n",
    "        x : 输入数据\n",
    "        t : 监督数据的标签\n",
    "        Returns\n",
    "        -------\n",
    "        保存有各层梯度的字典变量\n",
    "            grads['W1']、grads['W2']、...为各层的权重\n",
    "            grads['b1']、grads['b2']、...为各层的偏置\n",
    "        \"\"\"\n",
    "        # forward\n",
    "        self.loss(x, t)\n",
    "\n",
    "        # backward\n",
    "        dout = 1\n",
    "        dout = self.last_layer.backward(dout)\n",
    "\n",
    "        layers = list(self.layers.values())\n",
    "        layers.reverse()\n",
    "        for layer in layers:\n",
    "            dout = layer.backward(dout)\n",
    "\n",
    "        # 设置\n",
    "        grads = {}\n",
    "        grads['W1'], grads['b1'] = self.layers['Conv1'].dW, self.layers['Conv1'].db\n",
    "        grads['W2'], grads['b2'] = self.layers['Affine1'].dW, self.layers['Affine1'].db\n",
    "        grads['W3'], grads['b3'] = self.layers['Affine2'].dW, self.layers['Affine2'].db\n",
    "\n",
    "        return grads\n",
    "        \n",
    "    def save_params(self, file_name=\"params.pkl\"):\n",
    "        params = {}\n",
    "        for key, val in self.params.items():\n",
    "            params[key] = val\n",
    "        with open(file_name, 'wb') as f:\n",
    "            pickle.dump(params, f)\n",
    "\n",
    "    def load_params(self, file_name=\"params.pkl\"):\n",
    "        with open(file_name, 'rb') as f:\n",
    "            params = pickle.load(f)\n",
    "        for key, val in params.items():\n",
    "            self.params[key] = val\n",
    "\n",
    "        for i, key in enumerate(['Conv1', 'Affine1', 'Affine2']):\n",
    "            self.layers[key].W = self.params['W' + str(i+1)]\n",
    "            self.layers[key].b = self.params['b' + str(i+1)]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 训练卷积网络\n",
    "\n",
    "https://github.com/oreilly-japan/deep-learning-from-scratch/blob/master/ch07/train_convnet.py"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "train loss:2.298363787083355\n",
      "=== epoch:1, train acc:0.236, test acc:0.242 ===\n",
      "train loss:2.296706295857967\n",
      "train loss:2.292691633503392\n",
      "train loss:2.283486646299622\n",
      "train loss:2.2712700076618564\n",
      "train loss:2.2683018929074716\n",
      "train loss:2.245100547565853\n",
      "train loss:2.2322225204352915\n",
      "train loss:2.1991232390015876\n",
      "train loss:2.1727214016885195\n",
      "train loss:2.131718501610873\n",
      "train loss:2.127501808556071\n",
      "train loss:2.0289674452738438\n",
      "train loss:1.9551985141533075\n",
      "train loss:1.9742221542354057\n",
      "train loss:1.9149409338213055\n",
      "train loss:1.8301853280415825\n",
      "train loss:1.7530020867317313\n",
      "train loss:1.6647660060671734\n",
      "train loss:1.601356028969523\n",
      "train loss:1.5679587230348904\n",
      "train loss:1.4159968499373827\n",
      "train loss:1.3169464052797568\n",
      "train loss:1.2086799726708408\n",
      "train loss:1.242231169600962\n",
      "train loss:1.0944993978961328\n",
      "train loss:1.005298101248119\n",
      "train loss:1.042028790923519\n",
      "train loss:1.0455733782610912\n",
      "train loss:0.8306126466285181\n",
      "train loss:0.8416701810830282\n",
      "train loss:0.7874273039678238\n",
      "train loss:0.8051545466331909\n",
      "train loss:0.9075315409346397\n",
      "train loss:0.602938349254673\n",
      "train loss:0.6711749978913139\n",
      "train loss:0.5641790273829502\n",
      "train loss:0.572075077298855\n",
      "train loss:0.6628985513888829\n",
      "train loss:0.6303626017157182\n",
      "train loss:0.5705140103298275\n",
      "train loss:0.5680071762322668\n",
      "train loss:0.537882047287877\n",
      "train loss:0.5032938668598487\n",
      "train loss:0.4718117293008968\n",
      "train loss:0.71929854189441\n",
      "train loss:0.5271214594997665\n",
      "train loss:0.5292196171172581\n",
      "train loss:0.5491514015809268\n",
      "train loss:0.41313500334587877\n",
      "train loss:0.4234761721163104\n",
      "=== epoch:2, train acc:0.827, test acc:0.812 ===\n",
      "train loss:0.5321294254062401\n",
      "train loss:0.4420064011839947\n",
      "train loss:0.7256431392904311\n",
      "train loss:0.3660459441452637\n",
      "train loss:0.41920311462811655\n",
      "train loss:0.36146796600276715\n",
      "train loss:0.5908673922077924\n",
      "train loss:0.3549411361481255\n",
      "train loss:0.3686723222510694\n",
      "train loss:0.4120300197080942\n",
      "train loss:0.49118488397532756\n",
      "train loss:0.3623153947204574\n",
      "train loss:0.4886499025016353\n",
      "train loss:0.5200909358657938\n",
      "train loss:0.42101952136773235\n",
      "train loss:0.32172856696802027\n",
      "train loss:0.4340217312811027\n",
      "train loss:0.560223101203597\n",
      "train loss:0.45601290427853075\n",
      "train loss:0.42935082811464764\n",
      "train loss:0.4426042825080424\n",
      "train loss:0.36234255786493774\n",
      "train loss:0.3762799247107819\n",
      "train loss:0.4062171562570029\n",
      "train loss:0.3163157791844462\n",
      "train loss:0.6196981788449392\n",
      "train loss:0.31535106032856697\n",
      "train loss:0.24547779460894495\n",
      "train loss:0.4717873083919109\n",
      "train loss:0.5504117565529426\n",
      "train loss:0.3533188168550381\n",
      "train loss:0.4607348507095405\n",
      "train loss:0.38927711631572026\n",
      "train loss:0.3252222977206538\n",
      "train loss:0.3818219008722343\n",
      "train loss:0.5664601834076103\n",
      "train loss:0.7118077451757548\n",
      "train loss:0.35952043579753884\n",
      "train loss:0.46544516188055723\n",
      "train loss:0.39389715387228036\n",
      "train loss:0.3464678513281301\n",
      "train loss:0.3089527579374802\n",
      "train loss:0.27566931581725723\n",
      "train loss:0.41619742623245065\n",
      "train loss:0.33980794420344124\n",
      "train loss:0.2383177959493215\n",
      "train loss:0.3610519785041161\n",
      "train loss:0.3432861025992972\n",
      "train loss:0.5333508079466371\n",
      "train loss:0.33139493867504805\n",
      "=== epoch:3, train acc:0.867, test acc:0.859 ===\n",
      "train loss:0.27045435869636114\n",
      "train loss:0.2909785537863987\n",
      "train loss:0.42472940574154217\n",
      "train loss:0.4175268278034292\n",
      "train loss:0.3816255160974491\n",
      "train loss:0.37928997535255904\n",
      "train loss:0.2950591489213768\n",
      "train loss:0.3909527784080565\n",
      "train loss:0.332704269788745\n",
      "train loss:0.34753332546569365\n",
      "train loss:0.43094281003430945\n",
      "train loss:0.3618725699637907\n",
      "train loss:0.2619643184487527\n",
      "train loss:0.3760109393351989\n",
      "train loss:0.38118774302776304\n",
      "train loss:0.1944582664637024\n",
      "train loss:0.4128512394620514\n",
      "train loss:0.23089653822393497\n",
      "train loss:0.34275187930000633\n",
      "train loss:0.35426545933894393\n",
      "train loss:0.5550644364827926\n",
      "train loss:0.277170516291244\n",
      "train loss:0.2547525273433443\n",
      "train loss:0.3274614068733391\n",
      "train loss:0.20376591893614346\n",
      "train loss:0.2897222972867181\n",
      "train loss:0.30239657448451973\n",
      "train loss:0.2960438376144627\n",
      "train loss:0.25557039789694047\n",
      "train loss:0.4055199177951174\n",
      "train loss:0.21888031863296714\n",
      "train loss:0.24913619757812466\n",
      "train loss:0.2988272109600734\n",
      "train loss:0.3033363503742066\n",
      "train loss:0.24968129046211257\n",
      "train loss:0.3695628195651042\n",
      "train loss:0.2970865437136062\n",
      "train loss:0.30257276067092076\n",
      "train loss:0.33764735976717986\n",
      "train loss:0.27187510425026734\n",
      "train loss:0.45450293359030675\n",
      "train loss:0.294854501124109\n",
      "train loss:0.43094399085055407\n",
      "train loss:0.47082123195610187\n",
      "train loss:0.28343485051691797\n",
      "train loss:0.30686421941292114\n",
      "train loss:0.38982010030263126\n",
      "train loss:0.26197345159591523\n",
      "train loss:0.35754556307105806\n",
      "train loss:0.24832493106972017\n",
      "=== epoch:4, train acc:0.894, test acc:0.876 ===\n",
      "train loss:0.22167678147684974\n",
      "train loss:0.4036333758731914\n",
      "train loss:0.31253738853085\n",
      "train loss:0.2950212371720269\n",
      "train loss:0.3664054362748233\n",
      "train loss:0.2811695597094662\n",
      "train loss:0.2821138213847936\n",
      "train loss:0.3673064414484862\n",
      "train loss:0.4167718284168142\n",
      "train loss:0.3045152415887788\n",
      "train loss:0.16405904546570452\n",
      "train loss:0.20494304256293247\n",
      "train loss:0.2202536725240931\n",
      "train loss:0.5000483376304359\n",
      "train loss:0.28922957882631445\n",
      "train loss:0.36642882618316613\n",
      "train loss:0.22331044276760909\n",
      "train loss:0.2817264839557146\n",
      "train loss:0.2068207227473324\n",
      "train loss:0.34989249862862565\n",
      "train loss:0.1718396695704357\n",
      "train loss:0.3869162674887516\n",
      "train loss:0.2923169702068286\n",
      "train loss:0.24425537120109714\n",
      "train loss:0.3101690546638858\n",
      "train loss:0.20798603251682962\n",
      "train loss:0.22056246234697852\n",
      "train loss:0.2861807810143322\n",
      "train loss:0.21416516408968722\n",
      "train loss:0.30781377635706325\n",
      "train loss:0.4255179349984363\n",
      "train loss:0.3061542224621116\n",
      "train loss:0.150077317773754\n",
      "train loss:0.15211358244941545\n",
      "train loss:0.2933732241496615\n",
      "train loss:0.3232918769435705\n",
      "train loss:0.1355986117879442\n",
      "train loss:0.10298221152894707\n",
      "train loss:0.40143445041091097\n",
      "train loss:0.2265269240258579\n",
      "train loss:0.2437546305780294\n",
      "train loss:0.2506985332660804\n",
      "train loss:0.18502109810811893\n",
      "train loss:0.37549129154811\n",
      "train loss:0.34509661165668093\n",
      "train loss:0.22251341693723364\n",
      "train loss:0.20046673343620106\n",
      "train loss:0.18350023211631225\n",
      "train loss:0.2507516968056417\n",
      "train loss:0.16498192212803672\n",
      "=== epoch:5, train acc:0.902, test acc:0.89 ===\n",
      "train loss:0.2115959368909907\n",
      "train loss:0.33272571207640317\n",
      "train loss:0.33273909365424315\n",
      "train loss:0.10988477370234474\n",
      "train loss:0.3329719941026648\n",
      "train loss:0.15103622505871522\n",
      "train loss:0.17515074698098274\n",
      "train loss:0.2784683036042123\n",
      "train loss:0.1748501098551544\n",
      "train loss:0.27461999340684967\n",
      "train loss:0.15154510081015723\n",
      "train loss:0.27077690099603513\n",
      "train loss:0.3531982895706138\n",
      "train loss:0.21277598260710512\n",
      "train loss:0.32298688235917544\n",
      "train loss:0.13109553288419812\n",
      "train loss:0.24786920994027029\n",
      "train loss:0.27818300925104456\n",
      "train loss:0.25006680807487647\n",
      "train loss:0.21709443263942252\n",
      "train loss:0.3036099721758572\n",
      "train loss:0.2966055567726944\n",
      "train loss:0.21646536398355568\n",
      "train loss:0.16358092297290902\n",
      "train loss:0.19570657007428355\n",
      "train loss:0.27257288176501676\n",
      "train loss:0.18504512253432337\n",
      "train loss:0.24212770554787283\n",
      "train loss:0.2758954182358904\n",
      "train loss:0.24624424019682387\n",
      "train loss:0.2075989514580132\n",
      "train loss:0.2555280603806696\n",
      "train loss:0.1987002770763729\n",
      "train loss:0.14768354700002082\n",
      "train loss:0.252433645490588\n",
      "train loss:0.22379585883457456\n",
      "train loss:0.2353372054278557\n",
      "train loss:0.31693950784486413\n",
      "train loss:0.2792401695900329\n",
      "train loss:0.286537395297531\n",
      "train loss:0.2622717991456154\n",
      "train loss:0.11066089270642113\n",
      "train loss:0.2015502837625691\n",
      "train loss:0.1637551728763275\n",
      "train loss:0.224217577465566\n",
      "train loss:0.25005323807689184\n",
      "train loss:0.22339935221642038\n",
      "train loss:0.22429272570226047\n",
      "train loss:0.2048177726245227\n",
      "train loss:0.18379116695484435\n",
      "=== epoch:6, train acc:0.927, test acc:0.91 ===\n",
      "train loss:0.20535718940564743\n",
      "train loss:0.21303839474071637\n",
      "train loss:0.25606987063522885\n",
      "train loss:0.30317151805087994\n",
      "train loss:0.20312523754868853\n",
      "train loss:0.2429395675395467\n",
      "train loss:0.14090653494090355\n",
      "train loss:0.13997535038815684\n",
      "train loss:0.19104572431169395\n",
      "train loss:0.16620453948790534\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "train loss:0.2589129122172732\n",
      "train loss:0.162236678305782\n",
      "train loss:0.1465414594767902\n",
      "train loss:0.2607771342947665\n",
      "train loss:0.26365685578634496\n",
      "train loss:0.21342857932053505\n",
      "train loss:0.16173510261501522\n",
      "train loss:0.21744686805112515\n",
      "train loss:0.2513207942136274\n",
      "train loss:0.15552116438257538\n",
      "train loss:0.2569810946368218\n",
      "train loss:0.21920676187214827\n",
      "train loss:0.09200889275801635\n",
      "train loss:0.14614299390323204\n",
      "train loss:0.22191125720633537\n",
      "train loss:0.2267114343839381\n",
      "train loss:0.11678994673482845\n",
      "train loss:0.34108313551352815\n",
      "train loss:0.17426178007735643\n",
      "train loss:0.27874995391253826\n",
      "train loss:0.13216643458261168\n",
      "train loss:0.19409784332765517\n",
      "train loss:0.23895014341809886\n",
      "train loss:0.1339855160020341\n",
      "train loss:0.345054321391781\n",
      "train loss:0.17454927304604695\n",
      "train loss:0.17639362141232492\n",
      "train loss:0.23469079020563288\n",
      "train loss:0.14699044516076448\n",
      "train loss:0.2337049835131188\n",
      "train loss:0.16597438539501583\n",
      "train loss:0.1948590446619291\n",
      "train loss:0.13415290026525725\n",
      "train loss:0.25379194175127306\n",
      "train loss:0.30146466618526113\n",
      "train loss:0.19340068773542476\n",
      "train loss:0.13570578658447205\n",
      "train loss:0.1325575194461198\n",
      "train loss:0.19100717032810313\n",
      "train loss:0.161439619571715\n",
      "=== epoch:7, train acc:0.933, test acc:0.918 ===\n",
      "train loss:0.2585269188412215\n",
      "train loss:0.19399947866627554\n",
      "train loss:0.1706045902767645\n",
      "train loss:0.18316346536338815\n",
      "train loss:0.11772007279670721\n",
      "train loss:0.2629934342318727\n",
      "train loss:0.1864162464827461\n",
      "train loss:0.3195156943590111\n",
      "train loss:0.22587568050794618\n",
      "train loss:0.11355077084391602\n",
      "train loss:0.12269775732959999\n",
      "train loss:0.19910998752602893\n",
      "train loss:0.10725760434080428\n",
      "train loss:0.13320759936262724\n",
      "train loss:0.15175196803201305\n",
      "train loss:0.13996330983382985\n",
      "train loss:0.1747217574761367\n",
      "train loss:0.07079534782354871\n",
      "train loss:0.20761930060813977\n",
      "train loss:0.14765656351341855\n",
      "train loss:0.25707801257828233\n",
      "train loss:0.2589458145252898\n",
      "train loss:0.22070724011125903\n",
      "train loss:0.11881616141081398\n",
      "train loss:0.21724073685218317\n",
      "train loss:0.20597400353743164\n",
      "train loss:0.11899483607087485\n",
      "train loss:0.2332029024877637\n",
      "train loss:0.1420453133020976\n",
      "train loss:0.17495520736405884\n",
      "train loss:0.19485121388546328\n",
      "train loss:0.2239328856945584\n",
      "train loss:0.21948202921973148\n",
      "train loss:0.18898531687365674\n",
      "train loss:0.12280277524543216\n",
      "train loss:0.12776991793767373\n",
      "train loss:0.08343877909169649\n",
      "train loss:0.18130117188196798\n",
      "train loss:0.171468150397762\n",
      "train loss:0.2645256946961994\n",
      "train loss:0.176422601624333\n",
      "train loss:0.15232876102696646\n",
      "train loss:0.09306391169676735\n",
      "train loss:0.21832142795179565\n",
      "train loss:0.13864486828587386\n",
      "train loss:0.23771345697188045\n",
      "train loss:0.12070130600322733\n",
      "train loss:0.2793786217825137\n",
      "train loss:0.09795784524600379\n",
      "train loss:0.12450156871575327\n",
      "=== epoch:8, train acc:0.946, test acc:0.928 ===\n",
      "train loss:0.15954412570721913\n",
      "train loss:0.17699809030605299\n",
      "train loss:0.08655294544495172\n",
      "train loss:0.21966206035616398\n",
      "train loss:0.10297373854251238\n",
      "train loss:0.12533377163957748\n",
      "train loss:0.10288623703870732\n",
      "train loss:0.1716911023955802\n",
      "train loss:0.13427770058785474\n",
      "train loss:0.10247235544207037\n",
      "train loss:0.17403275685098113\n",
      "train loss:0.1374497696038621\n",
      "train loss:0.15950067793462414\n",
      "train loss:0.14767145810071752\n",
      "train loss:0.14902620847909712\n",
      "train loss:0.17850858283377313\n",
      "train loss:0.11731323789636783\n",
      "train loss:0.11627416446230612\n",
      "train loss:0.10251218986574907\n",
      "train loss:0.1346191891093086\n",
      "train loss:0.15109855439716105\n",
      "train loss:0.09597164654390071\n",
      "train loss:0.15445166924970155\n",
      "train loss:0.12223767759601023\n",
      "train loss:0.1075432018709235\n",
      "train loss:0.10238798367577034\n",
      "train loss:0.15879097740255252\n",
      "train loss:0.14708254115484293\n",
      "train loss:0.08343078007235549\n",
      "train loss:0.13814840486714927\n",
      "train loss:0.10670301550348556\n",
      "train loss:0.2745434549474264\n",
      "train loss:0.05312706770780243\n",
      "train loss:0.13707236291163383\n",
      "train loss:0.11168636712435917\n",
      "train loss:0.11893472524384724\n",
      "train loss:0.26717265759656383\n",
      "train loss:0.1926810744084294\n",
      "train loss:0.09357736673807512\n",
      "train loss:0.1802802081824247\n",
      "train loss:0.1436494282113467\n",
      "train loss:0.19540538675908362\n",
      "train loss:0.12834904707028594\n",
      "train loss:0.18191959409661826\n",
      "train loss:0.19185892421309092\n",
      "train loss:0.10190748982768198\n",
      "train loss:0.09232282130876887\n",
      "train loss:0.1119104279536281\n",
      "train loss:0.12765516901486326\n",
      "train loss:0.10111528676334976\n",
      "=== epoch:9, train acc:0.95, test acc:0.916 ===\n",
      "train loss:0.13842957682795454\n",
      "train loss:0.21269193608151718\n",
      "train loss:0.27660258793312187\n",
      "train loss:0.1763182868368385\n",
      "train loss:0.19254612089919873\n",
      "train loss:0.1475583139981719\n",
      "train loss:0.14472373237711544\n",
      "train loss:0.09971188624067638\n",
      "train loss:0.11573481633424672\n",
      "train loss:0.20138091789818455\n",
      "train loss:0.13651701354058518\n",
      "train loss:0.08422138199080771\n",
      "train loss:0.16397334353369963\n",
      "train loss:0.0800948140197632\n",
      "train loss:0.16992819649524674\n",
      "train loss:0.13420826678221262\n",
      "train loss:0.2281233828923441\n",
      "train loss:0.08981512526786553\n",
      "train loss:0.11677294304014132\n",
      "train loss:0.0741087454976897\n",
      "train loss:0.18231812192111765\n",
      "train loss:0.16078303106405678\n",
      "train loss:0.09103905241212963\n",
      "train loss:0.18303923680277237\n",
      "train loss:0.12514406762490346\n",
      "train loss:0.12244146319204349\n",
      "train loss:0.0940680219759064\n",
      "train loss:0.04127041068887755\n",
      "train loss:0.13710335311245195\n",
      "train loss:0.04557158074524939\n",
      "train loss:0.12867446212571407\n",
      "train loss:0.08193497609595889\n",
      "train loss:0.09720082911611475\n",
      "train loss:0.13345459462518008\n",
      "train loss:0.09653612226642375\n",
      "train loss:0.0748323360714889\n",
      "train loss:0.07526701684358182\n",
      "train loss:0.0844084940796348\n",
      "train loss:0.07716458912713266\n",
      "train loss:0.055302869130076385\n",
      "train loss:0.07924472044608247\n",
      "train loss:0.11065677287691207\n",
      "train loss:0.14638254571902437\n",
      "train loss:0.07275614789549936\n",
      "train loss:0.05616620455702076\n",
      "train loss:0.12424413523156425\n",
      "train loss:0.07085985756931697\n",
      "train loss:0.11419230831968198\n",
      "train loss:0.08031413241656836\n",
      "train loss:0.09397283346383098\n",
      "=== epoch:10, train acc:0.958, test acc:0.935 ===\n",
      "train loss:0.08016137465991555\n",
      "train loss:0.13698395014677323\n",
      "train loss:0.09531695805095536\n",
      "train loss:0.05974630525970996\n",
      "train loss:0.1485499163554529\n",
      "train loss:0.21878915749809758\n",
      "train loss:0.19865835276878793\n",
      "train loss:0.10203892880513166\n",
      "train loss:0.033081341979438586\n",
      "train loss:0.11915791651036768\n",
      "train loss:0.0853543832158285\n",
      "train loss:0.07949262860028974\n",
      "train loss:0.08185515735333385\n",
      "train loss:0.12193010374947429\n",
      "train loss:0.14403273467564653\n",
      "train loss:0.10348712663772527\n",
      "train loss:0.10886459436219122\n",
      "train loss:0.08948267084226512\n",
      "train loss:0.13204390290822104\n",
      "train loss:0.10011184212516891\n",
      "train loss:0.09938282169492928\n",
      "train loss:0.058174753622183356\n",
      "train loss:0.16034856045294574\n",
      "train loss:0.06027683869530968\n",
      "train loss:0.08203869581646007\n",
      "train loss:0.12733371507636154\n",
      "train loss:0.10092256416841362\n",
      "train loss:0.10734858425619881\n",
      "train loss:0.09091464163234515\n",
      "train loss:0.057590032759871713\n",
      "train loss:0.12222508948769605\n",
      "train loss:0.07807634402384221\n",
      "train loss:0.0931579890685339\n",
      "train loss:0.10870502640039939\n",
      "train loss:0.11857696939977291\n",
      "train loss:0.13403517376632784\n",
      "train loss:0.07573325544471358\n",
      "train loss:0.08675980983431514\n",
      "train loss:0.0798993624182201\n",
      "train loss:0.12558050699369327\n",
      "train loss:0.1186436721781443\n",
      "train loss:0.08752048515721077\n",
      "train loss:0.15890515665514388\n",
      "train loss:0.11396531851624124\n",
      "train loss:0.07197736805837518\n",
      "train loss:0.04873809652856913\n",
      "train loss:0.07249691583300528\n",
      "train loss:0.12747785864674596\n",
      "train loss:0.0836603637663641\n",
      "train loss:0.04644588456042978\n",
      "=== epoch:11, train acc:0.962, test acc:0.942 ===\n",
      "train loss:0.06526768052391571\n",
      "train loss:0.09800491364016914\n",
      "train loss:0.06767978698244242\n",
      "train loss:0.09889416732040926\n",
      "train loss:0.10740169833206467\n",
      "train loss:0.20211253997703238\n",
      "train loss:0.09877488461367076\n",
      "train loss:0.11909188520423034\n",
      "train loss:0.11067748650818458\n",
      "train loss:0.1266876973461307\n",
      "train loss:0.17506039632354461\n",
      "train loss:0.12941171587173028\n",
      "train loss:0.06256867922411337\n",
      "train loss:0.10957509387429136\n",
      "train loss:0.14542090086099077\n",
      "train loss:0.08580130210467769\n",
      "train loss:0.04869188104755333\n",
      "train loss:0.08492317229869352\n",
      "train loss:0.16826265710009702\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "train loss:0.11035769641299797\n",
      "train loss:0.11564225016294692\n",
      "train loss:0.16940180230802773\n",
      "train loss:0.1089501633281955\n",
      "train loss:0.1254712282190877\n",
      "train loss:0.15559382586658102\n",
      "train loss:0.04265389809342625\n",
      "train loss:0.1235317708471172\n",
      "train loss:0.2111467424059652\n",
      "train loss:0.08581750351318358\n",
      "train loss:0.06732561779345562\n",
      "train loss:0.07553519369524685\n",
      "train loss:0.11891045931924417\n",
      "train loss:0.10107521274056828\n",
      "train loss:0.09146939665171198\n",
      "train loss:0.07102665580097742\n",
      "train loss:0.08610981805256086\n",
      "train loss:0.06847414165908443\n",
      "train loss:0.10361732677026086\n",
      "train loss:0.08597233820396025\n",
      "train loss:0.04777052161264324\n",
      "train loss:0.0545511317064051\n",
      "train loss:0.07153014425887706\n",
      "train loss:0.08898243636535375\n",
      "train loss:0.13523610800735716\n",
      "train loss:0.06128331459033694\n",
      "train loss:0.04361672127039994\n",
      "train loss:0.060184590812242374\n",
      "train loss:0.06982835800597037\n",
      "train loss:0.1041844420214919\n",
      "train loss:0.08015925256074027\n",
      "=== epoch:12, train acc:0.96, test acc:0.937 ===\n",
      "train loss:0.04415440178219812\n",
      "train loss:0.03625669810513213\n",
      "train loss:0.05037521777753533\n",
      "train loss:0.01683795393009731\n",
      "train loss:0.13960404235770313\n",
      "train loss:0.08548167621199264\n",
      "train loss:0.11022851707228952\n",
      "train loss:0.1332061226538069\n",
      "train loss:0.15023723246459564\n",
      "train loss:0.14942385151876045\n",
      "train loss:0.11962571947473488\n",
      "train loss:0.09704928524134926\n",
      "train loss:0.04175089586793167\n",
      "train loss:0.08091352962157915\n",
      "train loss:0.06268097740334772\n",
      "train loss:0.06424337056139201\n",
      "train loss:0.017257551289748846\n",
      "train loss:0.08087659690589852\n",
      "train loss:0.08471816514016811\n",
      "train loss:0.018025846377140397\n",
      "train loss:0.08520099662490166\n",
      "train loss:0.042251821197699935\n",
      "train loss:0.04492060046148439\n",
      "train loss:0.08094160649227686\n",
      "train loss:0.11717795300819152\n",
      "train loss:0.10857790301434089\n",
      "train loss:0.15354464127909043\n",
      "train loss:0.05427067963952445\n",
      "train loss:0.04571902656496287\n",
      "train loss:0.05216002113496615\n",
      "train loss:0.13273607114542865\n",
      "train loss:0.0805616209044171\n",
      "train loss:0.0349630075201572\n",
      "train loss:0.07798508765530458\n",
      "train loss:0.053899200828936136\n",
      "train loss:0.044556010222855075\n",
      "train loss:0.11859000411023186\n",
      "train loss:0.1120745769558936\n",
      "train loss:0.19009611608174545\n",
      "train loss:0.04450886429653881\n",
      "train loss:0.07984743578092633\n",
      "train loss:0.12465855187955519\n",
      "train loss:0.1077369765364738\n",
      "train loss:0.10502782397810685\n",
      "train loss:0.09333046944568832\n",
      "train loss:0.04102711748847552\n",
      "train loss:0.045843473359623195\n",
      "train loss:0.08825012433122739\n",
      "train loss:0.036246452782605715\n",
      "train loss:0.08679885793188914\n",
      "=== epoch:13, train acc:0.964, test acc:0.93 ===\n",
      "train loss:0.04798652057921897\n",
      "train loss:0.09669766264989779\n",
      "train loss:0.05109177529463369\n",
      "train loss:0.0913160129937928\n",
      "train loss:0.05655341554195293\n",
      "train loss:0.04222834089995632\n",
      "train loss:0.06821034749231826\n",
      "train loss:0.04091401223092398\n",
      "train loss:0.03387170982244015\n",
      "train loss:0.071999896350665\n",
      "train loss:0.17635224695669172\n",
      "train loss:0.1177228860879043\n",
      "train loss:0.04955754302831794\n",
      "train loss:0.09739431414554288\n",
      "train loss:0.01783106124733131\n",
      "train loss:0.05511272204029262\n",
      "train loss:0.07166045995196244\n",
      "train loss:0.1253262067207805\n",
      "train loss:0.04431799000882822\n",
      "train loss:0.04447079648123019\n",
      "train loss:0.07984104629844763\n",
      "train loss:0.12712876666200892\n",
      "train loss:0.040770403455969356\n",
      "train loss:0.06459586356749769\n",
      "train loss:0.06827261433290252\n",
      "train loss:0.07416513915905967\n",
      "train loss:0.05099899790742203\n",
      "train loss:0.14768385741818701\n",
      "train loss:0.02409757097722914\n",
      "train loss:0.10997057212879792\n",
      "train loss:0.06306852213988433\n",
      "train loss:0.07271524228902444\n",
      "train loss:0.05599664316090085\n",
      "train loss:0.1549198678562489\n",
      "train loss:0.0610730550771168\n",
      "train loss:0.09914743932633495\n",
      "train loss:0.055375426878720574\n",
      "train loss:0.04949268507022384\n",
      "train loss:0.05579546942513135\n",
      "train loss:0.06702781445623261\n",
      "train loss:0.10914028260750537\n",
      "train loss:0.08288820416280926\n",
      "train loss:0.06589554712420392\n",
      "train loss:0.0416137258309712\n",
      "train loss:0.043557679745779565\n",
      "train loss:0.05852457062509308\n",
      "train loss:0.06498289851925763\n",
      "train loss:0.07224483757379335\n",
      "train loss:0.11019853509828138\n",
      "train loss:0.05587253545866177\n",
      "=== epoch:14, train acc:0.976, test acc:0.946 ===\n",
      "train loss:0.05895033303121759\n",
      "train loss:0.04867553401421029\n",
      "train loss:0.07965245210554844\n",
      "train loss:0.16824613848058692\n",
      "train loss:0.07079991915714902\n",
      "train loss:0.20711698288388836\n",
      "train loss:0.13456661996098768\n",
      "train loss:0.08117170785477709\n",
      "train loss:0.09211107094949694\n",
      "train loss:0.09763091823889111\n",
      "train loss:0.03793164010733401\n",
      "train loss:0.09126597655393134\n",
      "train loss:0.09436050908363913\n",
      "train loss:0.11559191266789595\n",
      "train loss:0.06433927184155257\n",
      "train loss:0.17223160757759193\n",
      "train loss:0.04277531767948758\n",
      "train loss:0.060756647867484685\n",
      "train loss:0.03147995814847307\n",
      "train loss:0.047846558053071496\n",
      "train loss:0.035992942411274556\n",
      "train loss:0.08654727189644466\n",
      "train loss:0.06890015378138915\n",
      "train loss:0.09736349142236221\n",
      "train loss:0.06211830584095909\n",
      "train loss:0.05160184706631782\n",
      "train loss:0.07036963305061862\n",
      "train loss:0.0770633240548235\n",
      "train loss:0.059421251981727596\n",
      "train loss:0.03419438897034449\n",
      "train loss:0.10207691455705864\n",
      "train loss:0.15304634856008661\n",
      "train loss:0.1042511414300101\n",
      "train loss:0.15599093923143734\n",
      "train loss:0.030280188041525324\n",
      "train loss:0.031556450797416855\n",
      "train loss:0.043498785163131916\n",
      "train loss:0.05719258040261706\n",
      "train loss:0.08605578447424328\n",
      "train loss:0.06337383425466618\n",
      "train loss:0.055124761727340287\n",
      "train loss:0.08666434042446815\n",
      "train loss:0.10096236712673676\n",
      "train loss:0.03491373572255912\n",
      "train loss:0.07867941387702641\n",
      "train loss:0.04662092509879864\n",
      "train loss:0.06865791732609991\n",
      "train loss:0.04142493673241757\n",
      "train loss:0.03990881126981762\n",
      "train loss:0.033570428771217735\n",
      "=== epoch:15, train acc:0.972, test acc:0.943 ===\n",
      "train loss:0.09523245318086185\n",
      "train loss:0.10440037932963944\n",
      "train loss:0.06495836298542824\n",
      "train loss:0.03841758076957447\n",
      "train loss:0.050775829056005826\n",
      "train loss:0.05517065305537807\n",
      "train loss:0.12234274649284849\n",
      "train loss:0.050476003426900465\n",
      "train loss:0.041460796887185423\n",
      "train loss:0.056983849613933844\n",
      "train loss:0.15334616120049083\n",
      "train loss:0.07079776245873984\n",
      "train loss:0.06203984675666134\n",
      "train loss:0.053609172120192304\n",
      "train loss:0.03901351941244089\n",
      "train loss:0.04015866521196088\n",
      "train loss:0.09638123252735312\n",
      "train loss:0.05296819508460634\n",
      "train loss:0.08155669798546794\n",
      "train loss:0.06923696342985051\n",
      "train loss:0.0779404714349039\n",
      "train loss:0.08055224689259838\n",
      "train loss:0.042762807593236314\n",
      "train loss:0.028515101039942498\n",
      "train loss:0.07795088577629333\n",
      "train loss:0.057895996117776266\n",
      "train loss:0.04812022454541039\n",
      "train loss:0.08961334189326572\n",
      "train loss:0.09671415999219608\n",
      "train loss:0.10812109848266768\n",
      "train loss:0.08178322812070413\n",
      "train loss:0.014576380525395206\n",
      "train loss:0.09894775863720202\n",
      "train loss:0.039532861743863404\n",
      "train loss:0.04729503170541956\n",
      "train loss:0.024005836582515906\n",
      "train loss:0.022620918447582416\n",
      "train loss:0.03651426308934287\n",
      "train loss:0.04961345189570768\n",
      "train loss:0.079697455165954\n",
      "train loss:0.053542934075999575\n",
      "train loss:0.10754379580856083\n",
      "train loss:0.03830536456953658\n",
      "train loss:0.04816520491192351\n",
      "train loss:0.03897937386458395\n",
      "train loss:0.03850225228320485\n",
      "train loss:0.028705688635472335\n",
      "train loss:0.011598380466888526\n",
      "train loss:0.04185831339207616\n",
      "train loss:0.056065996916891826\n",
      "=== epoch:16, train acc:0.974, test acc:0.946 ===\n",
      "train loss:0.03039846868001112\n",
      "train loss:0.0757724905996278\n",
      "train loss:0.08593756534344679\n",
      "train loss:0.10832130657819788\n",
      "train loss:0.06326399731115898\n",
      "train loss:0.08512855694402671\n",
      "train loss:0.07658017886613547\n",
      "train loss:0.17030137246629604\n",
      "train loss:0.07813450651574358\n",
      "train loss:0.09954067675334359\n",
      "train loss:0.05940105333123882\n",
      "train loss:0.04555051662235443\n",
      "train loss:0.041810574577379474\n",
      "train loss:0.05847451048517355\n",
      "train loss:0.12265337151834861\n",
      "train loss:0.07851734098341023\n",
      "train loss:0.0797707798644207\n",
      "train loss:0.04900940038342218\n",
      "train loss:0.07495783441605292\n",
      "train loss:0.04514596704381853\n",
      "train loss:0.020996112328054265\n",
      "train loss:0.0797691434959014\n",
      "train loss:0.07971376244064864\n",
      "train loss:0.021360509837679394\n",
      "train loss:0.10217282346802806\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "train loss:0.06655199730948627\n",
      "train loss:0.016422426996247453\n",
      "train loss:0.07842796366137143\n",
      "train loss:0.10832190128905898\n",
      "train loss:0.025993356367930463\n",
      "train loss:0.12290254434842461\n",
      "train loss:0.03146163211997479\n",
      "train loss:0.038798978469390503\n",
      "train loss:0.024573474044901022\n",
      "train loss:0.016839253780020033\n",
      "train loss:0.05233039716066787\n",
      "train loss:0.01374982454522698\n",
      "train loss:0.03352671997861438\n",
      "train loss:0.08812714279944489\n",
      "train loss:0.12860681151682984\n",
      "train loss:0.05049432178646271\n",
      "train loss:0.03808411927617014\n",
      "train loss:0.04472016768053562\n",
      "train loss:0.09153074366722817\n",
      "train loss:0.036394998078866043\n",
      "train loss:0.06267111811186266\n",
      "train loss:0.033884989878451593\n",
      "train loss:0.11113652932740728\n",
      "train loss:0.04427707624955467\n",
      "train loss:0.017354029072004885\n",
      "=== epoch:17, train acc:0.979, test acc:0.951 ===\n",
      "train loss:0.04789742620145846\n",
      "train loss:0.10574011821440125\n",
      "train loss:0.09101259123347459\n",
      "train loss:0.028778440655208647\n",
      "train loss:0.011400664983457202\n",
      "train loss:0.044438645901361776\n",
      "train loss:0.06793910720982405\n",
      "train loss:0.058786021715097315\n",
      "train loss:0.05265679940183964\n",
      "train loss:0.07051258616751917\n",
      "train loss:0.07841192050258541\n",
      "train loss:0.03139544593607028\n",
      "train loss:0.09900400384773965\n",
      "train loss:0.05705626033593369\n",
      "train loss:0.027954492700367366\n",
      "train loss:0.012686304992296768\n",
      "train loss:0.06868431619375129\n",
      "train loss:0.02760037655280644\n",
      "train loss:0.031009031787403784\n",
      "train loss:0.033126427041912355\n",
      "train loss:0.041055533527797665\n",
      "train loss:0.08842037328935871\n",
      "train loss:0.029904388898383347\n",
      "train loss:0.09055074513145805\n",
      "train loss:0.06576991545781029\n",
      "train loss:0.05550324840261328\n",
      "train loss:0.026947111551953448\n",
      "train loss:0.027048184122092565\n",
      "train loss:0.040942951679017694\n",
      "train loss:0.017360113216187432\n",
      "train loss:0.13250839535934605\n",
      "train loss:0.04361959265452345\n",
      "train loss:0.0821219619450072\n",
      "train loss:0.07027343237622877\n",
      "train loss:0.019819293866332594\n",
      "train loss:0.039014380410971675\n",
      "train loss:0.017267520796527137\n",
      "train loss:0.03728273724659227\n",
      "train loss:0.028245779259300087\n",
      "train loss:0.028223023952179956\n",
      "train loss:0.05900569198001244\n",
      "train loss:0.08266735756270573\n",
      "train loss:0.03779758425942148\n",
      "train loss:0.027307042165203604\n",
      "train loss:0.038914166464310174\n",
      "train loss:0.03779964855097887\n",
      "train loss:0.01647000756207392\n",
      "train loss:0.028722991102134724\n",
      "train loss:0.0699192918427378\n",
      "train loss:0.023371997850552703\n",
      "=== epoch:18, train acc:0.985, test acc:0.955 ===\n",
      "train loss:0.06981598815646685\n",
      "train loss:0.030881231570797738\n",
      "train loss:0.029958730092484686\n",
      "train loss:0.03422449259286091\n",
      "train loss:0.03295404618631588\n",
      "train loss:0.023779229021360107\n",
      "train loss:0.011236520339392491\n",
      "train loss:0.03825808293156564\n",
      "train loss:0.0374794946310728\n",
      "train loss:0.0237844253919095\n",
      "train loss:0.05413941261265703\n",
      "train loss:0.0727610925231775\n",
      "train loss:0.028697071140773956\n",
      "train loss:0.037032302753403995\n",
      "train loss:0.025448176560030875\n",
      "train loss:0.05028237245525873\n",
      "train loss:0.027711881006337275\n",
      "train loss:0.022576241564595468\n",
      "train loss:0.020957919000826962\n",
      "train loss:0.016577611311988257\n",
      "train loss:0.010066375506439964\n",
      "train loss:0.03380130875419509\n",
      "train loss:0.019461362321761086\n",
      "train loss:0.0420710463100308\n",
      "train loss:0.01584706159705038\n",
      "train loss:0.025272547078995414\n",
      "train loss:0.030435509070780055\n",
      "train loss:0.03206685250055001\n",
      "train loss:0.04707641223725306\n",
      "train loss:0.050641917741492876\n",
      "train loss:0.03302937115753998\n",
      "train loss:0.02174265660382413\n",
      "train loss:0.022249979615429388\n",
      "train loss:0.013713545086793873\n",
      "train loss:0.03139825475268561\n",
      "train loss:0.019036580281456456\n",
      "train loss:0.030689417364770585\n",
      "train loss:0.038374156250803204\n",
      "train loss:0.025339132029487822\n",
      "train loss:0.0363886549852441\n",
      "train loss:0.07080571788765574\n",
      "train loss:0.013525501634940587\n",
      "train loss:0.04170648010969747\n",
      "train loss:0.018961562624689506\n",
      "train loss:0.016185945082216276\n",
      "train loss:0.06352430264134995\n",
      "train loss:0.02422799138391704\n",
      "train loss:0.06261494671449734\n",
      "train loss:0.019589961018961058\n",
      "train loss:0.02187212606448322\n",
      "=== epoch:19, train acc:0.991, test acc:0.955 ===\n",
      "train loss:0.03734037757399384\n",
      "train loss:0.020778434457897798\n",
      "train loss:0.038692112518032055\n",
      "train loss:0.01676495310676168\n",
      "train loss:0.015565285820182642\n",
      "train loss:0.021744533500130964\n",
      "train loss:0.012009095449588485\n",
      "train loss:0.04284409824765142\n",
      "train loss:0.032409211123299836\n",
      "train loss:0.019054974314953167\n",
      "train loss:0.03653871596294546\n",
      "train loss:0.009876231256260276\n",
      "train loss:0.0265098886689662\n",
      "train loss:0.01634161070914556\n",
      "train loss:0.023756539348351847\n",
      "train loss:0.011891099842305124\n",
      "train loss:0.013794893537889258\n",
      "train loss:0.05772953222959168\n",
      "train loss:0.014612093470821232\n",
      "train loss:0.021263493758442123\n",
      "train loss:0.01896510630294652\n",
      "train loss:0.028262013021029948\n",
      "train loss:0.013585359503519561\n",
      "train loss:0.02370151946899202\n",
      "train loss:0.025840948022241118\n",
      "train loss:0.007402981468744765\n",
      "train loss:0.04820016795548287\n",
      "train loss:0.01985521022093298\n",
      "train loss:0.017442162826263128\n",
      "train loss:0.02741862621650525\n",
      "train loss:0.02924273756338079\n",
      "train loss:0.026942021900584246\n",
      "train loss:0.01100624807595602\n",
      "train loss:0.026418933883823735\n",
      "train loss:0.023155501397475874\n",
      "train loss:0.012663249101038037\n",
      "train loss:0.016865855276942777\n",
      "train loss:0.012095632293643285\n",
      "train loss:0.010381643970975582\n",
      "train loss:0.015720932948658565\n",
      "train loss:0.044019004990489004\n",
      "train loss:0.005799323370274048\n",
      "train loss:0.015137329495560537\n",
      "train loss:0.02871780409114261\n",
      "train loss:0.02656838962647572\n",
      "train loss:0.020673965653458783\n",
      "train loss:0.019769588156851096\n",
      "train loss:0.017458943462165397\n",
      "train loss:0.06033893305330516\n",
      "train loss:0.03366614927915605\n",
      "=== epoch:20, train acc:0.993, test acc:0.957 ===\n",
      "train loss:0.03150669325041356\n",
      "train loss:0.009234494009584615\n",
      "train loss:0.019346037088043087\n",
      "train loss:0.04211040400852745\n",
      "train loss:0.022662442827684398\n",
      "train loss:0.046476552376457744\n",
      "train loss:0.0444066781972484\n",
      "train loss:0.018403402086312724\n",
      "train loss:0.014375951796946274\n",
      "train loss:0.029455431791498534\n",
      "train loss:0.023275745540055535\n",
      "train loss:0.01490659806811512\n",
      "train loss:0.022304569621427333\n",
      "train loss:0.015047707281051998\n",
      "train loss:0.011973416665242466\n",
      "train loss:0.011172282518722387\n",
      "train loss:0.030900123300444355\n",
      "train loss:0.024740439434126946\n",
      "train loss:0.037537309257722654\n",
      "train loss:0.01007363034523539\n",
      "train loss:0.017811614310733623\n",
      "train loss:0.0501891529075222\n",
      "train loss:0.031738592008237455\n",
      "train loss:0.05687418040206445\n",
      "train loss:0.033355637329276176\n",
      "train loss:0.02127110610999277\n",
      "train loss:0.05397120548452609\n",
      "train loss:0.027489856692615295\n",
      "train loss:0.009252006652120781\n",
      "train loss:0.027986177070697157\n",
      "train loss:0.038430615086898184\n",
      "train loss:0.017830606140851406\n",
      "train loss:0.02496418260263789\n",
      "train loss:0.00891436698684426\n",
      "train loss:0.014513449038845208\n",
      "train loss:0.033013082526707704\n",
      "train loss:0.03843176183237187\n",
      "train loss:0.008088798823635755\n",
      "train loss:0.016720808540868447\n",
      "train loss:0.010376456698734453\n",
      "train loss:0.05808196337474324\n",
      "train loss:0.009581622118121308\n",
      "train loss:0.02358223433209592\n",
      "train loss:0.019381932759842447\n",
      "train loss:0.004523171987720411\n",
      "train loss:0.02226738154915409\n",
      "train loss:0.058773445097517474\n",
      "train loss:0.013137493580506103\n",
      "train loss:0.0865457556138701\n",
      "=============== Final Test Accuracy ===============\n",
      "test acc:0.956\n",
      "Saved Network Parameters!\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# coding: utf-8\n",
    "import sys, os\n",
    "sys.path.append(os.pardir)  # 为导入父目录中的文件而进行的设置\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "from dataset.mnist import load_mnist\n",
    "#from simple_convnet import SimpleConvNet\n",
    "from common.trainer import Trainer\n",
    "\n",
    "# 读入数据\n",
    "(x_train, t_train), (x_test, t_test) = load_mnist(flatten=False)\n",
    "\n",
    "# 如果处理耗时，则减少数据\n",
    "x_train, t_train = x_train[:5000], t_train[:5000]\n",
    "x_test, t_test = x_test[:1000], t_test[:1000]\n",
    "\n",
    "max_epochs = 20\n",
    "\n",
    "network = SimpleConvNet(input_dim=(1,28,28), \n",
    "                        conv_param = {'filter_num': 30, 'filter_size': 5, 'pad': 0, 'stride': 1},\n",
    "                        hidden_size=100, output_size=10, weight_init_std=0.01)\n",
    "                        \n",
    "trainer = Trainer(network, x_train, t_train, x_test, t_test,\n",
    "                  epochs=max_epochs, mini_batch_size=100,\n",
    "                  optimizer='Adam', optimizer_param={'lr': 0.001},\n",
    "                  evaluate_sample_num_per_epoch=1000)\n",
    "trainer.train()\n",
    "\n",
    "# 保存参数\n",
    "network.save_params(\"params.pkl\")\n",
    "print(\"Saved Network Parameters!\")\n",
    "\n",
    "# 绘制图形\n",
    "markers = {'train': 'o', 'test': 's'}\n",
    "x = np.arange(max_epochs)\n",
    "plt.plot(x, trainer.train_acc_list, marker='o', label='train', markevery=2)\n",
    "plt.plot(x, trainer.test_acc_list, marker='s', label='test', markevery=2)\n",
    "plt.xlabel(\"epochs\")\n",
    "plt.ylabel(\"accuracy\")\n",
    "plt.ylim(0, 1.0)\n",
    "plt.legend(loc='lower right')\n",
    "plt.show()\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 梯度检查\n",
    "\n",
    "https://github.com/oreilly-japan/deep-learning-from-scratch/blob/master/ch07/gradient_check.py"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "W1 5.935449328823674e-07\n",
      "b1 6.069060608834397e-10\n",
      "W2 5.478828239380243e-11\n",
      "b2 5.345778755017965e-05\n",
      "W3 1.32725632003911e-10\n",
      "b3 1.7990897208247648e-07\n"
     ]
    }
   ],
   "source": [
    "# coding: utf-8\n",
    "import numpy as np\n",
    "# from simple_convnet import SimpleConvNet\n",
    "\n",
    "network = SimpleConvNet(input_dim=(1,10, 10), \n",
    "                        conv_param = {'filter_num':10, 'filter_size':3, 'pad':0, 'stride':1},\n",
    "                        hidden_size=10, output_size=10, weight_init_std=0.01)\n",
    "\n",
    "X = np.random.rand(100).reshape((1, 1, 10, 10))\n",
    "T = np.array([1]).reshape((1,1))\n",
    "\n",
    "grad_num = network.numerical_gradient(X, T)\n",
    "grad = network.gradient(X, T)\n",
    "\n",
    "for key, val in grad_num.items():\n",
    "    print(key, np.abs(grad_num[key] - grad[key]).mean())"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 权重可视化\n",
    "\n",
    "https://github.com/oreilly-japan/deep-learning-from-scratch/blob/master/ch07/visualize_filter.py"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 30 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 30 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# coding: utf-8\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "# from simple_convnet import SimpleConvNet\n",
    "\n",
    "def filter_show(filters, nx=8, margin=3, scale=10):\n",
    "    \"\"\"\n",
    "    c.f. https://gist.github.com/aidiary/07d530d5e08011832b12#file-draw_weight-py\n",
    "    \"\"\"\n",
    "    FN, C, FH, FW = filters.shape\n",
    "    ny = int(np.ceil(FN / nx))\n",
    "\n",
    "    fig = plt.figure()\n",
    "    fig.subplots_adjust(left=0, right=1, bottom=0, top=1, hspace=0.05, wspace=0.05)\n",
    "\n",
    "    for i in range(FN):\n",
    "        ax = fig.add_subplot(ny, nx, i+1, xticks=[], yticks=[])\n",
    "        ax.imshow(filters[i, 0], cmap=plt.cm.gray_r, interpolation='nearest')\n",
    "    plt.show()\n",
    "\n",
    "\n",
    "network = SimpleConvNet()\n",
    "#  随机初始化后的权重\n",
    "filter_show(network.params['W1'])\n",
    "\n",
    "# 学习后的权重\n",
    "network.load_params(\"params.pkl\")\n",
    "filter_show(network.params['W1'])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 应用滤波器\n",
    "\n",
    "https://github.com/oreilly-japan/deep-learning-from-scratch/blob/master/ch07/apply_filter.py"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 16 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAUoAAADuCAYAAABf005JAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADl0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uIDMuMC4zLCBodHRwOi8vbWF0cGxvdGxpYi5vcmcvnQurowAAIABJREFUeJzsvXdw3PWZP/76bO9Nu6td9W5Jlivu2IApNoQanEAIM0nw5UIO7kiBucv9cTOZHFcyV2Yuk3JkMiQEkpBgjhATatxiGxdsLNmyrd5X2t57/f2hex59VuQOK/clSPPbZ8Zj7Wq1u5/n87yf/rweoVQqoUIVqlCFKvQ/k+Tj/gIVqlCFKrTcqaIoK1ShClXoQ6iiKCtUoQpV6EOooigrVKEKVehDqKIoK1ShClXoQ6iiKCtUoQpV6EOooigrVKEKVehDqKIoK1ShClXoQ6iiKCtUoQpV6ENItpQXq1SqklarRalUQrFYhEQigUajgVKpRLFY5Nfl83mUSiX+txQSBKHssUQigUwmQzKZRC6XK3sdvbZUKiEcDkOj0cBisfB3KRaLKJVKUKvVCIfDCAQC5W++jEilUpV0Oh1/Z0EQyngrCAJKpRIKhQK/5v8lb/P5PADwZ4tfGw6HoVary3hLMrASeKtWq0sGgwEAmGdSqRRSqbSMh/8TT+k5ugdE9HgxX+l39LfieyomupdKpbLsvel1qVQKRqMRk5OT/lKpZPujGfARE+kFACyrKpUKGo0GhULh/yS7f0gegXnZlUqlyGQyyGazZbwTvzYSiUAul8Nut6NQKPB7lkolyOVyxGKxa5LdJSlKrVaLPXv2IJlMQq1WQxAEbNy4ER0dHQiFQpDJZCgUCggEAshms3z4xF+QmKZWq5mJhUKBlatcLmdG0EE0mUyYmJjAxMQEstkspFIpZDIZM0SpVGJoaAhnz57Fvn370NDQgEKhgEwmg5aWFuzduxe7d+9eyqX+yYl4m06n+eCIeSuVSlEoFBAMBpHNZtloEG+BeZ4VCoUP8JaEk3gLzAu0RqOhg4iJiQnkcjlWICRwSqUSg4ODeO+99/Dwww+zwGUyGXR0dGD37t245ZZb/uT8WgoZDAY89NBDyGazAACZTAaZTAaHw8GHTBAE5HK5MmMMLBwqQRBQLBYhl8vLnILFyo1IKpVCLpcjnU4jHo/ze0gkEn69IAgIBALI5XJoaGiARCKBXC6HIAi4fPky7rzzTmSzWTz66KOTHyV//q+k1Wqxd+9eJJNJKBQKyGQytLe3Y/PmzQgGg5BIJCiVSggGg0in08xj4i05XXTexU5WLpdDqVSCQqFAsVjkc6BWq2E0GuF2uzE6Oop0Ov0B2ZXL5fB6vTh8+DC2bNmCXbt2IZfLIZvNoqqqCp/61KeuWS8sKfQulUrIZDJQKpV88TKZjP+n5yQSCWt8+kePyYspFAosTCqViv+eiISRGCiTydjDksvl/Fr6rHXr1kGlUuHHP/4xyHvo6OjAJz7xCaRSqaVc5sdGuVyODwqAMj4t5q1MJivjLwkI3Q+5XA6FQgGVSvUBvtKBFR9+rVbLQr74MG/YsAEKhQI/+MEPYDKZUCqV0NXVhd27d7PyWe6Uz+eRz+chl8tRKpWYpwCYD4IgQKFQ8PPE0z/0mIwOGSFgQRaBeQNG94v+joi8q0KhAJPJBIlEglAoxPf+6tWr+MQnPoF0Ol1mCJczpdNplh/SBwDK9IIgCHzmxbwkp0cmk7GxlkgkUKlUZfcJWDBMwDyP5XI51Go1lErlB14rkUjQ1NQEq9WKt956C1qtFjKZDHa7HZ/5zGeWpBeW5FECgEKhgNlsht/vLxMw+uLiUIOeo8Oez+chCAIikQiSySQ8Hg8ikQhkMhlMJhM2bdrEynixgEmlUigUCgBANpst8yaVSiVKpRJuuOEGvP7667h69Sp27tyJm2++Gel0mpXCciZSbgaDAaFQiL1DuvHE02KxyDzP5/NsRMgjTyQSiEajCIVCCAaDKBQKqK2txebNmznEFh9oAOw5AkAmk2GvX8zbG2+8EW+88QZGR0exfft2bNu2jT9zufMWmA9jDQYDe9pAubdIz4l5Sp5PsViEVqtFOp1GJpNBPB7nkLm5uRnJZBKRSIQ/S+x5io06sHBGSJHQ78LhMGw2GyYmJnDHHXcgn88jl8tBo9H8SfjzfyHiYXV1NbxeL4AFR2ex903ySz+LlSN513NzcwiFQlAqlTAYDGhtbYXZbEYmkwGwoHxJ8ZITlU6nASzIs0qlAgBs27YNv/nNb/D222/jrrvuwr333ss65Fpld0keZbFYRHNzM1QqFYfLANh7Ed98uhjKAyWTSWYOPadSqaBUKpFKpTA6Ooof//jH6O3thU6ng1KphEKhYK+IhJoYo1KpoNVqWRAlEglaWlpQXV2Nt99+G7feeisymQy79csdJalUKqGpqQlqtRrFYpGNitgrJIEkCyyVSlEsFhGPx/le0EGXSqWc3xwcHMQPf/hDDA0NMc/IkycFScZMJpNBqVSyh0nGqqWlBRaLBYcOHcKOHTs4pF8JvC0Wi8hmszAYDGUHV+z1AWAeAvPevV6vRzKZRCqVQi6X43wYMC/juVwOFy9exPDwMOrq6tj7o0NM0RDxkAwceY4ymQxyuRy1tbVQKBQ4ePAgdu/ezUpycZi/XKlYLKK2thYmkwm5XI5lkOSLvEy6drE3mUql+BpJXpVKJXQ6HZLJJCYmJnDw4EGcPXsWOp0OCoWiTH5JL5CXT39LPBYEAQ6HA6tWrcIbb7yB+++/n78j6atroSUpSplMhl27dgGYFzIK/xQKRRkjKAdEB1qr1aKmpgYajYYvyGw2w2AwcM6BLH1vby+ee+45GI1GmEwm9iIFQWBl63Q6+TkKY/L5PNLpNO6++24cPnwYqVSqzNtZ7odZKpVi06ZNnFckpaVQKNizJGNEfBcEAQaDAXV1ddDpdJBKpdBoNLBarTAYDKzE6O9Pnz6NH/3oR6itrYVery8L85PJJADA4XDw9wHAvE2lUnjwwQfxox/9iC07Gb7lTqVSCS0tLazIyAhRCkmstIjHgiDA7/dDr9fDYDAgl8ux0ZLL5awM6P3Pnj0LAOjp6WGZJYeB5LBQKJS9vzg/n0gksH//fshkMs5TKxQKhEKhj4dpSyCpVIobb7wRANh4kuySfJHTQ/qBIqaamhrodLqyiMpkMrEzRV59f38/nnvuOaTTadhsNo50pFIp0uk08vk87HY75/fJa83n88hms1i7di0uX77M+X06P9ea2liSolSr1awU/5ASosd0QMmCGo1GdrPJHSZvk8JGlUoFhUKBdDqNmZkZ/NVf/RXkcjn0ej0LXrFYRCaTwbp169DU1MShH31OLpfDQw89xAqULHooFGJFsFxJrVbDYDCUXc8f6hwgAaLrIw+UBJAstbgqSMJTKBQQCoWwf/9+2Gw29hqBhfzzhg0bUFdXx0UOcZi/Z88e2O12vr8U6sfj8Y+LbddMRqPxA3lCMa/ESpR4a7VaORcpNk70cy6XQzKZRFVVFeRyOcbHx/Hyyy8jn8/zfSLvkFIeKpWqzPsnr9Zut8NgMCCdTkOr1aJQKCCbzWLNmjV/emYtkVQqFeeuiUjexAVcAGWhd21tLTtcFDWRsyWXy1mhqdVqZDIZuFwufPvb38bVq1dhNBpZlxCvmpqasHr1auRyuTIlmMvlcN999zF/yXOlwui10JIVJXknJAjkEVIYSEwihmi1WhY+hULBQpNKpdiK6/V6xONxeDwezo3pdDo8/vjjqK6u5jyNuLXjpptu4rCRDvLatWvR1dWFdDoNQRCQyWRQKpXw/vvvl+WQliMplUpOTpMiIt6KQzCxF63RaJjXdD+y2SySySQLm0ajQS6Xg8/nQy6X43zwww8/jJaWlrKwnryBm2++mb0BMkBbtmzBxo0bkclkOOwWBAFnzpxBLBb7mLn3v5NKpeLqKbBQdKFrFofg4lCZqvt0sMh4UNeG0WiEwWCAx+NBMpmERqNBqTTfqkbeKnnm1O5lMpkgk8nYcZDJZBgeHsaGDRv4/ROJBARBgNlsxjvvvPNxsu6aSKvVcjcK6QHi5eJcOPFPo9FwgU2pVCKTySCdTvPZVSgUsNvtyGQymJmZ4ZBco9Hg+9//PusW4i8wr4RvuOEGVtp0Nmw2G2688UbE43HI5XJ2moaGhhAMBq/pGpdc9SYLID6w4oosAA6jF4c38Xgc8XicQxCNRgOVSoXZ2Vnk83no9XrOvVGYs2/fPtTU1JQpkGKxiK6uLvT09JT1bX3uc5/j70fK2+Px4OzZs8s+RKSDKS480WO6JgDsmajV6g+EzrFYjA2NRqOBTCZDKBRCPB6HXq/niiMZpz179qChoaHsfhaLRaxevRqdnZ1smQ0GA2677TYAKKssTk9Po7+/f8VUZsWFMfJq6ECLW6lSqRQbKcpFkmeZzWZRKBTgdDrh8/nYSJCnqFQq4ff7MTk5WcYrknmtVlvWNTI9PY29e/fy55Axs9vtOHXqVFlL13IlsR4Q81isxOi5UqnEXjN55/F4nNNqVMVWq9UYHh5GLpeD0WhkXpLsP/bYY5yLpO8gCAIXLim8z2azeOKJJ/j7UedDMpnE4cOHr1kvLHkyRy6X88GgwyVu3qUwMJVKQa1WM+Oy2SxSqRQkEgm0Wi3UajX6+vowNDQEg8HAHgwJL7UMlUolPP3008xYCmdKpRLWrl2LM2fO4PXXX8cXv/hFZDKZsmRyJpPB888/j0gksuwVJYAyawyAeUEeCPHW5/MhkUiwIGYyGeatTqeDIAgYGRnB+Pg4VCoVe43EN8qzKZVK/NM//RM0Gg3fP8oX9/T0YHx8HG+//TY+85nPQKlUMm9zuRzC4TDeeustBIPBZZ//BVDm2QAL/afiPG6pVEJ9fT0ikQi3CVFqKB6PIxqNshfe39/PB1oqlXIVlf4uGAwiFouVFRspt/ulL30JiUQCcrkcjY2NUKlU7GEqlUrU1NRw8SKRSPzJefXHkDgfSXJMTg+wUCxUKpWcmiBFRh6eQqFAdXU1BgYG0N/fz+G1uJBLsiyVSvH1r38dZrO5zIkoFArYtWsXTp8+jTfeeAO7d+/mvK84Iv3hD3/I3QvXQkv2KOkG0wdQ0lbc9wjMJ1/JQyQXm/qd5HI5zpw5A5lMVmZZ1Go19Ho959QodDx8+HBZddVoNOL3v/89Hn74YYyPj8PtduPdd9/lQhHlLl5++WUkk0kkk8kV0cKSTqfZGwYWeEvXTZ7f9PQ0zGYze5vEK5VKhXQ6jcHBQTZUAPh3RqOR0xVkud9++23o9Xp+H71ej7feegtPPPEE+vv7MT09jffee4+FkxTBsWPHMDMzsyLyk+JcJMmamLf0s0QiQTQa5eo4HTAqQm7ZsgXnz59HIpFguS0Wi8jlcpyWEvdojo6Olim6XC6HL3/5y/jHf/xHBAIBDAwMwO12c4uMTCZDNpuF1+tFsVhEOBxeEXILgA21OI0hniqj6xgfH4fZbGbjlM1modVqoVKpYDabcejQIdYVwPy9U6vV7D3SGVer1RgfH8fQ0BArUp1Oh/Hxcezduxejo6Nwu9145513WHbpHh08eBDRaBSxWOyjaQ8ib45+plAhk8mwMqSLm52d5RwkgLJG6KGhIZhMJm6olUgkMJlMHJIAgF6vh9FohEwmg9VqxT//8z/z+73wwgt46qmnEI1GodfryzwwSrK73W5cvXoV4XB4SW0AHxfRgQPAuUYxb8mjlEgkiMfjLDhkoORyOTKZDNxuNwso8UOtVkOlUrHAUmWRQrxvfOMb7MH/+te/xt/93d/B5/NxqE69leTZjo6Ooq+vD8lk8gPh1XIkyi8C5ZV68vDIE8nn8wgEAtDr9QDmvSCtVotgMIjGxkacOnWKi2fxeBy5XA4Gg4FzbCqVCvF4nKfWpFIppqeneQrKYrHgK1/5CrxeL3v6gUAA8XicleSaNWtw+fJlxOPxMo9sORPJLhl0ijzofItTR6Ojo9Dr9fyYPEytVouLFy/CbreX9bJardayxnOtVguLxQKpVAqLxYIf//jHbKhef/11PPTQQ5idnWVDRjl2KmwCwMmTJxEMBpekF5akQUjIcrkcC5/YitJrNBoNMpkMwuEw5xfI+rpcLpjNZvh8PvYwxYeflCbNdpvNZgDzbvnQ0BB8Ph/S6TT3aREj1q1bxw264XAYv/jFL8oU90qwzCRUVGAQW2cSQo1Gg1QqxUaCbnY6nUYoFIJGo0E2m4VMJuMGayqcSaVSGI1G7geke2OxWHDx4kVkMhlEo1H2bqgNpqmpCalUCjqdDoFAAEePHuVDIB4lXe4kzjmKeyaBBSNLio68zkKhgLq6OgwPD8NisSCZTKJQKJT1+tEBlkqlsFqtXJQhT1Gj0WBubo4Vsk6n40EIMvSUNjl+/DgUCgUSicSK6FElot7fYrHI7TrAQk8pMC+7MzMzCIfDqK6u5utTKBQYHR1FVVUV3G43jzHKZDKkUikuWhoMBmQyGWQyGZ5oUigUGB8fRyQSQSwWQ6FQ4IiyVCph586d/FwikcCPfvQj/t1S9MKSFOXidorFOUO6cPIex8bG0NXVxTmLYDAIq9UKr9fL7RoUmlDfmFarRSKR4JYD+ty5uTnONVI+TaFQoKenB1NTU2hpaWFFfuTIkbKWoMUtNsuVxK0/4kkRAOxdSCTz427Dw8Po6enhYkQkEoFWq+XQUSKRcPVUPGFD/Ka+y2w2y7wSe44qlQodHR2oqanBxo0bOUw9d+4cxsfHeQoCWP49qkTUcrJYHojXMpmMPUqFQoFsNguTyYSZmRlYrVZ4PB6o1WrOD9N7Uf6WHAdSsnV1dQiFQpiZmUEmk+F8ZiAQgNVqxejoKD73uc9xxbhQKCASiTBvV4rcAgsgI2SMyNiLizqCICAajWJycpKLuRKJBNPT06iursbc3ByMRmNZOyC9xmAwIBqNspdP0anP5+PKNTkNhUIBGzZswMDAAO677z4+Q6dPn8bc3BwX4MQThB9GS45JyXIQicfsiMh7/G9UGahUKgSDQVgsFni9XlaG5OEQAo1EIkE6nS5rlo5EIlwwIAudy+WQTqexZs0a1NTUYPXq1QDmlUomk8G5c+c4ZyK2aMudxEaHcrviiRz6fSaTQSKRQCAQgFKpRCAQYEGiMJDyv0qlEsFgEEqlEul0Gjqdjj0lAmwQFzaoINbc3Izq6mrs2LGD/zadTpd1EKyElAawYGQoVSDuyiDFls/noVAoYDKZuNhgs9kwPT0Nu90Oj8cDg8HAnl6xWEQymeTqqri7Q6VSIZvNclhNaQ1qdNfr9TymSqkRiUSCvr6+spSJ2GCuBBL3JC5WkOTVURO93+9HdXU13G43GyGTyYRoNIpiscgKs1gsIpFIIJPJcAQkk8kQDAbh9Xr5vlGon8/n0dnZCYfDUdbnnU6n8bvf/Q6pVIqdgaXQkiVdDKFGwibutCdBCQQCSKfT8Hq9cDgcMBgM8Pv9qKqqYo+xra0NTqcTQ0NDaGpq4s799vZ2HD16FGfOnIHH4+ELpnydIAjo7u7G/fffj9OnT+PP//zPWSmeO3eOlTmFm3SzljuJ878APnBDKbFNI4tut5v7TCk0JhSV1tZW2O12XL58Ga2trRwiO51OnDhxAgMDAwiFQqwkKRwXBAGrV6/GXXfdhStXrmDz5s0c4pw6dYqFdjF4xnKnPzR7TAqUnkulUojFYpwzo6Zzr9cLnU6HbDaLeDzOh9jpdGJiYoKHJBwOB/bu3Yv+/n4uxNC4ryAIyGazOHbsGP7sz/4MnZ2djIZVKpVQVVXFITcpycUOyHImSsctHo6gbgvidSgUYiwCnU4Hm80Gt9vNs9yEPFRfX49oNAq3241169ahUCigvb0d09PTePXVV+FyuVgZk4IsFAqor6/HQw89hImJCTz++OOso8bGxgCA01L0/EcSehMTxP+LmSNu76H8GrnYMzMzqKqqYmZYLBbYbDZUV1fDarWyAm5ra8Pk5CSGhobg9/vZqtBnOJ1OfP3rX0csFsORI0ewefNmzmnk83mcPn0akUikbNpiJSTExS0q9JiuWzzZIPbuCE0pHo/zNAf1T5pMJjgcDpjNZuTzeRiNRtTW1mJychIzMzNcrBG/d319PR555BEUCgWcO3cO7e3t6O7u5urk1atXeTR0sSFazkTfl/hLvF58UFQqFXt3ZrMZGo2GD3QqlUI6nUZtbS1GRkZQVVWFxsZGRCIR2Gw2rF+/Hhs3bsSJEycwNzfH3iXx98SJE3jkkUfw8MMPY2JiAt/73vfwyCOPoFQqwWAw4OrVq9zCIlaSKyH0XiyzQLlBEvc/m0wmZLNZ6HQ6GI1GTE1Noaqqime+a2pqYLFYYLVa0dbWBr/fD4lEgo6ODrjdbhw9ehSzs7NliFqCIMBkMuHJJ5+EVCrF8ePH4XQ6eRy3UCjgrbfeQjgcBgCOGD7SYg7lKUnAqBIr9i6i0Sh6enowMTGBxsZGBAIB2Gw2zuk0NjZi3bp10Gg0cDqdmJ6ehsvlQmdnJ9544w28/PLLKBQKMJvN/H4dHR343Oc+h6mpKbz44otIJBL46le/ii1btnASua+vj4sR1AO3UiqHFG6JD5eY38TnaDSK9evXY2ZmBm1tbVwEo9dbrVZ0dXVBp9OhtrYWExMTiMVicDqdOHXqFA4cOIBwOIyqqiqUSiXEYjGsX78eN998M5LJJA4ePIhsNot9+/bhnnvu4abzCxcuYGJigpGdKA+9Eg4ysDAyS/wlT1LcTSAIAqqqqjA7O4uZmRmYTCaYzWZuB/J4POjq6kJTUxNKpRKmpqag0+lQV1eHubk5vPLKK4zJSCH+0aNH8cQTT6CxsRF+vx8vvPACNBoNvvSlLyGTyfCBpyIR0WIDuZxJ3L5G100kPnvJZBK33norDh8+jPr6eqhUKnaegHlYxM7OTpjNZtjtdgwPD0OpVKKxsRHHjx/Hiy++iGQyycMT4XAYZrMZTz75JMbHx/Haa69hdnYWjz/+OPbu3cswanNzcwgGg8xvceH5WmnJvj0lqskqUxmfiAb8H3nkEfj9fhSLReh0OlZera2t6OzshMVi4RYYs9kMp9OJQ4cO4ezZs0ilUgxQQAeVniOr9N3vfhdbtmxBOp2GXC5HNptFf38/5yDoEFMb00ogcaVVzFs63MTbL37xi3C5XGUNzVKpFC0tLWhtbUV1dTXDTFVXV6O6uhrHjx/HqVOn2EunYoQgzIPHWiwWJBIJOBwOPPXUU7jrrrs4z5vNZjE1NVUWOVCyfiXgUZI3I/bSFj8HLOTbCWGJpmwAYHZ2FmvXrkUsFkMymUQikYBer4fJZMLIyAhisRh8Ph+PnFLXwGc/+1lMT09DIpHg8uXLWL9+PQPI0r0dGRn5g2E29SCuBKL0DACWDXFHBHVfbN26lfPger2eC45WqxUdHR3cCpROp1FTU4Ompib09vbi4sWLiEajPJEGlI9E0vPf/e53GVycpqVOnjzJyps83UKhsCS9sGRFSc3NYgYsDm0ILCAajcLpdHILj9VqxXXXXYfVq1cjFArBarVibGwMVVVVGBgYwPDwMEZHR+FwOFjYQqEQIpEIRkZGoFKpsHXrVnR1daG6uprxAakqHgqFmBHiCvJKIZrsIM9HjBAv5q044U19aA6HA52dndi4cSMikQjMZjP6+/thsVgwOzuLqakpXL58GTabjYsLsVgMsVgMExMT0Gq12LFjB3p6erB+/XoWZgAcri+eyFopJJ4YIe8cQFnEQQUHo9HIxocGH7LZLNra2vjeAODCj81mQ7FYhN/v57YUwq7M5XLw+/0wGAxQqVRYtWoVp5nIwFgsFhQKBYTD4TIvUgyMshJIrBfo+5PcUlqNCi4qlQrt7e1sHDQaDW666SYem5XL5RgcHITVasXVq1dx8eJFXLx4ETabjYuUgUAAiUQCs7OzSCQS2LZtGytWmlQrlUqMe0sKmSKJxW13H0ZLUpSLb5w4ryZuOo3H44jFYujq6oJSqUQymURPTw9uv/12APOJ35qaGkxNTSGfz+P111/HL3/5SwQCAfaiTCYTgsEgV7hpkqGjo4MPOOXJstkszp07B4/HA5lMhqqqKuh0Ou6zXCnVWXGuR5yjJAVZLBYRi8UQDofR3NwMh8OBTCaDzs5O7N69m8EJ6urqMDY2BplMhjNnzuD555/HzMwMADCOZzQaRTab5YLb0NAQOjo6uFJLEUCpVMLIyAj8fj/3CVI1HcCKmUWmgwosGFExxB8A7sNTKBSwWq38fG1tLaxWK+MPENqVTCbD5cuX2esslUowm82ca8zlckilUkgmk9BqtbxDRqxESMapd5M8HXrdSlGUiyv09N1p5LVUmgcLSSQSWL9+PfdKNzQ04J577uFz7HQ64fF4oFAocPToUbz44ouYmJjg3lWLxcKrZlKpFCKRCPr7+9HQ0MAOGhn4XC6HCxcucG7SYrFAp9NxHpoM5LXQkj1KcfhC85uUGAfA5fpUKoV9+/YhGAyirq4OGzdu5DCZRsMcDgfUajVeeuklJBIJuFwudHR0AAD8fj+H39TD19vbi1JpfmqFDrlKpWKvKB6PI5VKIRAIIJPJQKPRwGazca5zORPlIoEF0AbiLR0WEpZsNotPf/rTCIfDqK2txfbt28v6WqnoIJPJ8POf/xxerxfT09NwOp3cq0f5JDIifX19LNCE5KJUKuHz+TA0NMS89fv9DGBis9lgsVhWBG/FyDbiSr+47YqKY7FYDJlMBl6vF01NTewd0j2iiZH+/n6eoHG73TAajVz0ofNB8GASiQR2u73MWzWbzRgdHUUkEuGODplMBr1ez2sLVgLgiLgOIM75Lu4HVqvVyOfzaG9v52WAmzdv5rY2QtCqrq6Gw+HA888/j3Q6jbm5OaxatYpb3Uh2yUj39vYyTiXJLuXtz58/D7/fzwMwpDOqqqq4af1aaMmKcnGuhyyfGJ5KEAS4XC7s3bsXANDV1cWhhkajgUKhYC3/i1/8gi3I+vXrUVtbi89//vNYt24dg2XY7Xbs27ePkbUpfKJDPjAwwB4QNa0D89Mq0WiUCxDLncSFMXHDrjhXSY37u3fvhtFoRF1dHWZnZ9lroTasSCQf5yeZAAAgAElEQVTCvCU4/ebmZvzFX/wFGhsbWSlUV1fjjjvuwJo1axjwhA4yAIyNjZUhu2i1WkbYIYWyEopl4nwffV+SH3HudW5uDnfccQei0Sg2b94MnU7H1W9SWm1tbRgeHsbc3BxmZ2cBzBciGhsbWaHKZDLuI6adQzKZrEweqVVF3GoHgKHs5HI5dDrdn5RPfyyR0wTgA7JLRkomk2FychK7du2CVCrF7t27y4wsOUFqtRrPPvssJiYmEI1GsXHjRjQ0NODBBx/Ejh07OJVhMBhw77334v7770cmk4HVauXCkEQiwdTUFL+3QqHg+fF0Os1tbh/ZZM5iaHdCtKGDrFAocPjwYahUKpw/fx6f+cxnMDQ0hEgkgvr6ehiNRgwODmLdunX4zne+g76+Pmi1Wjz99NNwOBwYHBxEQ0MDLl26BKfTidHRUXzta19Da2srWltbubSfyWTYU+zv7+flTOIeNKpkrgQlKc5PivNT4uKZUqnExMQE8vk8zp49ix07dmBubo7BKpRKJS5cuICtW7fiJz/5Cfr6+qDRaPDoo4+is7MTV65cQWtrKxobG5HP5zE1NYX9+/eju7sbnZ2dPI4XjUZRVVXFwpZMJsugwcStL8DKaWEhmRAn8YnHUqkUg4ODEAQBP/nJT7B//36cP3+eZ+qNRiN7I8FgEOPj4xgbG8PWrVvR2NiI2267DS+//DJsNhtXsNPpNL71rW/hwoULyGaz3NoFgOfKxUac7rlUKuWQdaXILnmEYqKCFumIgwcPQqfTobe3Fw8++CAGBweRzWbR0dEBiUQCl8uFpqYmvPrqqzh//jzS6TT+4R/+AS0tLRxFDgwMoLa2FjMzM9i/fz927tyJ6upq/rx4PA6bzYZSqYS+vj4EAgHWT8VisWx53FJ4u+S7QGNY4twOFVVommF2dha33XYbNm7cyFXqQqEAh8OBYDCIzs5ODA0Ncf7glltuQSgUwtGjR/Hyyy/jZz/7GfR6PbvqWq0Wb7zxBjKZDBQKBXtN5BFQh764uZwgnahfayUQwZ1RAhyYb5DNZDKQy+UwmUwYGBjAzp07sXr1au7By+VyqKurY++lv78f58+fRyqVwpo1a2AwGHD69GkcO3YMzzzzDEwmEwNl1NfXMyIOTU6IgURo/E58mFcabxcbIbFil8lkmJmZQWdnJ1paWhCLxfDUU09hcHAQNTU1MBqN0Gq1XKVtbW3FxMQE/H4/9u7diy1btsBqteLQoUMIBALMV7fbjf379+P06dPI5XLciyn2OAnwFwCH5GIPR6fToa2t7U/Orz+GaAuiOLVBskupoHg8ju3bt+Oee+5BOp1GVVUVpFIpqqurUSwW0djYiFgshuPHj2Nubg6f/vSnEYlEcOHCBbzwwgv4z//8z7Koyel04rXXXkMoFGLciGAwyJBvo6OjZcAtZCgVCgV0Oh1MJtNHk6MkgRODmIoRy0dGRtDb24t/+Zd/QXd3N86cOcOAAFNTU5BIJLBYLMjn83j++ecxPT2NL3/5y5iamsLPf/5znD17FqVSCaOjowgGg1x0OHLkCOrr63H58mUUCgVYLBb2EFwuF7daUP6IYMUoTDxy5MiyR+Em3lKLBVVGqZjV39+PkZERPP3009i0aRMGBwdhMpng8XgwPDyMYrEIq9WKRCKBX/7ylxgbG8Mdd9zBQACnT59GIpGA1+tl3no8Hhw6dAg6nQ6Tk5MoFArQ6/U8STEyMsIje2LQX1KSuVwO77777rKHWhPnzYCFLgIaB7Xb7XjuueeQyWRw991348CBA9x8Pjo6CkEQuC/10qVL6O3tRXd3N1KpFPr7+zE5OckKj7xVvV6PgYEB9sLJkBNSEA1JiCvx4rG/trY2xGIxvP/++x8P05ZA5FHSJgLia6FQQDKZRCAQwDvvvINvfOMb2L17N44dOwaTyYRIJMKFMIvFArVajZ/+9Kfo7e3FY489Br/fj9deew2HDx+G3+9HIBDgsV0AePPNNzlCpZ05ZOTJKIn1AmEYaDQaqNVqHDlyBNFo9JqucUmht1Qq/cDmQwDcMJvJZNDS0gJgfgC9ubkZOp2OvcPR0VE0Nzfj2LFjOHPmDDNDEASEw2FWplarFbFYjOGu/H4/pqenoVAoGDmErMbVq1cRjUbLLAMpnFwuh5MnT66IPkqJRAKNRlO2pY6eT6VSnHuVSCQ4ceIEmpqaGG6KlJ7NZkNvby/vhdZoNPD7/QiHw5iZmeGG6lgsxsACHo+HvaZVq1bBZDLx/Z2ZmSnbNUThPyGxXLhwAfF4fNmH3mSEiMS9qblcDlNTU2hoaEBLSwvC4TB7hYODg9BoNAgGg7yimQ6qGNCXujUkEkkZZmIwGGSjQwg4YsALMeAJsBAKOhwO9Pf3I5vNroiuAolEwlV7ekz8TaVSCIVC6O7uhsViwZkzZ+BwOJgfjY2NOH36NLZu3Yr33nsP7777Lvbv388GIxaLYXp6GsC88aHWN5oZn5ub4/toMpngdrsBgPOb4lQRFYflcjkvILzma1wqQ8S4hjReSPPFq1atgiAIjPLT0tKCYrGIjo4O1NbW4ne/+x10Oh2OHTuGNWvWYPPmzbzdjuCmCCSAFCGN5FGPG1UPp6amkM1mMTw8XDYtQopGpVJxJTebzX4AVmu5EV03HRrKqdAOEVJiVFRoampCLpfD+vXrsWrVKvzqV7+C2WzGb3/7W2zevBn33nsvUqkUlEolIzVRMzV5OdTiQmGlRCJBdXU1gsEgisUiZmdnuYGfkvHkWU5MTCCRSHAry3ImcW6VHgPzqY7a2lpks1m0tLQwpmRLSwtSqRS6u7vR1dXFqNqRSAQHDhzA9ddfz3PxlPMUt8OQpwiAc2tikGAAcLvd3NBP95wOcygU4nTLUg7zx0ViYA8xBi1FKA6Hg4tbANDa2sr8bWxsxPDwMORyOY4cOYJisYi77roLuVyOIxcybFSAlEgk3O9KrVW02dXj8SCfz6O3txfRaJQdOvLclUolLl26xDCB16oXlpyjpDUCJGyklBKJBOrr66HRaLiPkvZTjIyMYGBgALfddhumpqYwOTmJf/u3f0M0GmUtTxVYjUbDoANKpZI33EkkEjQ1NSGZTHIyl1YSkCsubiT2+/2cu6RdMsudCGpOXCUkfhKiEgBu1ZFKpYzFd/3113NqYv/+/UgkEmXrQW02G3v3SqUSarWawZKlUilqa2s5lzQ5OYloNAqXy8UGi3irVCrhdrvhcrmQz+e5Ir6cSTx2K+7MSCQSGB8fZ8Xo8/kQjUYRjUZhMpkQDofR29uL3t5eNDc347XXXsNXv/rVsuulsJMUo8FgYGNNBkQul7MSValUsFgsjB60OGdKh51m+MUz08uZ6LyKea1Wq+HxeNDR0QG5XM5TTfF4HEqlEmNjYzhy5Ai6u7sRj8dx4cIF/OIXv0AoFOJzQD2tBoMBer2eFafNZuP7WVdXx3lg6qCZnJxkp454TL2aExMTKJVKS1o4uORZbyqti3MvxWKRc4+09c9kMuHSpUuw2Wzo7u7G6OgonE4nHnnkETz99NPIZrNwu92c2KbWFMrVaDQaVFVVobOzE1qtFna7He3t7XC5XKipqcH09DSuXr2KWCxW5u3Qe126dAmlUond7+VOxFsCdCXE+EwmUzaRkEgkYDQaMTc3B7/fj/r6eni9XjQ0NOCLX/wiHnvsMZjNZs6PUV6GDlypVOIeyK6uLmi1WjidTrS1tcHj8aCmpgYTExMYHh5GJBJhD1IcVhLqTSqVYg9qOVOpVGIoLrGMkVet1WrZ+/b5fAzUUizOb/57/PHH8d3vfhdbt25lPEqxl0L5L41Gw2s1APC6ga1bt3KLF+XKxR6pOIogtCwCW17ukRCwkEagHeRU0EmlUoxQTtek0Whw6dIlmEwmdHZ2Ip1Oo6enB1/5ylfwla98BYIg8OizTCYrWydBHqHBYEDTf8/ba7VabN++HT6fD1arlQ08FXHoHzkMfX19rCSXMrm35GIOAAQCAf6ZhCmVSvFkCIXIpKxyuRyOHTuG733ve9i+fTvq6uq4GkZJboLGp4bfqakpvPTSS7hy5QpkMhnWr1/Pi4iozH/27NmyRUXiyaFEIsFjTCuBiJ8+n489HkIDIt5SQUEQBJw4cYLTEu+++y6ee+45PPDAA9i0aRPS6TRSqRSy2SxX+chbjUajGB8fxyuvvIKrV69CIpFg1apVXJCjudmBgYGyTgJgoW+WGneXe8gtJuqpXSwnMpkMOp2Oc1tyuRzpdBrj4+PcHP3v//7vGB4ehtVqZQVGhkdcIIrFYggEAgiFQoxwQ+kUar2iwpm4d5K+C+X6ic/iQY7lTPQdyfEhQ087nAwGA1KpFGpqaiAIQhmW5KVLl/DNb34TDQ0N6Orq4qkwkl3qsCA80enpafz617/Gu+++C6VSia6uLhQKBcTjcR5iOXr0aBkAhngSy+/3/1Gyu+RijljQaMyOhCcSiaCtrQ0zMzNobW1FVVUVJicn8etf/xoulwu33347br31Vh5zFA/Sk6tOLUGUV2xoaIDFYsHatWs5EV4oFKBWqz+wAVBcYBLvCF4Jo2Biz0LMW8KeDAaDqK+vh8fjwerVq1FdXY3JyUm88sorPBa2ceNGlErziEBi3lJuksbvyPpbrVb2LGkqRxzCEFGIQ1FEKpXixt7lzleg/PuLZ5Ep+U/jtqTUYrEYFAoFGhsb8cMf/hDd3d245ZZbuDeSvEFxcz5NO01PT8Pn8yEUCmHDhg3YsGEDK1VaExwMBlkJiivdxF9x0/RKCL3FqSLxWDPNbedyOaxatYpXQOh0Oly+fBknT57EkSNHcPfdd+O+++5DqbSwvJDWjtD8OHn91Fva1NQEvV6P3bt3l/HSarVienr6A3JJnj/NgYt12LXQkhvOBUFgRUYfQnkdq9WKZDLJIAuUZB0YGMCOHTvQ1NQEjUbDI2EOh4MtfTabRTgcRkdHBwqFAnw+HxwOB1paWrB9+3YACy6+uAWAhF7cqJ1OpxGLxfjxSghfxKkM4i0JDuUQM5kMJiYmoFQq4XQ6EYvF0NfXh7a2NlitVn5NoVCA3W5HPB5HOBxGNptlqLpisQiv14v29nbU1dVhw4YNvPtFjNqdSCT4O5BSIN4mEgnm7UrxKmn/D4XeWq0WY2NjiEQi2LFjB+LxOBvnWCyGnp4e/Nd//Rc+9alPoa+vj8dv6R9Vs0mRUSjY3d0Nn88HAOw50nZNqg7TQSXvS+xRLkYJXymyK94+QIUzr9fL+AsSiYRbgQgg59ChQ9i1axccDgfndIvFIjfte71eDuHpNdFolOsh1KhOBbNUKgWTyYQrV65wegVYmHgrFAoMKLNUvbAkRSmRzK8KMBgMLCT0gRs2bGC4LloUZLFYEIlEUCgU0NXVhdbWVi7yAPNrZ1tbW9lrokorMdbpdEKpVGLr1q2M/kEejSAIbHkJkINW4ZLHSpaaciTLmagNiBq4KVQolUro7OzkAoPX62UD5Ha7IZfLsXHjRqxbt46NAy2wamlpKetxnJmZ4QIPgVts27aNlWI2m2VgCGoLIkVILTH0GjrglMBf7mQymfigEvBETU0NNBoNhoeH4XQ6ea+NTqfDyZMn0dPTg2PHjmHz5s1sJCispHQRGbZwOAydTse/u3z5Mm666Sb2sijqojNEz9PrydCTkaTZ6JUwHkp1BppPpwjOZDJh7dq1cLvdaGlpYXxJo9GIYDAIQRBgt9uxefNmNtTE19raWgQCAfZMfT4f732qqqpCMpnE3XffzXP1lJajTgHST9TJQIUhaiWkFNNHNsJIfXPEHACorq6GzWbjyRKn08mrO6PRKG6++WZs2rSJQyCqClJ+c8OGDXjvvfd4CyAVCzQaDW6//XZe26pSqTjpTt+H/ierplAoOCQXh7LL/TALgsDLqcRhrs1mg16v57n32tpaRKNRBAIBxGIx3HLLLbjpppsAgA8ipR/kcjl6enp4hxDleDKZDHQ6HXbt2lUGbkKWeXFVVzyNQwl7YGFmeiXwlpCyqdWM9hGlUilUVVVx8cDlcqG2thZarRZXrlzB9u3b2SOh6RrqGaTCDHkrgUAAbrcbuVyOATBIuSYSCS760Oup7YWiIiqEUmFSDHW3nInSFRR1iDsAqIVQEATYbDao1WrMzs4iFothw4YN2LNnDwCUFVyINxs3bsSFCxcYJYj6euVyOe68806GbaO6BrVxUcsW6RlSkiTftHJ4KXphycWc6upqVkyUB3O73TwbTDtCCG+OWgKohYgOJr2OhLeurg7V1dUwmUzceE1WilphCI2FZmmbm5vR2NhYhiGoUCgwMTHBj1UqFW6//XZuhF+uRM3gYp7QXKrD4WBrm0gksH37dsjlclaYWq2WQ0fK39KBIxzEpqYmmM1mPuSCIPD6VQAsdOSlNzU1oba2tqyPTaFQcFWWKpKf+MQn0NDQ8HGy7pqIjLoYuHV6epp7dslz1ul0mJiYQDgcxsaNG3lslnKT4jnsnp4ebN26FR0dHdDr9awkmpubcdNNNzFvSQnQZ1AKgLwo8WEVAyIfPXoUzc3Nf0o2/VFEskutZGQQzp8/D6lUCp1OxzUIpVIJrVaLfD4PnU7HxoC6J+hekLF2Op1wOp0wm81c8CUIRfECQZLfVCqFuro6tLW1wWKxsDdJERidKaVSifXr118zf5ekKEkbNzQ0cPI2EAigubmZIY5cLhe2bNnCnfFr1qyB0+nEgQMHuJJH3iQxdGJiAhcvXoTf70cqleL2lcnJSXzhC1/A1772NYagooZTwp3TaDTo7u7Gpk2bsGrVKtjtdt4ZrlAocN111+Hv//7vlz24QC6Xg9FoRH19PVf7qP/TbDYjl8thenoaW7ZswaVLlwAA1113HcxmM5555hk+YHQg6f6MjY3hvffew/T0NJLJJHQ6HSe89+/fj6eeeooFmYBnM5kMh040GNDW1oaqqireHS6VSrFlyxY88cQTZVMvy5HELU6k8H0+Hzo7O2G32yGTyWA2mzE+Ps6pBbvdjlKpxBNnYsAHMuButxunTp3CyMgI9zyqVCq4XC688sorMBgMHNJTq1cqlUJ1dTVqampQV1cHlUrFeT2aWZZIJBgZGcG3v/1tdHZ2foycuzaiImtrayvLXSgUwqpVq7Bq1SpW/ARyE4lEuHj2zjvvMDIVKTxKD/n9frz//vvwer28PM9ms8HlcuGb3/wmvvrVr2JwcLBs9DcYDHIrW0tLC9avX89bGSk9IpPJUFtbix/84AfXjFW7JO3h9/tx4sQJ2O12fo667ykfQPmrmpoa7N69GwAwNDQEjUYDl8vFISZZ24mJCT6k5G3K5XKEw2G2Rr29vbj++uvx93//9xziU1HC7XZjdHQUbrcbEokE8Xicl6R3d3fjX//1XxmlZDmTz+fj8S6iYDDIEzPEm3Q6DZPJhPb2dhSLRUxNTcFut8PtdnNVj6r9k5OT3KICgFFcKH+r1+vR19eHNWvW4Dvf+Q43pQOAx+OBy+Vi3tLuavKorrvuOvzN3/wN3G73suctreW1Wq0c2q1btw5zc3PQarWYmprCe++9B7PZjLq6OvT09DCAcXt7O2KxGE/NkNKlfdMAeOqGQmdqBfrd736Hl156CQ6Hg1vfCA80EolwXyGhatF3c7lc+Ou//mvE43HOqy5nCgaDeOONN1h2i8Ui52uBeSdgcnISgiBg27ZtuOGGG5DL5XDlyhXU1dWht7eXxxApovF4PFw4AxawWCnHaDQaceXKFXzhC1/AX/7lX6K2thZqtRqZTAYejwcejweTk5OYnZ3lmgVNPNntdsZDEO+n/99oyaH30aNHASyMZgmCgJqaGiQSCdjtdtjtdjQ2NsLpdHIjuslkglQqxZkzZ3iyh3II4+PjMBgMbLkpkS2RzC89p73fcrkcP//5z9HX1wer1cq5G0rk0sxtIBCAx+OBVCrFvn37MDQ0hLm5uRXRxnLo0CEAC7yVy+W8ttNoNEKj0aC1tRXt7e1Qq9WwWCzQaDSw2+04e/Yse9E0ETU1NcU9bNlstqyKKp7Bl8vlePbZZzE6Ogqj0VjGW9ohTpNOY2NjEAQBu3fvhs/n4znn5U6UMqDK6uTkJMxmM0qlEpxOJzZt2gSdTsetQfTzq6++ynlFOlTJZJLbgQjBn/JidKBpskQikeDAgQNsTMQFNzLoNInidrs5beXxeOD1elcEOr9CocC7774LqVTKPZDpdBptbW3I5XKw2Wzo7OxEbW0t5ywNBgOqqqrQ19fHa17E7Toul4uHWFKpFBfgqFWIwm+FQoE333wTzz77LGw2G9dQqDBGoBzRaBQDAwMAgHvvvRdDQ0OMf3AttCRFqVKpEAqFGKxCjGRMlVaXy4V4PF62nIk66CORCDeZk4dCc7LUbkHTCOROiydCJBIJHnvsMdjtds73kNsuk8mQTCbxzjvvIJ/Po66uDnV1dYhGowwAsZyJ+kIJdJi8P0pRGI1GBAIBJJNJDs9CoRC/XiqVIpVKcbGLPEsx+II45UFAsqQE5HI5Pv/5z7OwEdE9iEQiOH78ONLpNJqamtDQ0IBoNLoiDjK1lYjTL4S0RCOwJL/k8ZHnSB0HlFMvlUpl1W0y6tS1QTIsbv+hYpLJZOKpHgBc0bXb7fB4PMhms/D5fGhpaWG5Jg9rOZNGo0E4HGa9QEUSUnIGgwEej4cLVlQU1Gq1XF8go0xnPRaLMXIWsCC7hAxG3ibl85999lk4nU5EIhEukJF8S6VSvPXWW4x8vnHjRsRiMS7qXAstSVGKh9CBcpQQStCSN2K1WmE0Gnl50pUrV+BwOLjgoFQqMT4+Dp/Px1aWPFQCASU4KjFUknj8iJhAeaexsTFMT09DEATs3LmTZ5GXe1UWAOcWKU9IB44S4OR9lEol5iuF5RQ20v0g0ApawESGiBQiJbjpfzEfibfimWianR0ZGYFUKsX69eu5YizuA1yuRJ4dsAABRghKdFi9Xi+kUilMJhO3tQFgr4RWF5C3EggE2MtZnP8m/pEREQQBMzMzDC5N34kOciKR4G6E+vp6yOVyBkteCVVvagmidjY6b2azmY0KwcuRXqitrUUkEuFhCq/Xy4pvdnYWc3Nz3PModqIo/UF6gWS3VCphcHCwDLiFzoLX68X4+DiKxSKuv/56/j6LgYb/N1qyohQfOvqZXO50Oo3u7m6erqFwnPb0PvjggygWi9xbOTo6CqlUirm5OVYCxHjyegjCnZBuZDIZXnzxxbI2GPoeMpmM0Uo6OjpQKpVQU1PD41DLmahhl3hLvXdkaaVSKVavXs29e+l0mkFEqqursXfvXhSLRfh8PqRSKe6ZpBwieeDEQzJGFHqTEP7sZz8rM4AUJlKY43A40NXVhXw+D6vVCmBpSNEfB1EDNPUvipGpJBIJPB4PH2i5XI5EIoG6ujpIpVLU1NSgtbUV6XQaPp+P10LEYjHujaQ0EnVdkIKkzwXmFWMwGOQODjJEcrmc0XPy+TycTicrx0QiwRiPy5noekgZAgurN+i8U3U5mUwimUwyyEVVVRUeeOABjlpoAyMVLxfLLnmSpBeod1omk+H73//+B1ZS0BCF3W6HVqvFli1bAICHY65Vdpck4eIeMHF/ErWmUFsPgSVQ9W7r1q1ob2/HwMAADAYDGhoa2GK0tbXx3KcY/p6ESDwDTiHNM888g927d2P9+vUM4uv1elFdXQ2LxYKWlhY4HA6ukK8EYSMhoBYUGjsUJ5sJY5O6B2ZmZrBu3TqsXbsWvb29DBEWCoUQCoXQ3NzMW/6oJUXMW7lczsJGB/f555/Htm3b0NPTw+C8Pp8PNpsN9fX1aG5uhtPphNFoLDvwy51o+ghYGA1MJBKIx+MwGo3sZdC+boPBwO1SV69ehVarhc1mAwBeKBaNRvleLJ5gEqeLSG69Xi/nOr1eL99bo9GIlpYWbpHJZrMwGAwfuP/LlUgP0J55SjkEg0FGMgdQlv4YHx/Htm3b0N3djStXrsBoNKKhoQFerxd+vx91dXWYmprintfFeoE2YpKHKZFI8Nprr+GGG27Apk2bGEiGQvympibU1dVxf6tarea01bXQkos5NEVDYTANq6vVah5/E3t3Esk8Crler8eaNWt4UH5wcBBarRYNDQ2MuSfOd5GlIIaIPZ+pqSmoVCrYbDbs2bMHjz76KFpbWzE8PIzZ2Vns27cPDQ0NrHhoidNyJolkflcN9T8Ssg0VDUjJUQ6SsCnj8TgsFgsj26TTaQwPDzMkFVVsxbBfVGwQe5VUfJiamkJVVRXq6+txzz334NFHH4XdbsfMzAw8Hg9uv/121NfX87if2IIvV6LmbbHc5nI5xGIxaDSastl4UqJXrlyBxWJBPB5HW1sbBGEe5ZzCxBtuuKGs6EaFCPHMPh1iipRSqRQaGhrQ1taGdevWcYGOwk2DwYD6+nrU1NQw8MxKGBGl6CcWi5V50+QRJ5PJsuKWRqOB2WzG4OAg/H4/du3axUWeq1evolAoYPv27Yz8JfYQKcoUbzEgJyoajUIqlcJoNGLXrl145JFHsGvXLly5cgWDg4P45Cc/ifb2dhiNRo5qPxKPkpL8NJ1B6DZGo5GXJ1GrBICy0UYSKJlMhuHhYeRyOWzZsgWHDx/+AEw/hR5ihlASmLwYl8vF7UH5fB61tbWYmprC1q1bccMNN3BSnpqql7vXQy074XCY82AKhQIajYa9TMpvkfeybds2DA0NIRqN8iGfnJxELpdDU1MTBgYG+KDRoRPzlpSkTqdjYaP3mJubg8vlAjC/YfDq1avYtGkTt3YUi0X2xJY7b8njE6O10/XTMAQRpRqo8kyFHDJCly5dws0334z+/v4yQAs60GLwDSKSbQKaptaglpYWDvmPHTuGW2+9FWq1GnNzc7DZbJxHXe5EsksbKWnijjou6BzTqHEmk0GTaA0wOUUjIyPwer3Yu3cv3nzzzQ9gDVBRkpQlNa/T30skEvz+979HMB+4+ssAACAASURBVBiEz+dDNpuF3W6Hy+VCfX09HnjgAQDzHRA04faRKEoa4woEApxHS6fT3Ecpdo/ff/99zMzMYHZ2lgEwCMhCp9NBq9XizTff5JwNjRQRQ0jbi/MRtDpUIpHg4MGDMJvNCIfDOHfuHOLxOD75yU/iySefhM/nw/T0NDo6OjA7OwutVrvsvR4qYpEHKUbyoVYSYF4Ie3t7MT09DZfLBZ1Oh+bmZvaaVCoVTCYTLly4wPO0YjQiCsMBsBdF7TA0W/z222+zx0qjpXfccQcee+wxBINBXuxEi5yWO28p9yieyhEEgUF26fexWAzpdBoWi4UVZnV1NecgbTYbtm/fjv7+/jI4NPpfPP5JYTc5B6RAxb2BwWAQLpcLsVgMf/u3f4tCoYCRkRHeN02TKsudqAgWDAaZz6lUCna7HYIg8ARSLpfD2bNnGawFmJ+8ETea19TU4Le//S33UYs7BCgEp44AsbIkyMGjR4+iqqoK8Xgcvb29GBoawn333Ydvfetb8Pv9GB4eRnt7O3w+H9+ba6EltweFw2GMjo7ylzeZTNzyQ4laSlzr9Xo0Nzdjw4YNvM/44sWLOHHiBE9BiFGgKU9HxQpqb6EEPOUVFAoFzpw5U9YcTZ3/U1NT8Pv9aGpqgtvtZqSc5V5wIFSZmZkZAChrQwEWdhEXCgVMTU1Bo9Ggq6sLa9asYd4SdNXAwABj7lGOjP6RcSPPkgo7Yq/91KlT3Orh9XqRTCbR0tKC2dlZxONxOJ1OeL1eeL1eNDY2LnuPkgYXqFmZ1oqQ4jQajSiV5gGNc7kcxsfHkUgkYDabuVc1mUziwoULXGAQj/ECC2kjchbEVW9xLpfkmVCKVCoV1q9fD5fLhbm5OdTW1sJqtcLlckEqlcJgMHw8TFsCKZVK5PN5XLx4ka+Rog3qWyXEfeKNwWDA6tWrGezmypUrOHnyJEZHR7mwS/wlGaaOBTEmAo3pUq59bm6OJ838fj97r2NjY3C73WhoaEAsFsPY2Bja2tquWS8IS2ntEATBB2Dyj+Tnx02NpVLJ9nF/if+JKrz96GiF8xao8PejpGvi7ZIUZYUqVKEK/f+Rlnc8WqEKVahCy4AqirJCFapQhT6EKoqyQhWqUIU+hCqKskIVqlCFPoQqirJCFapQhT6EKoqyQhWqUIU+hCqKskIVqlCFPoQqirJCFapQhT6ElgSpo1KpSjqdrmzVo0ajYfRnGl+isSXxUPu1NrYvHoej8aVUKlW2ooDwEum9CanaZrOVoZ/TsvtgMIhAILBsZ+2It2Ke0WgWjdot5q2Y6PHiFb2Ln6fniHcymYw3DBLR7+hnApy12+1l95X2xCx33qrV6pJ4Fz0ABi4muaXRRuLXtcqrGHOSHhPvCInoD4Hv0mvEYMzic0Wz/iaTCUNDQ/7lPJmj1WpLZrO5jL+l/96bTY8X/+5/4u9imV0sv/T3ABiZicZ8F79WqVQikUggGo3yJlPx/aVR60gk8qGyuyRFqdVqsWfPHqRSKahUKgiCwFvO6DDlcjmEQiGGoAJQxiASHmKiGBmEht3pYmn1qslkwtTUFO/7Fs/OEgL3xMQETp48iXvuuQft7e3IZrPIZDJYs2YNduzYgVtuuWUpl/onJ51OhzvuuAPpdJrn3levXo3Vq1cjFAoxTBotSBKDMRARVijtHwEWQI0X85Y2BhqNRszOzvKiJfG2QmB+vn9ubg5HjhzBXXfdhc7OTkaA6e7uxo4dO3Drrbf+6Rm2BDIajfjsZz+LdDrNsqNUKnnWGADDBRLeJyksAn8l3hLoLz1HMi7efkmoQhqNhrEnCTxafOBJhi9fvgyz2YzrrruOD/Pk5CTuv/9+lEol3Hzzzct6PNBsNuOxxx7jc06Ym2vXrmXEMZlMxtB85MCQ7NL8NwErEy/FO45IKZL+EK90DgaDfxA0WBAEdHR04MiRI5idncVdd93FwCUymQwOhwNPPvnkNV3jkkPvdDrNShJY0OqEICP29MQ7WYgZhI1IyOjkOdHriMRD8bQrQ4w9J/48Qv/W6/V49tlnodVqIZFIsHbtWuzatWtF7B0hwFZSksRT+p94Q2AM9Bz9LwZnIDAM8prEO1zos+h9aIeOVqvl1RDi95JIJGhvb4dOp8Nzzz0Ho9EIQRDQ09ODnTt3rohVBcRbgqoTHzTigxj1RyzDJF/AAm+BBbkTH066X2JoMHoNGSkxGg4wr2ANBgOvopDL5ZiZmcEDDzzAGJorhcR4nmJAZ7FxECM4iWWNwEQISIOwQ8XIVGS0iOhekJzTvaS/k8lkDDqdSCTK9u3U19fzObsWWpJHSWCyFouF0ZrFC9zpQsQuLlkQQv8gaPh0Og2/349AIIBSqQSbzca7fglinxQAMYCQiWk1gni/TqlUwo033ohXX30VfX19uPPOO7FlyxZesbrcEW4IUstkMsHv95eBtpIHIw55gQXsRPLkC4UCo3aHw2EEAgHeF75z504AYN4uFmRawUq/B8AL4vL5PPbs2YMDBw7g7NmzuPvuu7F58+aysHU5kyDMr0h2OBy8C0csp+LwWfxYvDhMrVYjEAhALpfD4/EgnU5Do9Ew1qLJZOLDTv/EDgQpSiLxyoiGhgaEw2FcuHABNpsN+/btY8zPpex1+Tgpn88zQhOFw4VCocxwi9MKFCGR90grG8izjMfjiMfjmJubw3XXXQeLxcIA38RTcehNn0MePTlqyWQSmzZtQiAQwMGDB3HnnXeivr6enbVrpSV5lKVSCS0tLbwSlQ4yrWugG09KjbS6GE9RfKEE8ZXL5TAxMYFf/epXOH/+PPR6PeMkEsI5CTZZXblcDp1Ox6GQIMyvzXU6nfj973+PHTt28A1Y7hBrRM3NzdBqtawAydumlRYkaGRE6GanUqkyKy6RSBg6DQDcbjd++tOf4vjx4zAYDAx0Kn4NCRlh/JESoPe0Wq2oqanBuXPnsG3btrL7sdyBVYiXZrOZlSMpQeIjvQ5YyI1pNBpkMhlWWiqVig0v7Uf3eDwYGxvD7Oxs2foHugdi3pDRpy2adC/lcjkaGhrw0ksv4b777mOA5ZWiJIlXRqORH1MKiORHvAlUjLhPcmQwGGAwGMqQ9w0GA9ra2uB2u/Hyyy+zTAJg/pIhAxaMEa2jIO9zYGAAbW1tyOfzqKmpKVPS10pL0iBSqRQ7duxgQRJjHYq1uHiPBUHoV1dXQ6VSQSKRcG6MgFMFQeAVCJcuXcIzzzyDXC4Hs9lcxhjaKFhbW1sm3KRYstks7r33XvzmN7/hxVyUD1nuh5m2G5JQiUM2UvbiFbMkcAaDAQ6Hg8GJVSoVb78jJUZ/Pz4+jv/4j//gVRLEW0EQWNnW1NSw4aH8ZrFYRDqdxv33348XXniB97iIE+nLmUqlEjZu3MhyIgaFpvBbbCToAAUCARgMBt4dRH9DYZ44FJydncWVK1e4MCPOaYoVh0ajKSuo0ffJ5XL4zW9+w04FAdH6fL6Pk3XXTLR9QJynpP9JV4hTPnTNFAqTMSIjRDqFQJW7u7tx4MABzM3NQa/Xc86T1nCk02lotVr+O3FxGf8fe+8dG/d93o+/bpDH23c87iGSokgNihJFatqyhpPYsWzHMw7SLDsNMpGmaLPqBkGCtEFSNC2aFjHaDCTOaJzIVhRbbWrLS45lWaYWNShS3BTvSB5v78X7/cHf6+H7aKcWA6Q64qsHMCxRJO/u+bzfz3w9rwdAVVUVHnjgAbFTWq1WyJevRZZlKGnR1TqMarmpGJVVm5eZYTi9LNM9nU6HdDotkU0sFkMgEMDXv/51zM/PS3Spdgk7OzvR0tJS0N2msX3ooYfA7jHfH7fmFbMYjcaCzidFbYQBeJPxYwSqpnL8Pkb+bJyxwfU3f/M30Ov1sgKCuk2n06JbtZNI4713715Zi8sLzlS/mEWn08FqtRak2GokzM+jpoxarRZ1dXVynhndqQ6Lzpkd9dnZWbz00ku4ePGi7HhauhyrsbFR1meoz3rt2rWyptZkMiGTySAWi6G1tfX6KG0ZotFoUFVV9Zb6Xdps5Nf0er3YBTWy43Ngpsn/Z7NZtLe3w+/348KFCxJ1MqCgs2ttbS1YoseAYnx8HBUVFcjn80L+y7Xb1yLLMpRcFM/Uj4rh11QoAC8zVxmwacBti1wBodfrZTG83++XtMXhcODTn/40rFZrgRcGFi7v/v37pXgMLFzy7u5utLW1ye9mffTkyZNFbyip26XNnLfalwxAdqMDhatBuTaYemTNl7t49Ho97HY7HnnkEdTW1sJoNBZAY9LpNG677Tb5OrBQf+ru7kZPT48wp/N9nThxouh1azabpem0NNLhOVW/xktGRm2m3DR6dOwul6ugZslVGkNDQ4jFYpIaUlKpFFpbW2V1Ch35pUuXsH37dsTjcUnpecn/8z//8/9eYcsURoRqo1aNIBnBqf/OPgMNJ3XLf6Pzp+NS0+qXXnpJ/p0/Oz8/j5GREaxbtw7l5eUFaJpYLCYZBZ2T0WjE1NTUNW+5XHbxTrXW/JDM+dVtaby4NGZM7yKRiNR7uMrB4/Egk8mI12dtzOFw4IEHHoDL5RKvwddsb29Hd3e3HGyr1YoHH3xQjAEjrqmpKVkEVeyi1s2ARV3TyfBBExLFWpdWu7CdMRQKSQOAUbXP50MqlYLVapWDqtPpUFNTIxsVGdkDC8+uubkZXV1d8loOhwPvfve7AUCcm06nw/j4OC5dulT0ulXP6dLOv5oG02jOzMzAbDYLpErdhcPfYzKZMDExIUZTbVqUlpbi2LFjUjrh67CG2dLSIlnS1atX8Zd/+ZditFOpFEpLS+FyufDSSy8VLD4rVlHhUur5pV1Qm768rw6HQ3TCz82022QyyQps6pNnLJvN4uabb8YvfvGLgmyTTmlubg7r16+XsqDFYkF7e7sEH1wZEYlEBBp2LbJsQ6k2bNQOFi0+L3UkEpHdFcBCxMeulsViQUlJCc6dO4fJycmCIi0Vz0NlsVjw1a9+FVartSDtnJ+fR1NTE9544w0cPnwYDz30EObn5wVjyJ0kBw8exNzcXNHXKAEUwBXUC8zIkjCpubk5JBIJOZDcL6TX66VhMTAwgKmpKVlFqx5mOi+bzYaPf/zjslaYnjmbzaKyshIDAwM4cuQIDhw4gNLSUokm0+k05ubm8Oyzz8Lv968IQ6mulAUWSxrUIbOPDRs2SDTNSCidTiOZTCKRSKClpQWxWAxDQ0Oyv0mF+vB+ZLNZ/Nu//RssFotkYIx+7r33XkxMTGBqagrt7e3Q6XQFTaL6+nr8+te/hk6nQzgcvg4aW76ozl09r0tLHMBCYMTSjuqEysrKRNcqLI4Zq6rb9evX44EHHpAAgEFGNBrF+Pg4YrEYzp07B4/HA5fLJZtYU6kUIpGI4JGv9ewuu+sdj8el08QPw2hHNUbDw8OSMjNtYQc7mUzKIiKCoxkhccEVvbHRaMTJkyflkDHC+dWvfoXPfe5zGBsbQyAQwPHjx8Wz0BD85je/QTAYXBFL5AGII2EazIuj1ieBhSYD0wvWu/R6PYxGI0KhkIDHeeEZARErST2VlZVhaGhIPD4A2Gw2PPHEE/jCF76Ay5cvY25uDs8995y8Bo3uq6++iqmpKVkOVcyiNsiARSev1st5Cd1utyxWAxZq5owYN23ahOPHj0sUyUvMJo+K/S0pKUFFRQXGx8cLwNU333wzPvGJT8Dj8WBmZkZ0y+wrnU7j7NmzshWy2HULQCJCYBGy9lawwVwuh9bWVsHi0pBmMhmYzWYp4YTDYXleHDphAMZsCgA++tGPwu12y+8qKSlBf38/Ll26hGg0isrKSjzxxBPyHmi0TSaTlEuuVb/LNpT0BDSA6n88NFqtFhMTEwLj4ddoUKenp2Gz2aSuQagPO475fF7gAnq9HtXV1fi7v/s7SRt//OMf49vf/jZmZ2elTqdCY7LZLMbHxzE4OFiwlL2YRb1MAGSyibpVO4ZerxdGo7Hgc5WWliIWi8Hn88FkMhWgEIxGo3QXc7kcHA4H7HY7SktLUV1dja985SviXJ555hn84z/+I4LBYEFnmLqfn5/H+Pg4Tp8+LZMuxS6MktVOM5tbajkjm83ilVdeKZhO4gVuaGjA66+/LiOyjIRYFgIWMoJUKiUlpdLSUpw6dUoMa1lZGT75yU9ibm4OFotFXoNpZzKZRE9PD3p7e6VevxL0yzsMoKBbDSxGmLzXg4ODUk4goqOsrAw+n082KtIYql1wNri0Wi1MJpNkpkePHpVnWVJSgoGBAcTjcdkM63A4ClAiFosFp0+fFt1fqyzbgvBCs/DKSJJvJJ/Pw2w2IxwOIxQKifcAFi6/1+uF3W6H1+st6B7ygDFiJLyHkVNFRQUuX76MWCyGSCQisAE2iAguNxqN8Pv9eOqpp+QiLFcp10uoV0InVN3yQhuNRszOziIYDMLpdMq/RaNRWX+aSqXESTFlZiSoNjZYJ6qsrERvb6/8HnpnnU6HYDCIzs5OJBIJmEwmeL1evPDCCxLhrwQwP4WOSD27Kg5Pr9djenoa8Xhcal65XA51dXW4fPkyysvLEQwGUVpaKnXfRCIh55D/8WzykmezWYyOjspzNZvNYrhvu+02yajKysrwgx/8AGVlZYhEIgUTQ8UsajlMTWXVMgfvYG9vL1pbW8WZ85zTDnCHOoMDNbKkscxkMuLY165diwsXLqCpqUnGqAFIt/ymm26S0UiWjJjV/skMJdM4KoEPUvXWAOTADA8PY/369SgtLUUwGEQoFILD4RB8mkajQSwWkxqOXq+H1WpFNBotwKOlUimpKahdybKyMnR2dmJqagrNzc0SLRw7dgx+v78g5S72A6emfyrMQo0ymTrEYjGMjIygo6ND0hAasmg0CpvNJjgxAFKntFqtom+ml8lkEoFAAMBiM0mn08FoNKKjowN+vx/btm2TWtzp06cxNjYmNR++32IW1hv1er1kMbzQvKSsY5aUlMj+b4vFApPJhPHxcdTW1sLj8QiKgEMQLEewPs6GQTqdRnl5OQKBAKanpwsA/V6vF+vWrcOpU6dw8803iy7ZeGMKqnbki1nUs8u/q40btRFGVAb3w0ej0QKUBpuRam1TxWrTsOZyOYka6+rqClANJSUlaGhoQDabxbp16+T5ms1mXLp0Sc7Acs7tHxVRqlgzhsMqdlKrXRiKD4VCYsFjsRisViuCwWDBGKLD4RALz/9YAAeAQCCAQCBQ4P2Zkm7evBmrVq1CW1ubFN3T6TROnTpVoPyVEvHwAKi4UL53Vb/pdBrhcBherxdmsxmBQECiEHpjnU4Hs9mMyspKBAIBGecym83yPalUSowpsDgTnclksHr1atTV1aGlpUV+NpPJ4LXXXiuAgqwUWTqypl5iRo4VFRWw2WyIRqPQ6XSorKzE5OQkqqqq4Ha74XQ6EQ6HBTzOqDGZTEo2QzYtlkFSqRQMBoMYUgBoaWmBwWCAx+OR32U0GnH48OGC96s6zGIXFekCLM5uMyvK5XIwm80AILbBZrMJCoafU61dVlVVIR6Pi1FkIzKdThecP3XohO9Fp9Pht7/9LTZu3IhYLAaj0Yje3l65T4xSr1WWPevN6JHCSFD9u9lsRjQaRTabhdfrRV1dHcxmsyiHkV5jYyOamppw8eJFVFRUCKyluroaR44cQV9fH7xerxwYXmIA2LNnD26//Xa8/PLL+OpXvyqRUG9vrwBUVaKNYjeWqm7VzvdST22xWGSKhuULg8GARCIBq9UqF3PNmjVwuVyYnJxEc3OzGMjGxkY899xzOHXqFCKRiBg9emedToedO3diz5496O/vx7333gudboFthw5opczPq8LLpMpb4f7obDSahUEFGkmbzSaOpbKyEhaLBRUVFTh58iQaGhrg8/nQ2NiI5uZmPP/88/D7/QAWUkBe7Ewmg5mZGezbtw+1tbU4cOCABB0Wi0UCCRWDuBJETbHVs6s6eY1GA5fLJWPJKpOQCiMi5M1gMKC5uRnnzp1DW1sbDAYDjEYjgsEgTp06JY6HdU5G942NjVK+O3DggDRv2Nhkyq02m65Flm0oVVlal1BD7Pn5eaHtMhqN0sBJp9MoLS2FzWZDS0sLKioqJMXJZDKor69Hf38/BgcH4ff7JWzma6xfvx7vf//7EQwGcfToURw4cEAwg6lUCq+//rrUd9RRwGKXpReDBlMF2TOqTKfTiEajqK+vl5qYyWSS9KKkpAQ1NTWora2Fy+USeMWqVaswMTGBoaEhIYdQL+X69evx4IMPIhwO48yZM+jq6sLatWuRz+cRjUbR19cntWOVuaXYL7QKLFfPqBpdZLNZBAIBqY83NTXB4XDA7XbDbrcjkUggHo+jqakJyWQS8/PzqKmpwebNm5FIJFBTU4OOjg787ne/w+zsrBgHXsiZmRm8//3vR3V1NcbHx/Hkk0/iYx/7mKSVR44cQSgUKoBxrRRh2YJOd2lJhud2cnISDocDY2NjWLt2rdQaOb7MwKampkamlFwuFyYmJuB0OuF0OtHX1yd3QJ2Fb2pqArAYyc7MzKCzsxPAAkqEhhlYnP5ZTkb0RxlKdl9pJBlV0jNHIhFs3boV4+PjaGlpQTAYhMPhkDfY1taGlpYWWK1WNDQ04OrVq5iZmUF9fT2OHj2K3/zmN9KxyufzCIfD2LZtG26//XYEg0EcOnQIyWQSn//859HV1SVNm0uXLiEUChVwVq6UgjhFPWj0sjRGGo0G4XAY+/btw+DgINauXSuRZD6/MO/KlNlkMqGhoQETExNIJpOoqKjAiRMncPjwYQSDQSEwiEaj2LlzJ3bt2oV4PI5nnnkGuVwO99xzj/D3abVaDA8P4+rVq+KRVZTBShAOIqj4SGAx0+D5veWWW/DMM89gZGQEJSUlKC8vFwiUy+VCS0uLOPaRkRHk83nU19djdnYW3//+9wuo8hKJBC5evIivfe1rGB8fx9DQEI4ePYqenh585jOfQTQahVarhcvlKsACqhjalSB876oQtqayA2WzWTz44IP44Q9/iEQigerqasTjcWlAtrS0YHp6Gj6fD5WVlZibm0Ntba1E3ZcvX5aIn7Xi+vp6iTbdbjd0Oh0CgQD2798v01FtbW148cUXpfTxx9TWl33KSSsFLIbabP1TYYlEAh/60Ifg9/sLoh2tVovm5mY0NTWhqqpK0hGXy4W6ujqcPHkS586dk/SRSgaAsbExOBwOocv61re+hU2bNklHN5PJ4Pz58wWQFUYRK4W4gV71rSI2VbcPPvggPB6PNGYYNbe1taGqqgoVFRVS+62rq0NFRQV6e3vR19cHv98vuqVe+vr6pCDe0tKCv/7rv8aBAweEci2dTmNoaAhAYaFehd0UszCiJOxnaZOE55iRitlsRkVFBaxWK3w+H3K5HKLRKNasWSMTI+Q5rKqqQjgclmYlgwYiDtavX4+5uTnY7XZMTU3hAx/4AG6++WZpvmm1Whw9elQMixqMLMUmF7Owc88/AyiI2BjBzc3NCcqFsCmNRgOr1Qqv14tcLofNmzfDarWiublZKOyMRqP0O4haAIBf//rXaGpqgtlsFvatPXv2CBoBWMB022y2gve0tEP/drIsQ6l6X/6dMAtCW/L5vLCcx+NxrFq1StLgiooKdHd3o7OzE6FQCBaLBUNDQ7Db7ejt7cXAwACGhoZQXV0th9rv9yMcDmNqakrYixoaGtDS0oJUKiVewu12S/cWeOtZ3mIW6pb1MQDCnq1iKbPZLObm5qDT6URP8/PzcDqd6OjowNatW5FIJGAwGNDb2wur1Yrh4WGMjY3h7NmzcLlcMpMciUQQiUQwOzuLVCqFXbt2obKyEh0dHXKQgQXdjo2NvWleeiVFPEsbI0sB0TzLdPj19fUS4QWDQWzfvh3l5eWSYvr9fpjNZgSDQczNzWFsbExIHux2O8LhMFKpFJLJJHw+H6qrq5FKpVBeXo5MJiNNtLq6OuRyOZnA4XtUR/5WgvD9qs1TdqnV5svg4CBWrVol5CBkKK+pqcG6devEOR09ehTJZBJutxuhUAh9fX3S0Sa0KJ1Oo6WlBSdOnEBTUxOOHTuGNWvWAFictDIajRgZGSlAaahZ2rXKsiNK9cAx8lGR9yyyRiIRdHZ2oqSkBIlEAp2dnbjtttsALJAD1NXVYXJyEpFIBIcPH8ahQ4cwPT0tyjOZTAgGg9LlDgQC6O/vR3t7u8xpsgaZyWTQ29uL2dlZABCaMU4GqeORxSyqblX4CoAC0uNoNIrNmzejvLxciBZuv/12YZypr6/HxMQEAOCNN97AE088gfHxcTk4RqMR4XBYUAbBYBAXLlxAW1sbZmZmRLfsBrOmSVyrzWYT5hWms8Usahq7NPVmOWh+fh6xWAz5fF7OLSE9u3btkotptVoRj8eRSqXg8/lw9uzZgpJEVVWVdHLJUD4xMQGHwyGNHd6TeDwuGRQbZktRGsWuW8pSeBvxvfw8xOjq9Xp0d3dLgyUajWLTpk2YmJiA2+2Gy+WC3++HxWKBxWLB7OwsxscXNmGoBC6sRRoMBgwNDUGv12NmZkb+jfcFgDguo9EIs9ksdmE5ZaNlG0o17Vo6y0ljye7Se97zHsRiMTQ0NGDdunVIp9OCqk8mk6irq4PL5cKhQ4eQSqUwOTmJLVu2AAAikYiMMhIs3dfXJwc2kUggkUjIpZ+YmBAvzp09RqNR0tBir6XxwakNHOqWF51TDPF4HLfffjsikQjq6urQ3d0tdF86nQ6xWAyrVq1CZWUlfvaznyESiWB8fFwoqEKhkHQfiVm7cOGC1HQ418yJCY6EpdNp+P1+SRsrKirApVLFLEvrvrxEarpLmA45IJPJJDweD2pqahCJROByuYSI1ul0Yvfu3XjhhReg1+sRDofFgaiz2YxaRkZGkEgksHr1aqnt22w2QXx4vd6C6TYSJ6vcmMUuPJ8sF9BIqpEml7lNTEwgl8shHo8LK/ratWtRVlaGcDiM+OZR9QAAIABJREFUxsZG3H777Th+/DhKSkowPT2NiooKSdlJDajX6wVhMzo6is7OTtEhae5OnTolJZFUKoVwOCzNI5Y+rkX+qIhSxS+pB08dFZqenpZpmbVr1wonH7uy9Lq//OUv4fV6odVq0dPTg5qaGnz4wx/Gxo0bhY27trYWf/ZnfyYGgXgsgqCvXLlSwDTCmiiBrctF4V8P4eF6K91S53yo8XgcHR0dsFgsskYgHA6LtwwEApiZmcHjjz+OmZkZWK1WbNiwATU1NfjzP/9zdHV1CTt8XV0d7rvvPuzYsUNA/gReA8DU1JQ0y0pLS4VVJ5FIIBwOF5BzFLPwfAKFnX5GHYxOpqensXfvXkxMTOCBBx5Ae3u7NCiJJV23bh1effVVjI2NYWhoCM3NzWhoaEBbWxva29sLJqJcLpfUe9euXYtYLCZ8CfF4XCbX1HSQUedyO7PXU3h2VWe/dKFaSUkJamtrMTc3B5fLhZqaGqndEkjOevjvf/97wf/W1tYikUhg3bp12LdvHwAIqcvGjRvFLuzZswdGo1EMYWVlJcxmM9LpdEHAxfJgJBK55tLGsrrePEw8cHyIKmmsVqvFpUuXoNPpMDo6ive85z2Ynp6G3+9HU1MTSkpKZGLnu9/9Ls6dOwebzYZHH30UDQ0NuHjxIpqamnD+/HmEQiFcuXIFX/nKV2A0GtHV1SXcdIlEQiLFc+fOyfjSUuIDpt7FbihV3apg6KX8lIODg5Jm3HTTTZiZmcHMzAwaGxsBAGfOnMH27dvxk5/8BBcuXIDVasUnPvEJrF69Gr29vdi0aRNOnz6N2tpaDA0N4ZOf/CSqqqqwbds2wUf6fD5UVFQIt2IkEpFnzIOlkj8Uu9AQLa3/ERmh0SwQZLzwwguwWCw4fvw4Pvaxj+HUqVOYm5vDmjVrpKbrcDhw4sQJXL58GYFAAB/5yEdQVVWFTZs24fz58+jv74fL5UIymURrayseeughRKNRAVwTEpPL5fD666/L7im1fkY9F/uZpTC6U8+CSsdH9Mbo6ChcLhe2bt0qk0hTU1NYtWoVZmdnZWKppKQEo6OjMJvN0qS5evUqenp68D//8z+w2+0IhUJwuVwIBALYvXu33P/JyUmZoyfigwZcdZbqCplrkWWTYqisKYx2WOsyGAyw2+24cOEC7r77bqxevRplZWVCgFFXVwe/34/W1lZcuXIFp06dQjAYxI4dO+D3+/Hyyy/j6aefxq9+9SuYzWZwh3gmk8Hvf/97mQN1Op2YnZ2ViJLdddX70vBYrVYhDy5moW5VcmNgkZ6Okfj4+Dhuvvlm1NTUCAlvJBJBTU0NfD4furq6cOnSJZw5cwapVArr1q1DKBTCs88+i76+Pnz729+G3W4Xajuj0Sg1SHKEMjUnAQZraKytcd8OD/JKaJbx0gKLtWDCTvx+PxoaGlBbWwuz2YxHHnlELq3BYIDT6RS4WltbG/r6+jA+Pi7UftFoFK+88gqOHTuGkpIS2O12nD17Flu3bsWJEycQj8eFwZ6oA61WK3R5AMTJ81JzcMPlcl1PtV2TqNnQ0to602s2b41GY0Gj1+fzwWq1IpfLCbB8enpaeEAvXryI48ePY82aNfjGN74hGWMoFMLMzAza29uFgMfhcAhrViKRkFFbPmu1Zsra+rXahWV3vbnsCoB0tObn55FIJNDf34/R0VF897vfxapVqzA+Pg6r1Yrp6WnMzc0BACoqKpBKpXDw4EF4vV488sgjmJ2dxcGDB3H8+HFkMhkMDw/D5/NJV/Gll15CTU2N1CgrKiokSpienpbQml6MtTdOqrz44ovLIum8HsKoho0nMj2Tqebs2bOYnZ3FV7/6VWzZskUaBJOTk8JeXlNTg3A4jIMHD2Jqagp33HEH4vE4Dh06hFOnTsHtdkuX1mw2Y2ZmBseOHZPyxfz8PMrLyyVlcrvdBR1LdfGTxWJBPB7H8ePHi163NIw8t0QPsOHicrnwq1/9Ck1NTbjnnnvwu9/9TqZAGEmXl5ejsbERv//973H58mXcd999CAQCGB0dxeDgoFxKYiN5XjUaDcbGxqSTzjpwU1OTGEw1UqehbGhowOjoKM6cOXM9VXdNwqyNUzVA4YDE9PS0bDLYsmWLBE1jY2Nob29HPp+XccXJyUnE43GUlpaivLwcRqMR2WwWR48eRWNjI2KxmAwIhMNhHD9+XAZTqqurEQwGZdabgRYdEJE0zEiXw6W67OViLJKq6YJGs0gVxXT4xIkTAny22Wyora3FyMgI9Ho9XnvtNbz88sv4i7/4i4U3odUiHA5jcHAQExMTQrpLjBWJBWZmZpDJZFBeXi4pU39/v6zI5HtRAaknT54s+p0uwKJuVYJSCqeZHA4HdDodXn31VVRVVcFms8HhcMDpdGJ4eBharRZDQ0O4cOEC3vnOd8LlckGrXVjKNjw8LIzc1C2hGX6/HzMzM0in0/IzWq0Wk5OTbzKCvBDhcBgXLlwQI13Mwtq1CmtT/z88PIwNGzagvb0dAwMD0pSIRCJobm7G6dOnMT8/j9HRUfT29mLr1q3CspRMJjE9PY3JyUlZO5DL5YTchY6JaA4iOdxuN3w+X8H7pJGsq6vDG2+8IdFQsYs6gaRCmljXNhqNqK2tRWVlJS5fvixLverr61FVVYXz588DANasWQOfz4fZ2VnU1dVJU4vnkRNpmUwGFRUVQlStEoKTpFuj0QifJxtMDK4Y4JEg+FpkWU9Bq13ggqNyWJPg7HZbW5scFhJn5nI5tLS0oK6uDkeOHEFZWRleeeUV7Ny5U2qOXE3JFRDs+DH9YEhvs9mQzy9wAJKcdnBwEKlUSqIepjAGgwEXLlwQpRR76s3PykPByJKz1R0dHQUd2VWrViGdTqO5uRmtra14+umnYbPZcPjwYXR3d+O+++6TfS/knuRh4nOjnnmxNZoFOrvR0VGk02mMjo6KB6cnZtNhYmICsVhMMopiFr5n6lYlgCXHgMFgEKPf3t4ukyNtbW3CFJRMJvHTn/4Uu3btAgA5b8Bi1Epj7HQ6AaCAd0Cn08melt7eXmmEqfhDg8GA4eFhyZJWCsO5upaEEVwmk0EikYDZbJaskr0FwtwqKirg8Xig0+lw9uxZDA0N4e6775ZIsKysTM4moW3AwlZF/lkdfBkdHUUikcCVK1dkTBdYhLERGE8c65/EUAIQyAKVwgueTCZRW1sLi8UieDQap6GhIQwMDOC9730vrly5Aq/Xi0cffRTBYFDSZYPBIF0qXmx2rlgXXb16tbwOI0+/3y9GgJFAaWkp5ubmBHaxnLWU11PYLaVn1mg0sNvtBetLuXCJ4G+3243p6Wm8613vQn9/P3Q6HT796U8jEAgUMMVzpp57RqhvAtwbGhqQTqdRUVEBt9uNeDyOq1evCu8fD6PBYJAiPCeFVoJu1cYNowwW/7u6ugAAs7OzyOfzknJ7vV48+eSTKCkpwbp16/D444/j6aeflpls/l4VyqM6HBrniooKgadwhJeYQgqdTXV1Ndxut6xTWSnCDQYqeL+xsVGar1qtFvX19UIwTV14vV5s2bJFpqFuvfVWzM3NFZRJaB/4e8j0pJZUWLecn5+Xch+DJtoijWaBVIasTcsJnpbdzEkkElKH4YulUik4nU4ZRyRFVV9fHy5fvoz29nZZu/n1r38df/VXfyU1MEYq9Ops79OTsCFUU1OD1tZWIUydmZnB4OAgYrFYQaRA5Vy6dElGz1bKmB11S8+Xz+eFrYZGNBKJoKqqCgMDA0JYajQaUVlZiW984xt4xzveAZPJBI/HAwByyPgfO9sVFRVoa2uD2WxGXV0d1q1bJ+TJc3NzmJyclMus6laj0WB0dFTA7YSxFLPwspHnlBeMIGR+jY5lbGwM5eXlsjFxx44d+Od//mc89NBDSKfTciHV2hcZzDk6x5314XBYUvWpqSmJQInhpPFmwEDECO9YsUfrlEwmIxEcI+SrV6/C6XRCr9fD5/Ohv78fra2tsobBYrFgYGAA6XQafX19OH/+vOAql06B6XS6ggk96tdgMGDjxo2IRqMYGBhATU0NqqurC4Y32LwhXRvt1nJ0u2ziXq1WC5/PJ56ZzMMkjI1Go3A6nfB6vRgYGEBzc7Nww+n1etxyyy1obGyUGoG6WF4l5JycnMTAwAC8Xi/279+PTZs2CUiVEAtSqvEyq+D3aDS6YqIdYPFhqgvvqVsCvAl4JrvSnXfeCY1GI3tc7r33Xmzbtk1G51iSUKmoUqkUPB4Pjh8/jpmZGezZswetra3IZrOIx+MyFNDX1yfTOUDhSCiJR5ZLVXW9hI7H7/cXnFuWN2w2GwYHB9Ha2opAIAC73Q6PxwOTyQS73Y6///u/x+TkJG6//XbhRVzaQSWAfX5+Hn19fQgEAlizZg3Ky8ulTszZ5hdeeEEMqtr8MJlMGBsbWxHORxWtVisUcer7ZjPWZDKhs7MTZWVlmJ2dlQAoFAohl8thdHQUmzdvRltbmzhfFbhOgDnRB+fOnZPmJaN71n0dDgeGhobErgCQKStmQ2810vp2sixDyVRCnc7J5/OCsyNMZXZ2Fo2NjbDZbLh06RJ+/etfY3Z2Frfeeiv27Nkjl5ILruhRjUYjbDabQCZMJhM6OjrgcrnQ0dEhNE7ZbBYGg0FSJVV4ERKJhMyBr4RDp6aFwCLkgqBYRpZutxttbW2oq6tDb28vLl68CL1ej56eHvT09EgUbTabZdkYpyLMZrOkPHq9Hu3t7aiurkZXV5dc8nQ6DYvFArfbXVAAV8sBsVisgNev2PXLiFiFB+XzeVitVhnN7OnpAQDpuL722mvo7OzET3/6U9x999249dZb4fV6pVNKomQV96qCmLkLfO/evfIetFotamtrhfF8aX2SNbdEIvGmxshKENVpElmQzWZl9QsJpm+++Wb84Ac/EBrA3bt3y9ROKBRCbW0t4vE4vF6vnDGbzSYwucrKSmnuNDc3w2AwCDdBfX09JicnxdgyktRoFtagqEiO5ciymzkABLICLM4gq3RJbrcbMzMzsthqbGwMq1atwvr16+F0OuVD1tTUoKKiQjqwpANraGiQ3cZr167FTTfdJIaEhASMtNSRP/6fo0q8xMXeyAEK16ZycRUPDsfrUqkU/H4/RkdHodVqJU1hDZMHaH5+Hi6Xq6BRkUgk4HK50NDQIKWR9vZ2bNy4sWAvdTQalZURanpCPbNZp44BFrtoNBqYTCaJ1LjTZnBwUKI/MsKXlZVhbm4Ot956K/r7+/HJT34SJSUlwjFJ3VmtVslkMpkM7HY7ampqhOauu7sbBoMB8Xgc8XhcmkH19fXSbVWjSWAh8lFRBislsqQjBhZZeTKZDNxuN8LhMPr6+lBSUoI1a9YIt8C+ffuwb98+jI+Po7e3V34X+w12u12aOYlEQmqOo6OjGB8fRygUEiw1S0rAQkM5kUjI39W7zxoqdfonYzgnDIgFa0Ym7MqGQiHo9XqcPXsWdrsd9fX1iMfj8Pl82L17N+rr6wXKwxGklpYWodfP5XKYnp6WFNrpdKKsrAzbtm0TMl7WREmtpoJG1WkAemUAK2KJvFa7sD6D8CrWwKxWK8rLy8XTcosd4U/T09N46KGHsHXr1gLdWq1W4QJldDo7Oys1tcrKStjtdtx0003SAWRzhrP49OaZTEbqd6z18Vmw2VPswkVsPBNsNqTTaVy5cgV6vR79/f2wWq1wOBy4ePEiEokELly4gN27d0u0zTTQaDTKPWDkz9KIyWTC+fPn8cgjj8i4HJ+n0+mUCIelDRpcOn8CpBlhrgQhJI/NRq1Wi+7ubhw/fhzBYBDNzc3Cql9TUwObzYZnn30WZWVl2LVrF6LRKMrKymAymaTPYDabEY/HhYgkEonAbDYL3nj79u0STPDck6iaz1olQS4vL192t5uybEOZSCTkQatvoKamRuqE9fX1sFqtCAQCCIVCeM973oONGzeKRwYgMCODwYCenh6cPHkSiURC6pOJRAIWiwW33347UqmUNHjImq6+PrCYXrEEwK6mOnJX7KLOxxLOYLfbZW2vRqNBdXU1DAYDrly5gkQigbvvvhvveMc7ACyOlKrjj1u3bhXdxmIxTE5OirPjQWOdl7rlxVWfM5s5sVisYCZ9OWNg10s0Go2sFFEL+zqdDrW1tQXOoKqqSlJjr9eLD3zgAwVnh/VenU6HmpoawUgmEgnMzs4inU4jlUph1apVwvbEvUZOp1P0S2ysiulkuq5Ov62EiBIArFZrQbmGwUpTUxOmpqak9r5t2zYcOXIEdrsdTU1NsrqB32+xWOQObNiwAWNjY5KdculbKpVCMBgUFIxWqxVDyyYPHY56J3gv+LXlUK0tezKnurq6YAqmpKREultMcbLZLFavXi11BT5wdZEVu9s8EOXl5QJh4fQPvXgqlSqIJtkEam5uxurVqyXFJDaOwHYyhDz00ENoaWlZzkf9PxcWq/lnRn6kw1dxjG1tbcJrODExIXRVNGwkN2UToaqqCtXV1UI/x/GwkpISafgAEPqwbDaLVatWoaWlBU6nU5oeBoNBoCssq7zvfe8TGv5iFqvVKh1uZkPPP/+8DEQYDAaZOSbkZOfOndL5V9nRmUaz3FFeXi7wLRrQpqYm2elCw5dOpwWlwf0xvKxarVYifmZIHo8Hd95553XW3NsLozmOFLPmOjQ0JNmLVqtFR0cHJicnsXnzZrS2tsp+bTZsiFdleeTEiRMyecdpME7kAIvNZU7gAQtN3NLSUtTX14uOAQg8C1iEinV1dQnBztvJsgwl2cTZyaYlJ91WNpuFx+PBzTffjP7+fuGhtNvtOHz4sMzWcp6Zh29qagqDg4MIBALIZrOw2WxwuVzweDz44he/iM9+9rNwu90CB6B3YeewtbUVPT096OjoQE1NDQKBgNC57dq1C5/5zGeKPqJMpVIwGo1obW0V70cMGscKZ2ZmsHv3biEx7ejowKpVq/DYY48V4PqoY61WC6/Xi4sXLwrYl8Vwn8+HRx99FB/84Adl7zF1FgwGxZC2tbVh27Zt2LhxI6qrq+H3+2WCaPfu3fjgBz9Y9DuJGPER26jT6ZBOp7F//37B41HH4XAY09PTsFqtqK2tlSiJl5iZChdkkRyDk2lOp1P2Ob3yyisYGxsrKP243W5YLBYYDAa4XC5UVVXB4XAIKJsG1+Px4Dvf+Q6am5uvn+KuUZLJpEwxMWqrq6sTDDRr7LW1tRgeHobFYoFOp8O+fftw5MgRWK1W4QxgpM9NoACERESr1cLhcKCmpgZlZWX40pe+hIGBAeG2LC0thdvtRiqVkgmdhoYGNDc3S9pNw1lZWbksdqZlWQ+/348XX3xRXgRY4I2kpWcDQKPRoLW1FZs2bQIAXLhwQQqxGs0CVyVB4GNjY1K0pSchUwsJBi5duoT7778fjz76KKqqqoTheG5uDjMzMxgdHRUAdDAYFJhLV1cXvvKVr2BsbEyaT8Uqc3NzOHr0KKqqquRrbrdbImumZaz3tLa2Qq/XY2BgADabDf39/VLnZJNrdHRUakfqJee0Den33/Wud+HjH/84ampqYDQahShiZmYGIyMjmJiYkAYZG3mbN2/G5z73OUxPT4unLlZJJpM4deoUqqqq5CJ3dHRgYGAA+XwegUAAvb29MJvNuPXWW7Fz504kEgmcOXMG9fX1eOONN6TBxrRbhaUxKjKZTIjH44LKGBkZwQsvvICf/exncDqdAq0KBoOIRqOIRCIyPZVIJMRBjY+P4/vf/z68Xi/Gxsaut/reVnK5HF588UWhNwSAmZkZ9Pf3yy4cckPu3LkTsVgMpaWlOHv2LCoqKvDcc8/JBB11HAgEZO0MG4YMBtg8uuuuu+B2u3HbbbfB5XJJ+YMReTgcluaPx+MR/Gsul8PatWvFYV6LLHvW+9ixY5JGEBfW0tIiBL1Wq1VqDwyLSZ7LRgQ9hMFgwOjoKOx2uxg3fhBO/LATWVpaiqeeegrHjx9HVVWVKI/F8mg0ikAggLm5OUxPT0Or1eKuu+7CxMTEW8KIik30ej1eeeUVKfJnMhlZvhaLxQTT19jYiMbGRpmrr6ioQF1dHa5cuQKj0SgjiWRhcTqdUqpQu9QWi0WYyvV6PV5//XU8//zzqKysLDB8HEMLh8NiOHU6HXbv3o3Z2VkppBezlJaWYnR0VDIZ7rxp/v93stTU1OC2225DSUmJgKaZ6h08eLDgwvIikw2HtTWeQ9aGOUlCUoynnnoKDoejoHFG4pNwOAyDwYCpqSnkcjns3bsX/f39mJqaWhENHe6zYcqdy+XQ0NAgXJterxfl5eXS+GMwtXr1atETCaQBSK3R5XLJ2WLjligMlv1KSkrwyCOP4O///u/hcrmkjsnyFcso2WwWw8PDktoTi32tsixDWVZWJru5gcVVEMzzzWYzPB4PAoGAwFe8Xq+wfbDYzW4p2/jqSB7Bu2z7Uxms0X32s59FeXl5AWyFbN2RSATPP/880uk0mpqaUF9fj1AoJLWpYhaTyYRwOCy65YVmI8VmsyEYDCIYDAoFndvtFigPi9zcl8MpH15GYFG3bCiQEYjp6Oc//3lJUVTIil6vRzAYxPHjx5FKpdDc3Iw1a9asmDlkde85sIgwIClISUmJ4CNZciC/YVNTE0ZGRiT1BiA1XtbdVWiXirxQZ7/9fj+qqqoKaqSMkOrq6jAwMIBMJoPZ2Vl0d3eLs1Pr+sUq7FTTMALApUuX5OsMdHw+H8rKyoQZqbm5GZcvX0ZLS4s0ivlsgsEgksmkNA/ZRGNGoGJ7AUidM5VKSfbEYQsOqrDXsXXrVsnOrrUktyxDycVKnMNkAZebFokXI6uH2WyWdZNut1uYPgjvmZiYkOVhtPI0iLygKiyFKTpTQBVErNfrMTU1BbfbDY1Ggx07dsiUD5VazMJoh16OB4yHLZFICGckd380NjYKZyQ9OSFbk5OT8Pv9Mi/MFIPeu7S0tADsz7EwHh61I1hSUgKfzyd0YevXr5eRy5Wwj2gpLEjt9FssFpjNZoyNjcnmRa65SCaTaGhowPz8vDQmAMiMMlmpmOqprFqEuLDhQLwlo0yV6Fqj0cDj8QjMjsvGzGbzisIA07mSJ7WyslIi54mJCTQ3NyMUCiEWi6G7uxuRSAQ9PT0YGRmR2jedTTwel3UkjNRpB1QHxOd68803yyQbv8770NbWJpkPm8vz8/Ow2+1/GlIMWnUWrumBaeRyuRxaW1tliVIqlZJOtsViwQMPPAC9Xo9QKIRsNivpkMfjEQYipi6MdmgYVIP5ox/9SA4csLirx2AwwOFwoLq6GuvWrYNGo0FNTY3M1Raz8BIyiqSRJ3RFp9Oho6MDOp1OeCrNZrPo6N577xVCh1QqhatXr0r6TUwpD5s6n0xcGnX7r//6r285YcFnWFlZia6uLuEFXQljjERDcIKMWEcatqtXr0pazFKDSrqwb98+aLUL2xjJruT3+4UmTaUYUy+zCrXS6XR46qmnxECrz2NkZASVlZWIRqPo7OwU9EggELjmruz1FqvVKmkvAf1EVGSzWRlnDIfDcDgcmJmZgdfrFb4Ck8kkARR7HaFQSBqNLB2pkDUyjBF3/S//8i/iWNQVHyaTSbCY3d3dElz9yUgxmE6o+zEMBoOkeOw8cUwul8vh4sWL2LlzJzZv3ozh4WGYzWY0NzfD4/EIForb09LptNQieJnJZ0fjqdVq8cMf/hC33nortm3bJvhNr9cLp9MJu92ONWvWCJ8d32OxX2ZGFoyAtVqtwH7UmWLWGEOhEM6fPy8TDn19fTAajWhubkYgEIDf70d9fb0Ut1ny4IVm1K46IWBhT/KePXuwY8cO1NTUAFhoNJWWlqKxsRHNzc2ora2V3TorIVoHUKBLrVaLRCKBmZkZhMNhWRnr9/vlQk5NTWH9+vWor6/H4OAg7HY7amtrEYvFhF3I7XZLdKoSLPDy8j8axtOnT6Ourg5VVVVSr2PZqampSXDA0WhU8LMMIIpZGJX5/f6CMWefzydICqbEBoMBFy9eRG9vLw4cOICHH34Yg4ODyGazsFqt0Gq1CIVCiEQi8Pl8EigAi3AgOhlVx/l8XvokXV1daGpqkmm/06dPo62tTSJO1kBVaNzbybIxM/S6fHPEJxmNRqTTaekk6fULG9JIeaTVarFp0yapF125cgVWqxXV1dVS5+HvZARAQ8moktRr09PTyOfzcDgc2Lt3Lz7+8Y9j7dq16O/vh8fjwXvf+17U1tZK+M33Vcyi0WgKyhKMNuilqZdkMomysjJUVFSgvLwcQ0ND0Gq1uOmmm0S3Q0NDsNlsqKqqQjgclhSQHpS1X+qTjkiv18uSeavVij179uBTn/qUrO5wu92466670NjYKEMAy4FYXC9hA4YErxqNBnNzc7K5kheceMf5+XlUV1djcnIS/f396OzsxPz8PIxGI65evYrp6Wl86EMfkoif5QcVbE1Hr3bGTSaT3Inm5mZs2LAB3d3dmJqawqVLl9Dd3Y2WlhasXr1a5pyLvbYOQBojPp9PDFkwGJQ6OW0CtyO6XC488MADePrpp3Hy5Els375dAgCS7fb09AgNI58JM0PqmOU3ZkmrV69GNBrF+Pg4MpkMtm/fjm3btmF8fBxXr15Fa2srqqqqZJKKi/KuRZZlKPkh1JSDzNsMkXnJeTg3btwozNqMCoeGhgAAnZ2dOHHihHw/u1psPrBJw4vLEbGSkhJ4PB74fD7MzMwgkUigsbER4+PjuPXWW2XiJJFIwG63r4gtjIRDcYKBnX+bzSZ/JoSKEX1bWxu8Xi/i8bik7tyBXFtbiwsXLhTULdXOLGfATSaTgPz1ej3Kyspw5coV+Hw+ifo7OjowNDSEm266CXv27EEikRCigj+GYOD/Whilq+OxBoMBFRUViMVigmukzM/Pw+FwCIs5CWO5uvf+++/Hf/3Xf72poQAsjtHS8dMZ0Vi+8cYbQipiMBhQX18vUK57770zcx9oAAAgAElEQVRX9lRz6mclpN7JZFIauWy2EDZIfdhsNmg0GlkjPTY2hqqqKsEKMwMlKJ9Gkmeff2YPhLaGWSPvysmTJ6HX62VnOkd7rVYrurq6hGrPaDQuK1pflqHU6/WCvVMvHnGNaoTR29uLq1evYnJyEiaTCQ0NDfIBqLyXX35ZPA69J+sQ9DBMEdVlVjqdDk888QScTqdg4MLhMN773vfiwx/+sECEmpqa4Ha7JfwvZuFs69zcnFzmTCYjhpLQKa1Wi9deew1XrlzB5OQkzGazpG188Hq9HufPn5d1n0AhqYX6nKhfNjU0Go3oloQGc3NzuO+++/CBD3xAWGDq6urg8/mk5lfMks/nZbEX091kMgmn0ynFfxqzwcFBAd3r9QvE0cBi47KzsxPHjh0roPBjtMPd3OzSMjNi84yG0mq1Ip9foH07fvw4nE4nHn30UcRiMVy8eBENDQ1SvuK0VjELm2I+n0+MXjqdRnl5udgJErf09vZifHxcVvXa7XYxdG63W7YmBgIBAIuNIjqjdDotMC0VmUFHRwA7bcxrr72G9vZ2mRJks4jZ2bUOSywbHhQIBDAwMCBfIwM399ywcRAOh2G322Vqxuv1Ip1O4/z58zh16pREmUwH6TUIYSFOk0h91tOYgp89exYmk0kiXI7qTU9PS93J4/HIutFij3rKysoQi8Vw5cqVgq/RqEUiEYRCIWQyGenGbtq0CRs3bsTs7CxSqRQuX76MM2fOYGRkRJ7JUkwZDTBrRkQWsMNuMBgExK7VLmwKDAaDos9EIoHKykpMT09jdnYWzc3NRa9bYu4uXrwoOrDZbAKdYipmNBqFKFqv16OlpQU6na4gmvR4PIIgUD83MylGPWqkxOhHr19YsMUhAjbeurq6MDExgatXrwr+cHh4GC6XSwx5MYtGo5FzwyhbJbEJh8OYnZ1FJBLBjh07sHfvXrS0tMiEmIpFpSFUcdIs8fEME1KolqRY5mhra5NSUiqVwvDwsDTq6CRZj6bDuqbPuJwaiEaj8QIY/2OUWQTSlM/nK6/3m/hDckO3fzpZ4boFbuj3TynXpNtlGcobckNuyA35f1GKmynihtyQG3JDikBuGMobckNuyA15G7lhKG/IDbkhN+Rt5IahvCE35IbckLeRG4byhtyQG3JD3kZuGMobckNuyA15G7lhKG/IDbkhN+RtZFnLTgwGQ95isQBYpJZSZ2T5deCt10GqpKUA3kRgulQ4VcJVtpyqWEoWEI1GYTAYhHT2rcgEotEo4vF40Y6QULekpQIWqKu4ixhAweTH2+Ff30oHqp7JxAKggKh36bOIRCLI5XKor68vYMjh9+fzeYTDYUQikaLWLcczOdHBJWtLdauOxvHrS+V/m0RSdxeRs/MPCXlYGxoaZCxv6euWlZVhdHR0rpgB539IvxxfBArPLv/+h2Tp2V2qb06accnbH/pezurX1NQUnF319SORyDWd3WUZSovFgjvvvLNg/K29vV2IETiCxFltXmiVN5IzxupoEglN+b1kC1GJe4eGht5EeMrfE4lEMDw8jLvuukvIgzmuRIKDn//858v5qP/nQt2m02khTNi6dStaW1uFHow7gRKJhIxikUhkKXM29czLr46A8d+MRiPKy8tx7tw5eDyeN7Gacx72yJEjWLduHR544AEhFeAY5ObNm/G1r33t+irvbYRrj7kSo7S0FD09PWhsbJT9QblcTli11T0qSw0ndcvnwTlllbhBp9PJ8rAzZ87A6/WKU1L5KlOpFI4cOYK1a9fijjvuEDZzju/29PRgy5YtaG1tLeqpF7PZjNtuu002FhiNRvT09KCurk62WPLscrQWWNTtWwU/PNcq8bFK8cjAqK+vD7Ozs/JzKptQPB7Hc889hw0bNuDd7363sKVzJr+rqwvf/OY3r+kzLjv1psfg0D8H/1VCTXoO1ZgBi8YNgOw05nw3laPyG6pcdBaLpeDQqlGV0+nE/Pw8fve732HHjh0F3rm8vFy2uRW7cI6b+0Oom6URuMpPCaDgIJHOiyt/OWOsRpD0xpxJNpvNBd/DQzo/Pw+LxYJNmzbh2LFjuHLlijy/bDaLzs5O7N+///ooaxmikoGQgYqXifokIYvKOQDgTReV/J2cj1cdD4CC807dqmsM+HVyI65fvx7Hjh0Twg7qfcuWLdi2bVvRz9FTuByMO37olIHF88mv896r542s5eQbKCsrE75T6pc/p+qXc93A4rPi87ZarVi7di1eeeUVTE5OFrynrq4u7Ny5809Ds0ZZygxMVm71Q5Oj760utMqyTYOpehZ+j6pgABJ1Uuhxcrkctm/fDr/fj8cff1xIOR0OB+rr68WDFbPk83nZ46JeYDLR8CKqOqQel5LEqutqjUaj/Bt/nkKnVlJSIheaPwcspjhbtmyByWTCj3/8Y7D0smnTJrzzne9cETtdgAUqsKUXT2X5US8t9aUS8JJYgyQXKoXaW7Enqc+Ol199DnzNrVu3QqfT4bHHHhPd8hKTYKbYhSUGo9Eo+qXj5f1WhTpTdf1WHJNcfKfqV6UZzOfzQpZDSkb17Gq1WnR3d8NgMODnP/+5UNZt2rQJt9xyy7LO7rJSbx4wUkktrRmq9Gjc260aKaaBJpMJfr8f2WxWqI+2bNmC6upqnD17VpRHhbA+SXJg9fepUWptbS08Hg8+8pGPYHBwUJY5rYR1BeTIs9vtstFO5egEINEG/87SghoRcd/5zMwMAoGAkPy2tbXJUiyVdZsHigdO5Zdk5AQAO3bswNGjRzE4OIi9e/di9+7d4iBXgm5JmsvITXVEAApKRSwHASj4cywWQyqVwuzsLPx+PwwGA6qqqtDe3i5lJ/Xi8+cNBoNEOdQ3yZMBYPfu3XjhhRfg8/nQ3d2Nrq4uWbeiOrZilXw+D4vFAqvV+pZnVyWNppAVi/+m1S5sKUgmk/B4PIjFYkKtuHHjRuj1etEJjS+NK50dl4ep9Hb5fB5btmzBq6++ivHxcdx0003YtWsXIpFIwXN6O1nWU9BoNAWb1mjRl9YYVCWRMo1pZSaTwfT0tNRhuFDs9ddfx8GDB9HW1vamWhkPi1p3YzSqRqVdXV2YmprCyZMnUV5eLu/7Wpof11vy+Tyam5tl3SwvFY2VmqowMqJXpfOgI2Ikw0VsFy9exC9+8QsMDAzInh2VJ5HRPA0od+nQQ2s0Gqxbtw4VFRV4+eWXsXfv3gJnWexC3dL5skxEHkO1fsvsR6db2BzKzYA0pNS1Tqcr0O3Y2Jjs4VH1y+ehLsfj9/G12tvbYTabceTIEXR3d4sRWVofLVbJ5/NoaWkpKLstpZdbminys5OBnzuvNJqFfUFciTE5OYlDhw7hwoUL0oBjdM+7AixmriSj5uvw7JaXl+Po0aO45ZZbEIvFCt77tciyDaXD4Sh4AabZqvFSmZ+1Wi3q6upkc2BpaSmsVivsdrukifxder0ezz77LKampvDud79bohs+AC6W58J4Ko0HK5lM4v7774fdbpf3q+7bKGbR6/W45ZZbAKBgwyQdEQ+XumpWq13YfV5TUyMcitz/bbPZCmqRGo0Gr732Gn75y18KKz0NLgBJQ+rq6gCgwNFx0dk999yDgwcPSqeRz7jYL7Ner8euXbsALGYhdDhqo0vdy6TRaOByuVBTUyM6p95sNpuUlmh0X3vtNTz55JOyhI1RDtNSjUaD2traAvQGa+nJZBL33XcfHn/8cWHjXxrxFrPo9Xrs379fjORb1dfpPNRmltlsRkNDgxhIjUYDm80maA9gMfI8f/48Dh06BK1WC4fDUbD9k4avvr6+IKgir2UymcT+/ftx/PhxWbGsds6vRZbNcK6uI6CRWlpLVA2lXr+wE1pNNfjv9DrcE8NVuHNzc/jyl7+Mj370o+Ix+JrJZBIdHR2yyZGGEFhIn1wuFyorK2E0GuUQU+HFLPSibwdPUeu5AGCz2eRZEEqkNiu4Q50M6MPDw/j85z8PrVYrF55GMZlMYtOmTWhqaipI94EF/X34wx+GxWIpqBH5fD5Z21qswtqZepFVtIBq9CmM5tUd8zyvNILpdFruQzKZxMTEBL74xS+iqampoEHGy9rd3Y3Kyso3QbxSqRQ++tGPwmg0FqSUc3NzePnll6+X2q5ZTCaTwKDUxiPlD5WP7Ha7ZE4mk6kgGCCyghlVOBzG8PAwvvGNb0iTkd/H7920aRPq6upEtwwQMpkMvvzlL2N2dlbuCgBMT08XRJf/myw7ogQgF1C1ytz5ze9Tu4iMOLmFTQ3NtdqFjXhcVxmLxSTq/NKXvoT9+/cXQIx4wVevXi1hOx/Qhg0bUFVVBY1GIysm9Ho93G637OEpVikrKxNoiVrw59dUz8eDZrVaC/aIEDLEaFur1cJutyORSGB2dhb5fB6lpaWw2+34whe+AJvN9iYGbYPBgFtuuaWgk5jNZtHd3Y1169YhkUgU1KlPnz4tXrpYRc14GKWpEDTqVjWeVqtV0m31jKnniusFgsEgAAiD+SOPPIKGhgYxfPzdBoMBe/bskWgRWIhId+3ahfr6elmDwK742bNn8eqrr14HjS1PysrKxCipn1eNntWv5fN5lJeXS6TN/TXZbBaxWEzSaqfTCZ/Ph9nZ2YJ1MN/85jfl7PL35/N5lJWVYdu2bXJfeHa3bdsGi8Uie+gJAevr67tmJ78sQ6m2+5d6ZGIhuT9Eq9XC5XLJ5aWCuAsEWAzHbTYb3G53QSTEjYLf+c530NraKpElf9/GjRtlEVE+v0Dtz9QfWLjwmUwGbrcbwWCw6NNDNsPU/UFq7ZeXlQ6K9WE+k0QiITgxdcUv95CwHME0cn5+Hg8//DDKy8sL0sT5+Xl0dnZiw4YNACAH+aGHHpIGHQ/h9PQ0ent7l7Uf+XqJmnKrTlyNMNTLrmZJiUQCoVAI8/PzUmNkxMdNjgAKuuAf/OAHpYzBaJZRz5o1aySSKi0txZ133imGnOm4x+PBpUuXEIlEro/CliG832o/gfqmPmkkGeExyAKAQCAgC+vKyspEx2NjYxI0AZC6ciaTwWc/+1m4XK4CfCXP7tq1aws66Q8//LBkrVwJEQgEcOrUqWtOvZfd9WbuT6utegtGlDqdDmNjYwXpNpXJdJxbB7mQqKysTIrnuVwOVqtV0kFV8Xx9g8GAdevWIZ1OIxgM4t5778XFixcLPFkgEEAqlSr6aJKyNLpRoSjUdUlJCfx+P5xOpxwQOgymMBqNBqdOncLs7CzMZjO0Wq3oQIVuJZNJ/Pu//zu2bdsml5m627hxIx5//HHMzc3hF7/4hThDtYH0y1/+En6/v+idELCoW15ktZamRo6hUAj19fWyuyWdTiMej0vZKZvNoq+vD36/X7IAtc7O8kk6ncZjjz2GDRs2FETgGo0GnZ2dOHToECKRCH7729/KDhie70wmg0OHDsHr9a4IJ8RoGUBBBKmeXRrSVCqF1atXIx6Pi/Ngt5oY1f/+7/+GVqtFeXk5IpGI3Am1AZZIJPDTn/4Ua9askSicOtyyZQv+4R/+ARqNBseOHcP09HRB0yefz+Oxxx570zTU/ybLTr2XgmtZG1OxfSUlJQgGgwLuZZ1MjT4BIBgMyhY2KsDpdMJgMEj0UlJSgu9973s4cOBAweGemZnBqVOn0N/fjzNnzgjQlYdWo1lYcLScOsT1lkQiIRdarYctraNdunRJDGImkxH9mkwm5PN5nDp1CtlsFkajUX4PF7Op3V2LxYLf/OY3BfUcu92OI0eO4FOf+hQGBgYwPj6O06dPizdn9Hnw4EGZEir2RhmwqFvC1Xhu1Y6/Vruw4ZJ1W2YvRAdkMhlcuHBBomyWkqhbdtDZVDty5Ahqa2vFwdjtdhw+fBiPPvooRkZGMD4+jkuXLiGfz0tQodfr8cQTTyASicjPFbvk83mB26j6VZu1/BxHjx4tCJ4SiQSsVqs4ohdeeAHV1dUwmUyi+7KyMsFK8v4bjUYcPnxYnAywYDuef/55vP/970coFEIkEsHly5flebEU8r3vfQ9Go1Gi+muRZRlKNT3kIeL0DD0rw2p2VdWai1arRTwe53wl9Hq9wC1SqZRsIlRxUKyz/cd//IcUwhOJBF588UUpklssFkxOTgo+Ta/XY2RkBHNzcysGwqJOIQGQMStCqNTU0e/3S9kCgHRwgYW93ozGmXrYbDbpiut0OjgcDtm5XF5ejn/6p3+SC/6Tn/wEf/u3f4uZmRn5GTW6zeVymJiYwODgIMLh8DWv+7yeslS37IYyalSdK8fh2LxRcboejwcWi0XKR1zzywucy+XgcDikkeNyufCNb3xDpkwef/xxfOc738H09LRM9aiDFdlsFkNDQ3C73XI/VoITYsTI6IzGi+PMau/i6tWrEgQBkKDI4XDg8uXLqKmpQSQSQTqdljFQ6imXy8Fut8twRHl5OX72s59JRPutb30L3/72t5HNZmVnuNodTyQSuHr1KkKhEPx+v9yZa5Fln3I1hWY4q6aLjDKrq6vl8HHtqgoINRqNiEQiKC8vRyqVEsVxhSSNMiMYAlIjkQiOHj1aUM9LJBKyDlen00m4vpwF58UgvNBMBalftUtrNptlda3D4ZDLlMvl4Ha74XA44PV65fKyUM4IlE0KFr+BhZRmaGhIvHA2m5UsIZPJYOPGjfI1v9+PJ598UozF0m5xsYoKdWIaTCfKEobZbIbBYEA0GpWpEDoqr9cLm82GmZkZlJaWSgODjUhCWzgmyQxHq9VidHQUwWAQ0WhUoiStVot0Oo3W1lak02l5rs8880wBkcZKiCiBRVSKGkAxkuTZtVqtMJlMCIVC4ihYlhgcHERlZSWmpqZgtVplD3gkEpEd80zFDQaDjDTrdDp4vV4MDQ1JSZCRYz6fx8aNGxEOh1FaWopgMIgf/ehHAN68L/zt5I8KB2jIaCD5hnnw9Ho9QqEQSktLUVlZibm5OQALRjadTktNkp0u1h0ZDtOTqylOOBxGLBaTlF6r1QpEwGq1IpPJwGw2I5vNIhwOC8nDSrjEQOGcu6pXVR/A4gMeHx8XIHo2m4XP54PL5YLX65XGjcoKpNfrYbVaEQ6HodVqYbVaZVLH5/OJgaThMBgMWL9+PTweD1pbW+W9cIJEdULFXqOkA1dLGGoUxDNSWloKs9mMkZERtLe3Q6NZmPYIBoOwWCwIBoMoLy+XGiTrvYws2XhhKplKpRAIBKRGxijfYDBgzZo1KCsrk7/ncjk8//zzknLzPa2E8VueXQASwDCgUps6paWlcDqdGB4expo1awBA9NvQ0ICrV6/C6XQiHo9LFkWnYjQaBRHjdDrldaempgqiQ2Y+GzduFHvC+vuTTz6J0tJSQWksB9D/RxlKFaoDLEaU/Hs4HJYLzsMRCoWg1WoFLM5wOxaLIZ1OIxAIwOVyyfSOTqdDKpWC0WgURiB6b0YEfM37778fqVQKmUwG8XgcXq9X6lEryVgyiqawhqK+f3ptv9+PQCCAkpIShMNh2O12+P1+mM1maT7Y7XYYDAakUilhJXI4HGI4+DvUcT02vzo7O7Fq1Sr09PRIZBmPx3H69GkBRa+EkgZFxe+p8DZ13I6loVAohGAwKJ/ZarUiFArJiCcJR4xGI0KhEHQ6negWWLgf4XAYwWBQDJ2KVd28eTMaGxvxrne9SyKqTCaDy5cvI5lMimEBsGJ0rI4vA4vNMjVyY009FAohEAhAo1lgp6qqqsLk5CTsdrs47IqKCphMJkSjUSlB2e12OJ1OQVzQphiNRmE0SyaT2Lt3L6qqqrB7926kUimBC46NjSGZTBaM716rLPspLMVLASggvaDnYPowPT0tH0Jt/pSVlaGxsRENDQ0Ih8N4xzveAZ/Ph3A4jFQqhf7+fqTTaYRCIcEF5nI5mfdMp9OwWCwwm804ffq0vP7IyEhBdKRCFopZ1CaXOvVE3fIyGwwGaaIQnmK32xGLxeBwOCT12LBhA+644w5UVFSgo6MD27Ztw4EDB/Dwww+jqakJPp8PHo9HuuGpVEombrZt24Z77rkH0WgUH/rQh+S5vv7661LXU4kjit0Rqbrl31Xd0kgyPcvlcvB4PAAgjsdut0uzZf369bjjjjug0Wiwc+dObNq0Cfv27cP73vc+AIDf74fH45HXpG7n5+exfft2vPOd78Ts7CzuuOMOKS0dO3bsTbXoYtcrhfpVIzQVYqV2rf8/9t48uO36TAN/JFmWdd+XZflU7MSJHecgAZpAwlkCaUuBdpdCl9DO0G2Xbbvbdrvb2SnbabtHl92WltmhMN0OQ1tggAZ6hDNAQpOYXHbixLctW75k674PS/r94X1ff2TaH3Fn+CHPL+9MBpI4tvTR5/sez/u8zxsMBpHNZhEMBrl5OD8/D4fDwX6lo6MD119/PTQaDTZu3Iirr74a1113Hb785S+jrq4Ofr8fc3Nz/PPS6TRyuRyKxSK2bt2KT37yk5icnMRXv/pVpiweOXLkPSpDqznjVWeUFJUpW6RfwHKZYDKZMDw8zEIMyWQSKpWKQVylUsm4j1arRVdXFwqFAm644QZ+0H/3u98hFArxB0H/LRQK6OjoYLxiamoKV199NYO/lKqL2cNacZQrSwGRfiXST+icZTIZtFotpqamoNfr+Qx0Oh1cLhcMBgOuvvpqbqo1NTXh3LlzGBkZYRKviIU2NTXhwIEDyGazeOedd3DVVVcxFlQoFHDq1CkkEgnuCFPgq3RbeYb0fkSIg5wnlWUymQxqtRozMzPQ6XSMl2s0GjgcDmi1WuzZswdmsxnZbBZWqxX9/f1IJBIIBALMVqCzbW5uxqc//WmkUikcO3YM+/bt4+BfKpWY/ExNC3HwoNJtZdVG71t0mlTFEN9aIlkS9x0bG4PVauUR2rq6OrhcLjgcDk6E0uk03G43zp49i9HRUYTDYWg0mrKf4XK58M1vfhPj4+P42c9+hvvvvx/j4+P8ORw7dgyJRIK5qqtNoFbNoxRLFToEckzUCcxkMohGo5ifn+dpDgAsCuD1etHZ2cndccIYKLr09fWhVCrBaDSyU6CZW4fDgbNnzyIejyMSieCjH/0ovz4SSaULRtjEWnmYRZwSAGO+4vuJx+Po7OzE+Pg4GhsbEQwGYTKZGERvaGhAe3s7tFotXC4XfD4fZmdn0djYiN/+9rd4/vnnkc/nYTKZUCwWEY/HsXnzZtx2220Ih8N48cUXkUql8I1vfANdXV3cWDp16hSXo8SFJcC80k1sOq08W/GOxONxNDY2wuv1or6+HsFgEAaDgbmqtbW12LBhAwwGA5xOJ6anpzE7OwubzYYjR47g0KFDWFxchMlkQqFQQDKZxNatW7Fnzx4kk0n89re/RTKZxNe//nVs2LCBYZTjx48zHCXi82slyJPRvfhjDpLgCJPJhNHRUZjNZkxPT6Ouro7hNo/Hg66uLqjVahiNRoRCIZw9exadnZ144YUXcPDgQRiNRoY44vE4PB4PHnzwQRw/fhxPPPEEIpEIHn/8cX5m6HxJIIYydnHi6lJs1Z8CeWR6gERmOzUWUqkUvvjFL8Lr9SKZTHIUUSqVsFgsaG1tRW1tLUqlEsxmM3dijx49iqGhIaYQEGWiqqoKNpsNTqcTw8PDrO9388038yVeXFxkcjWZOM2zFkyctFnJP6MsLplM4p577kEoFEImk+EGlkQiQUtLC9ra2mC1WpHL5ZDNZmEymVBXV4cjR46gu7sbmUyG6SqE06XTaej1ehQKBVgsFvz7v/87S31Rw2dgYICjvpj9VvoMPZmoRUkOSQxAi4uLiMfj+NKXvsT3T6fTsWP1eDxoa2uD2+3mxqJWq4XFYsG7776LU6dO8fQH8TMlEgmSySQsFgvro37ve9/Dhg0bkM/nOXscHR1lvA1Y7iCLWHylG3FsqYFDY83kNOnufvGLX0Q4HIZarYbNZuM7tW7dOg5CU1NTCIfDKJVKcLvdeOONN9Dd3c3rJajBCyx9doFAgCG9l156iUdWCZ/v6+tDMBgsG9xYSRl7P1u1oyTisTgKJj7MABgvCAaDmJiYgFqtRqFQgMlkQiaTQW1tLcLhMOx2Ow++u1wu2O12+P1+VhqiN5VMJjE2Nsaz3Lt27cL27dtZXUgmk8Hn8yEej5d141e+7ko3cWqGHCOVMeQ4yQHGYjG4XC4OTrW1tdiyZQva2toQi8VgNBrh8/lgNBpx9uxZ9PX1wev1wuFwcNkRiUQQjUYxODgIjUaDnTt3YsOGDfB4PGUrEaampnieGVhW8V4L2SSwnNkQRiUSo+lsATCFbXZ2lpsGuVwOdrsdHo8Hra2tCAQCsFgs3HwcGRmB3+/H7Ows7HY7nw1RrQYHB2E2m9He3o6NGzeisbERmUyGWQMTExPsYOm1roQK1oKt7AeQsxR/EaVscnISbreb/8zpdGL79u1wu92Yn5+Hy+VCMpmE0+nEmTNn0N/fj0gkAqfTyc/x/Pw8kskk+vv7YbFYsGXLFmzfvh3JZBLpdJqr2OnpadawFNk5qz3jP6uZI/4A6lBTVAaWRuOSySQSiQTq6upQKpXQ0NAAg8HAo3cqlQrBYJAnO06ePMkdRDpwcf8ONS6amprKJkJEsm4sFoNGo3nP1MpqDuTDMuqKiqUhXTY6A7pY8XgcGzZsYEZAV1cXrr/+eh4jo4BTKBRw8OBBPPfcc1hYWOAJEK1Wy/tLMpkMFhYWMDw8DI/Hg1wux6wF6g6fPHkSs7OzkMlkMBqN0Gg0TG1ZC7AGUP75iyRz8Zzz+TxisRhaW1t5Lrirqwt79+5lXqTJZMLs7CwmJyfx61//Gs899xzm5+dZQZ1oRHS2lCxs3ryZnwkiQufzeZw4cQKTk5OQy+UwGAy8xgPAmjlbYDkLJv9A50vEc+L6RqNRtLS0MNWvq6sL1113HWfgZrMZU1NT8Pv9+OlPf4oXX3yxDEcnoQxgiUBO57tu3Tpks1leKEbsmjfeeAMTExN8vkqlkv0DQQKXYn82AEI/QOZScHEAACAASURBVC6XI5vNll286upq5HI5PPDAAygUCrDb7WhubobBYOCsKJ/Ps9L5xYsXMTIyglQqhfn5eZZPo8OjfzMzMwOVSoXx8XGmrJRKJe6sA0tOOpVKoaamBlarFXq9nqWwKt0oMxfJunTx6LJIJBKkUins378f8XgcdXV12Lx5M79/+nur1YpSqcQzxX6/H+3t7SgWi4hEIixySmpOPT09LMVGD7lCoUAoFMLU1BQSiQQymQzC4TDDHBaLhelGlWyUUYpZsKgwIz6ImUwGn//85znIb9q0iRsrhB+SHuLrr7+ORCKB6elpdHR0MOZLI4/ULDtz5gyqq6tZxSqVSqG6upoHByjDjEQiKBaLUKlUfHfFyZJKNZEqJlaZIuGcntV0Oo377rsP8/PzaG5uxsaNGxmfpSYP9SaOHDnCVKBt27YBACvL0+ReoVDA+fPnoVKp4Pf7kc1my1SC6Pd0vjQ4YbFYykTI38/+rNJbzHxE8JkOixznjh07uEwcGRlhHqTFYkEgEEAgEIDP50MikUBjYyPm5+fR3t6Oa665BhcvXuTvp9PpkE6nEQqFYLFYOAITodTv9yMajfJrJA5mKpXiSL8WojM1GCj6UvAhEq1INr/llltQLBbR2trKQiKkL0k8x+eff545pZs3b0ZTUxPuvvtutLa28kNos9lw1113obOzsyz7pjJ1YGCAJ6po/wuNkRLHrdIbDmJXGyjnVFKHmc7d5/Nhz549iMViaGxsRCwWQ7FYhNFo5IdRq9XilVdewfz8PKqqqtDW1oaGhgbceeedcLlcTDi32WzYv38/GhsbuXEDLMuS9fT0MK+PsDc6W6K40R6dSjdqkgDLXW+qhCjQSCQSDA8P48Ybb0QqleLAbTAYoNVqIZVKEQwGYbVa8bvf/Y6pWrt374bBYMAdd9yBrq4uHo82mUy49957sWPHDg7kuVyOdVm7u7u5P0L+guAUIvZf6t39s+hBIpGbIgE95Pl8nkeRDh48iNraWiSTSZhMJoRCIbS0tKCvrw/t7e3o7e2Fz+eDw+GATqdDR0cHfD4fFhcXsXfvXvj9fo6+d955J9RqNWMO5KxXKoFTB5MyByKhVjooThxJMRKTg6RLJpFI0N/fDwA4ceIE9u/fj7m5OWQyGTidTtTU1HAZ8sgjj+Ds2bNQqVT4p3/6J9TX12NoaAgWiwX19fXIZDIYGxvD17/+dTidTqxfv54vDjUeJJIlAY5oNFrWaFrZma10I+yXup3U6RbX8kqlUkxPTzMN6tZbb0UymYTP52O+7+DgIDo6OvDjH/+Ycd2vf/3rTA1yu91oaWlBsVjEyMgIvvGNb0ClUmHjxo1IpVL8nFDT8eLFi5zd09mKnzXdgUo3iUTCugDi5BM5KGJunDlzBmq1GhcvXsTnPvc5jIyMYG5uDk1NTbDb7Zibm8OmTZvw3e9+F8PDw9Dr9Xj00UdRVVWFoaEhGAwGDA4OwmKxYGhoCD/60Y+g1WqZEQMs0edIXPr8+fPw+/1ld5dM1Ee4FFu1ehB1nkQqA/2X5mNHRkZgMBhYj4+2IlosFgwMDKCxsREDAwMIhUI87UAXk2gVNKbkcrlwyy23IJfLQaVScdQVS3MitAPL5SuBy6QRWOlZDwBW9xGFBKh8kMlksNlsGBgYwG233Ybt27eXKW87HA6EQiE0NTVhaGgI3d3dCIVC2LVrF4rFIt5991386le/wu9+9zuo1WoW7NXr9Xj55ZfZcdTU1HAWKpFIsLCwwB1a8QGmeVuaza90I1yKAr1UKuUMRKFQwGAw4A9/+ANuueUWNDU1wWg0cmZis9kwPT2NlpYW9PT0oLe3F8FgEFdddRU71oMHD+Lo0aNQq9Usv6ZSqXD48GEuwwk7pgwrl8uVlad0tvSwq1SqNaNXQBgusVAA8PlKpUvrYLxeL26//XZs2LChbF+R2WxGPB6H3W7HxYsXMTQ0hEAggDvuuAMXL17EsWPH8N///d84duwYT0WRGPXjjz/O1aLVasXc3ByfI1ULYjVJ04FqtRpqtfqDaeasbKsTKE50IZ/Ph0wmgx07djCNAljCFc6cOYNisQiTyYRIJIKxsTHu4MrlcvzqV79COByGTCbD3Nwc0uk0ampqWKmGcEbKGokicP78eSZBi9wzunCZTAbpdLriNSkpoyRMSlS3SafTGBsbw4ULF/Bf//VfqKurw+joKNRqNYLBIMbHxxkIz+VyeOaZZ7CwsIDPf/7z8Pv9ePLJJ/HOO+9AqVRieHiYtzPGYjG88soraGtrQ39/PxYXF1kkVSqVwuv1lgmWUIZQU1PDlK633nqr4ldBAGDSPbC8g5rUpwYGBjA+Po5HH30UFouF1X2mp6cxNTXFD3OpVMJvfvMbzMzM4J577sH09DSefPJJnDp1CiqVCr29vUyGjsfjOHz4MNrb23Hu3DkA4PKdhDKIYE5BnZyHVqtFJpPBsWPHcOrUqQ/z2C7JiFEgYuzUSE2n0xgfH8fZs2fxxBNPwGAw4OjRowCWZBanp6cZL5dKpXjqqacwNzeH7373uzhz5gxeeOEFHD9+HHV1ddzwpeTq5MmTuPLKK1lW0Gg0Qq1WM4RCbAIxaVIqlTxE8dprr30wqyBEEQAqvQgkJ94SdQB9Ph9MJhOsVivS6TQmJibQ2NgIqVSK+fl5HkekblUoFGKNPhLL0Gq1SCaTmJ2dxfT0NMLhMDcZzGYzR2Wxe0WllUKhQDQaRSqV4i5ZJZtMJuPtfCLALJPJmOpAuGJPTw9cLhfUajU0Gg2sVisveH/77bdx5MgRfOUrX+FoGYlEcOHCBVZYicfjTNlaWFjA2NgY/H4/8vk8L32TSqUYGhpiyhW9JsLSMpkMjh8/XoYNV6rRlM0fO1tav0B495kzZ+ByuVBTUwOz2Qyn08krR1588UWcOHEC//AP/8BBOxKJYGBgAF6vF6VSCalUCmq1mifPfD4fPxNGoxE6nQ6lUgk9PT2szr/ybNPpNK/YWAvCvTKZjCtNkdhPAxLpdBrNzc2IRqPo7e3Fxo0bWcW8vb0do6OjkMlkePrpp3HixAn8+Mc/xtDQECQSCSKRCIaGhjAxMcHPAnGHI5EIU9ey2Sx0Oh3v7j5x4gSr0gPLpTbJur355ptlkz3vZ6suvUUhWWBZey6VSsHj8WDbtm3cvbNaraiqqoLH48GOHTvw/PPPc/m9bds27Ny5kxVBampqmGcllUqZqE5OYnFxkUFwjUbDl4tSfrHdTxlQoVDgveGVblQSiHPH1O2XSqVobGyEXC7nBorVagUAtLW1wel04rXXXoNCocA777yDHTt2oLOzk/FjwsToZ1D0J2FlksACALPZzN3D0dFR/vmiKLNCocC5c+dYzKHSz5cCJwDWIiCIp1QqYcOGDVAoFEgmk0gmkxyEXS4XnE4nXnjhBWi1Wly4cAE33XQTWlpa+D6Kcl+kagUsE9wp0ymVSrBarRgeHkY2m8X09DTfdfF8FQoFent7mfmxFgIRVUPi3aVsvaamBm1tbdyYyefzqK2thUwmg8fjgUSypMavUqnQ39+PG264oUyrkspj8hF0xhT0iJ8tlUpht9u5UqXARdWlKPh79OhRZutc6t39syZzRIEL2mFRKBQ4emg0GsRiMe6ehkIh3l08MDBQprIiju1RV4oUprVaLSQSCeNpFG1pvGx+fp4dqTjbXVVVBb/fzw/CygmiSjXCeEUWgUwmQzKZRG1tLevzpdNpxg1HR0cxNDSEa665BqOjo5iZmcFDDz3EorqkEG21WqFWq2EwGBh4NxgMXPI0NjYilUqhrq6OaSzBYJA/CwqMCoUCfr+faVyJRGJNYJTi2ZJjoskNaoSFQqGykVev14uJiQncdtttOHfuHHw+H/76r/+a1bwpaNhsNqhUKpauUygUfN7kEIjZMT4+zsR2CuyUVRKFKBqNcpJAeF+lG90jYFnqjNgnHo8Her0eExMTSCaTHKyHhoZw+vRpfPKTn8T4+Dh6enrw8MMPI5VK8fkqlUo4HA6oVKqyvVhGo5GbMa2trYjFYnC73ZicnCxjFxAuDSz5LhKCoUmhS7VVd72pPCNOEnltiqzkrGhp2PHjx2E0GhGNRrFjxw788Ic/RFdXFxwOB7xeb9kKB4oQJOyg0WhQW1sLqVQKm83GkXvdunXo7+9HKpXiQwWWGzkSiYSB5D81qVOJRhxQ6s4SEdpsNiOfz8NisSCXy0GpVGJychKRSAQej4dH6R588EF8+9vfhlS6JHJM2TXxzegiU1PM4/HwRWxubsbs7Cxqa2sxOjqKCxcu8MZB8Rd1a0ulEgfDSrdSqcT3gR5eOl+Cd5xOJ1KpFEwmE3p6emAwGNDe3s6ir//6r/+Kb3/725BIJMy8oPUF4s/RarVQKpVoampivp7FYsH09DSsVismJyfR09ODaDTKzoR+yeVyXLhwgafRVlMafthGG1RJ7LhQKHAGmEwmYbPZmIx/+PBhNDQ0oKurC06nE9lsFt/85jfx0ksvMdkcWF41IyY7FNxbWlqYamiz2TA+Po7a2lr4fD6cPHkSiUSiLJskfP306dMAsOq7u2pRDIp0IiZIIrwymYw7exKJBAcPHsSOHTsYjxwcHMStt97KGaU4JldVtbwjPBwOs0gBrWSlHcCFQgEajQahUIgzTpGqRKRpumRr6bIVCgWWn6L3o1KpkEwmodPpGFtUq9V49913sWvXLkilUhw6dAgAcOWVV6K+vh65XI5JvBS0iMQejUaRy+UwNjaGqakpuN1ubNmyhZtGVEZR+SfSkwDwQ0yiwGvlbAEgGo3y+yDSvNfrhUajYQ3P6upqnD17Ftdccw2KxSJ+//vfQ6lU4tprr4XdbmeYiTqqVGaKsMjw8DDm5+fR0tKCq6++mrVU6W729fXx6C5QPrZIU23iTPJasFKpxE1VYFlfgRR7JicnYbFYUF1djcnJSWi1WhiNRjz33HNQq9XYtGkTfy0FMmocAkvPBkkw9vX1IRqNwuPxYM+ePXxmBCN1d3czNEIVBJ1voVDg17mas12VoxQbN+Lv8/k8IpEIUqkUVCoVcrkccyJlMhkmJiZgNptRKBSQyWSYLpRKpbjrXSqVOOIUi0uK5j09PZDL5aivrwcAFuelDEHcGUNlPr0e+rq1YqKsFj0g1ByQyWQIh8NMyq+rq+MR0Ndeew3j4+O48cYbsXv37jLNTjofwmZo3wsxFZqbm2GxWLBhwwZmBVAmQHvA6TOmbJKmK+jrxbnZSjXxXOm/pVKpbBNoY2MjZmZm0NTUBJ1Oh1AohEOHDiEUCmHbtm244YYbkM/nWVhXhEeoqSbuiHG73dBoNHA6nfxvKKsnlXkyUcWI7jiwPI1V6SZWdCvvrl6vRywWQ2dnJ3w+H+rq6nga7Cc/+QkmJyexf/9+PPjgg9x8pek+6jsQrEHnr1KpYLfbodFo0NjYiFAoxJknEd3FSSxKPKgbTwpYq0miVu0oiVZB2R+td6DybWxsDNPT05ibm4PH40EkEsGRI0fQ2toKt9vN88SkzpJOp/lwc7kctm7dypHf4/GUMfdplwmVqMSNEz8kYHkFApHS10LmIwYhUQSVQHG73c4NFiqdo9Eoenp6sH37dmzcuJH3D9HYaDqdRiwW40mEtrY25PN5hMNh1NXVobGxEbt27eKMkYSSq6qq+HMRifvEPRRXiK6FB1mcdiIHT5BRsVhkZW06W51Oh2QyiXPnzmH37t3YuXMnTCYT7053Op2ctdNd3LBhA1KpFAe0xsZGXH/99dypLZVK8Pv9HMjFbIbuKWVGZCJcUskmVkBUIVIQCQaDuOKKK5BOp3Hx4kXU1taiqqoKkUgEg4ODuPvuu3HNNdew80qn03A4HDx6SHe3oaGBmQQKhQINDQ342Mc+hmQyyfTEubk5Ho8Wje4o7WcnjuUHmlHG43FoNBreg0PNGrfbzSXiwsICDAYDotEo5ubm0NjYCJ1Ox9pz5BSrq6uZKE0ef3x8nLtdtJnRYDBwVKYLGo/HeTfMSgoAdbrpe65m+P3DMpqIIQI3OaZEIoGuri6GIbxeL7RaLfR6PQeUnTt3ora2lqk8i4uL0Ov1aGpqQiKRYEGNycnJsm12VVVV2L59Ozd+RFiFJPTp39J2S9ohTlkEdTsr2eje0tlSxjE/P49NmzZxJnju3DlWsaKx0KuuugoWi4XFLEqlEi91o6AELDV+SIKQxg49Hg9PNQEou8OihCCxSej819LZAkt3l2hRwLL+5/T0NDNbMpkMhoeH+XmkZ3L79u2Qy+UcdGQyGZ8v8awJkqKOen19PZRKJZqbm3n2WyaTlWXiALjhSb8o6SBGiVKpvGQO8KodJV0McZkPLQSqq6tDKBSC0WhEb28v9u7dy05r06ZNAJYBWlpdkM/nEQqFGCOiksjv90Ov16O5uZlFNgDwg04RQUz3qTykWWe6iGsh65FIJNylp64ssDRtYLPZeAqnvr6esdp0Oo3rrrsOmzdv5vJYHCmsrq5GR0cHlyyJRAI+n4/3wOzfv5+dBDlKwncpwyVAnLq5RPMQpx0q/WEmPh+wTHEDlqa+LBYLq+W73W7eyRSJRJgKRF1cGtUjBZpSqcQCIslkEpOTk8jlctBqtbj77ru5jKb7S4wRMrEaqqqq4uRDPM/VjNl9WCaRSBAKhfju0XNcV1cHq9UKjUbDmaJer0c2m0U0GsW+ffs4qJCiGA04qFQqVFdX48yZM6wK5Pf7uZl533338b4ttVrNiZe4Z0ok89MAAb1e8bVfiq2aR0lbzMQmynXXXYf169cjEAhAKpViYmIC999/P8bGxuDz+SCRSGCz2ZBIJDgjJGqGTCbDxo0bsWvXLqxfv54FBejndXV1cXQnWgwJQNCIF2VN9HqIsE5Ry2azVfyFo84+lSyUHfv9fh7nrK6uRjKZRFtbG6sA0QNL2E6xWOQxT8J3HA4HrFYrE9opUyG8kbIbWkBWKpXQ1NSE+vp6Vsqh1zMxMcG/r6mpwa233orGxsYP9/DexyQSCRwOBzs8ykwymQxXQlqtFvF4HFu3boVEIkEwGGTGAO2apyyEZMOoSjKbzVCr1TyMAYDxeArchMkDQH19PWpra5mkTa9ncHCQm2cKhQL79+/Ho48++mEe3SWZRCJBXV1dGTNCoVBgZGSERVS0Wi07SzpfYrhQc0WcuCMHZ7FYyqT9KFBRaU4shlwux022lpYWNDQ08JZW8lsDAwNMGVIqlfjoRz+K1tbWS3qPq3KUVVVVLLhJGUw+n8fg4CAikQgTPVtbW7nL5fF4YDQa8frrr3NkECd75HI5zp07hzNnzmBkZIQxmurqavh8PvzzP/8zFhcXcffddzPWQ1JJNTU1WLduHdrb22GxWAAsXVBSPCYqUVdXV8WrsNAeY7fbza89EonAbrfDbDYzBtPR0YGJiQkUCku7gxYXF/Haa6/xWF5VVVUZn3R0dBT9/f0Ih8NccptMJkxMTOALX/gCvvOd7yASifDXS6VSDkxarRZtbW3YsmUL1q1bB5vNhkgkAqVSCblcjiuuuALf+ta3Kj4I0SI6p9PJQSgQCHDJlslkMD4+jptvvhmTk5OIRqPo7OyERqPByy+/zCUxZUokptHf34/+/n5Eo1GGOwwGA8bHx/H5z38eP/rRjzgbpz3rxPTQ6XTYsGEDqzqZzWaEQiHmuO7cuRNf+tKXOEhWslE253Q62VlOTk5i8+bNjDdOTExgz549PGnT3t7O8/V0d6lrTfdwdHQUIyMjnAgZDAYYDAYMDw/jwIEDeOihhxAOh9kXSCRLcm40GtrW1obNmzejtbUVdrud94lXV1ejq6sL3//+9y+Zp7oqR0k/yGw2s6ObmZmBXq+HUqlkvMBgMGB0dJQFCKgEpo4XAM4mSU6KCMF0EQkPkkqleOGFF/Cxj30MW7duBQBoNBrGF/x+P/9cEjNobm6GRCJBa2srXC4Xg/aVbIFAACdOnGC+KACGJ4rFIis3U4bY1dXF2TsRaQk3S6VSjBFR5kcBhtbbVlUt7fl+4403cMMNN+DHP/4xk6QLhQL8fj9mZmbg9XqxsLDAOBJF7y1btuB73/seT0JUsgWDQfT29rJGJ7CEY7lcLqa1iJNH27dv5weVloURbkuY1tjYGFOpqDtNFCEStXj11Vdx3XXX4amnnoJOp2PHHAwG+Wzp7tLgRD6fR1dXF/7+7/8ek5OTmJmZ+dDO7VItGAzi9OnTrDcLLHF1W1paeBV1NBqFVqtlabR8Ps8BP51Oc0eaEiWv1wtgecCFGluRSAQKhQJ6vR5nz57Fpz71KTz22GM8vqhQKBAIBLCwsACv18vsjVAoxBtdOzs78R//8R9MI7oUW5WjTKfTuHDhAjs/KmUUCkXZXm6j0YhNmzbBZrPBaDQiEAhg79696O3t5bScSuO5uTkAy6OQFFWIB6VSqbgU/853voN7772XM0oArPpNasfhcJjLSavVytJvlY5TSqVSvPXWW/x6qTNKMnV1dXXQaDRoamqC0+lEdXU19Ho9dDodZDIZxsbGeOKBMsqhoSEmVBP5nhTpNRoNdDodjzH+5Cc/YQkr+lrC4GKxGAKBAILBIKvffPzjH8fExAT8fn/FB6FSqYQjR47w2dLDUVdXh2QyCbfbDblcjpaWFnR2dkKhUHC5R5srCSPOZrNQKBS8u4m4k3S2EomE6ULE43vkkUcQjUZhsVjK9sgQyyAQCPCDLJVKcdddd2F2drZMkLqSTSqV4t133+VGFXWeqXfR0NAAm80Gt9sNq9WKRCIBk8kErVaLYrGIw4cPcyZNVREJ4SSTST4D8g9qtRpKpZIbjD/72c8wNjYGi8VShjlSYKOzpUThs5/9LHp6euDz+S79Pa7mQIjPR3w6EjSlmp9W0tKDGQqFGDcjqgA1D2hbI7AsgUWZn1g6ikC6TCbD/fffz9M8ZNTRoqZQqVRiiTf6eZXecCA1cZptJayRtPXUajVmZ2eRSCRYZYgeLIVCgampqTKiLpUzYpeXaBGEUxIWRqX+F7/4Rea70nnRREM8Hscrr7yCXC6HhoYGOJ3OMoy0kk2hUCAYDJadLXFLgaX7Nzo6imAwyEK6CwsLDBMNDAwwG4BWDIgkZgB8tqQaL0r8yWQy3HPPPbBarWXNNirJU6kU3nrrLRSLRTQ0NECtViMajUKj0ayJ5W1VVVUIBoN8F+iO0R0kfJAwW8IlA4EAXC4XRkdHOZOsqalhhyjyIImKSHeXlrgRlPK1r32NAxEZfV0ul8NvfvMbAEvBkapRq9V6yQnUqh0lTYuI0Zk4lRKJhCc/aKyOVlL+8pe/RFdXF9xuN8sfBYNBdowUlUWZNHq4qQSXSCQIh8Nwu91c7tG/Ica/1+uFRCJhOIDoLZX+MFPXmT5oOgMSASHBkFKpxM0Do9GIfD6PgYEBGI1GZLNZznjGxsYQDAZZgYY+I3qQaY6Wfg4xEYLBIF9Usak0MjKCyclJSCQS1rgkdkGlG50t0VfofREvlzqtpVIJNpuNN4MqlUr09PSgs7OT2RgqlQperxeRSITHDKkspKBCTlI8W9K9JMgJWG6O9vf3Y2ZmBhKJBLt372aohRKMSjdKVGiKhu5wKpWCTqdDPp+HwWBAKpVCbW0tDAYDNBoNFhcXcf78eXR2djIVSKFQ8O5u2t1EnWzCIqn5RXhoVVUVEokEr1MWNxrI5XL09vbyOO6+ffuQSqWYf/mBdL1JoIE6UwROU6eZAGpKjanhIJPJ0NDQgIaGBgSDQczOznLzh5yZyB0krIiiMV04sVsp4pn0b8n52u12FtCgS1vpDzRx8Oiho+kjKulyuRxaW1v5XDOZDBwOB0wmExobG3HHHXfwRFMul+Pl77Ozs7wVkygTdCYKhYKXWRGE8tRTT3GnkDqRhDHrdDo4nU6mzDgcjjWxCqK6upp5oeI0GZHsi8Ui31uiupCQg9VqxSc+8Qmebc9ms/B6vZDJZFhYWOCsnX4ONW+oQ0uZe1VVFZ5//nkWj6AgSPQhlUoFp9PJy/jcbjd3gSvdaFEasLyLiOAxwh2bmpqg1+tRLBaxsLAAk8mE5uZm2Gw2fOpTnwIALrGHh4chlS7JMdL3Jb9AKk0k9kLZu1wux/Hjx1kZixIwKuXp7prNZkgkEjQ3N3OGeym2akcpzg6TGAY5Iipv5HI5rzel7Yg1NTU4f/48ZDIZ6urq+CE0mUxlCkTixITYMRTHkF5++WV2mHNzc4hGo0gmk2hubobZbIbJZGJhDeoCV7qjJHhB3OFSU1ODRCLBTRwC/akzODExgfb2djgcDoyOjsJoNKK+vh6zs7M8C0uXjTJ/ytSphFEqlVwiAsDTTz+Na6+9Fps3b4bJZIJEIkEgEEBdXR0MBgPWrVvHeo30Giv9bIllIQZcmimm6oZUyUlJv6+vDxs3boTdbsfMzAyPy5EqVUdHB0KhEJ8rqabT9yaRWHHdxMMPP4xdu3ahtbWVye+BQAANDQ2orq7G+vXr4XA4mA+4Fs4WWD7flepM1DsgpZ9UKoV8Pg+TyYSpqSl0dHRAo9EgkUjAYDDAbrdjfHwcmUwGmzdv5opTHDuk702BnoITADz88MO4+eab0dHRwXjlwsICPB4PCoUCNmzYALvdDmDJKdPAyqXYqhwlbY4jflKxWIROp2PNPPLkiUQCWq0WKpUKr776Knbt2oXOzk643W7U1NQwoddgMGBmZoYPgUwki4rpNVlfXx/v8b3xxhvR2tqKU6dOwWAwMLNfpVLxPhK6xJVsRDgnJ1gqlbjppdVquTwkR6pWq1k9Sa/XsxgqTUBoNBpurtE5UuZDpTdlPvRAU4ldXV2N2tpa3HTTTThw4ADq6+t5v8knP/lJVnQi7qX42VSi0WQOZdOEe+fzeYY1CDOnjIUCUl1dHdxuN68GHhoaglqtZkew8mwJOxOzSvq5tOO+oaEBN910Ez7zmc/AbDbD6/UiHA7j9ttvh8lk4mQhk8kw7a2STSKRUPH8QwAAIABJREFUsLixeL6lUgk6nQ6RSIS1ZEV6lc/ng8vl4k2pyWSSz5cU5YHy9S4U1CkLp7tcVVWF8fFx3vd07bXX4sCBA2hra8PFixcxPz+PT3/607BYLEwVJLbDpdiqvAdlfZRSE24gzk4S+Eo7Vfbt24c33niDNfyopKTmBYG3FHUJq6TIJGI/IhDucDgwMzODSCSCpqYmbNq0CePj4zAYDDCZTDxKJnbIK9loKiccDgMAl11EqhfHsKh03LFjB+vv0fkMDQ0hn89j69atOHHiRFlHWtzFQ1gl7Q+h7rdEIsHs7Cz8fj/8fj+qqqpYxmrXrl246qqrGEPT6/VrAqekAETamcQnpddPZ0uYeLFYxLXXXov+/v6yiafR0VFIpVJ0dHTg7NmzZYILBFFQtk4Zj3i2FKxnZ2cxPz+P6upquFwueL1e7NmzB+vXr2caEk2bVHoQAsAi23S+JFZsMBh4DJkgMkoIOjs7GSekZ7y/vx8SiQT79u3Diy++yH4AWF6NAoCdpVwuZ3YB3d18Ps9atcQcCYfDuOmmm7B+/Xrm0FJH/gPBKEmKS1RhIUoAlcoUOTKZDKanpzE2NgalUgm32w0ADOyazWYGYenBF0UhREcp4pMUqcbGxgAA8XgcY2Nj3CS69tpry5wtOe1KN4IxiFVA3DHqetMlKRQKuHjxIgKBAObm5lBVVQWHw4GqqireWKnT6XD48GHW5ANQdrbU1SZckoIaNXcOHz4Mg8GAcDiMEydOIJvN4vbbb8ff/M3fYGFhgVV25ubmWD29ko261aTRSU6TYBkA/OD19fXB6/UyDmm32yGXy5FMJhmqOHr0KHe+AXBDhzBmgqdW4sBSqRRPPvkkqxN1d3ejVCph3759OHDgAMLhMObn59H4f0pGGo2GG1CVbDKZDFqttowRQ3eRzoTwxYsXL8Lv92NqaoqDLfU+CD986qmn+OvJ11AiRhi+eL5UFUmlUrz88sswGAyIRCJMWfrMZz6DBx54AH6/HxMTE9i0aRNvefxAMEoSgBVpANRcIMyHLs3x48eZPEvE3mQyiXg8jsHBQQwODjJ1hw6bDgcAg+SESRBoS7/eeOMNnsul0n/Tpk1MMzCZTDy3vBZEMWpqahCJRJhoC4BLbiLlkrT+1NQUSqUS7HY7Ojo6uIHT19eHU6dO8boBcgIUaMgxiKLGRBFSKpVM2P3DH/7AVJr5+Xnmc1IG73a74ff7Gf+pdEdZU1ODQCDAs77A0gw9ZULhcJjL6NnZWRiNRl5r4vf7mT985swZViine0vZEFVVopADld8qlYpHHI8fP86K/aSo7nA4sLCwgFgsBpfLxSrnLS0ta6IaUigUmJ2d5fULAOB0OhGPx3kAguhugUAANTU1cLvd6OjoQDweRyaTwZkzZ3Ds2DEEAoEy1SoReiMtArrbVIIrlUqeuKElekQyJwhvenoafr8fzc3N8Pl8iEajaG9vv+S7K1kNbUYikSwAmFj1SVaGNZRKJeuH/SL+lF0+2w/O1vjZApfP94O0SzrbVTnKy3bZLttl+/+jVXbNdNku22W7bBVglx3lZbtsl+2yvY9ddpSX7bJdtsv2PnbZUV62y3bZLtv72GVHedku22W7bO9jlx3lZbtsl+2yvY9ddpSX7bJdtsv2PraqQVKVSlXS6/U83wrgA1fmEYU7/5SSNo060dIy8d/+3+vG7OwsUqlUxY7n1NTUlDQaTdk+cqVSyZNPJA9Gs97i9MKl2srPiaZ1xEkd+jr62lKpxCtALBYLT6TQz6dlc8FgsGLPVqlUlkgpht4jTWyIdqm6pXQn/5SJ5yfujxYX8gFgrVSaghLPn9TldTodxsfHA5VMOKe7S/eWxobF90v3euX5iuchfu3KsxKNfoYok0d/Ln4N7XEnERNx9Pr/XjetqXjfu7sqR2kwGHD//fcDKN9lIe7OFi+f+MGLKtviaJ0opCF+PQCWciPRDFEUgr4/fa/6+nr4/X6cOnUKV1xxBXK5HBYXF7F582b09PTg5z//+Wre6v/nplarcdNNNyGVSvFc8KZNm7Bp0yaEw2EOBqQSLwoXi0rQpVKJd+QAKBNFpnE4ceOdwWCAz+eD1+tlUVn6LEn0ZGJiAsePH8eNN97IC81yuRycTic+/vGPY8+ePR/WsV2S6fV6fPazn+XtibRB0Ww2syAyaaquVLwWHRgp3/wxVWxR8ZzUcxQKBeLxeNnuaHHDKO08AoAdO3YgEonwZzQ4OIjbbrsN+Xwen/nMZyp66kWj0eDWW29FJpOBSqVCsVhEe3s7B3ZSGRc3hYrPO4mJ0Agz/T2NQwNgjVQx0NEKWtpuCaBMOEcul2NoaAixWAz79u3jOf9cLoe2tjbY7XZ8+ctfvqT3uOrSWyKR8ApKUQiW3gQ5L7qUAHjemv6fRDDI0YrKLQDKZsABsCOmgxDVt2lWeX5+HgaDAQ0NDbz4vLOzExcuXODvX+lGGTE9dOK50J+Jmnz096JwCM0ci5qWdGZkdNkoSxclq1Z+rUQiQVtbG3Q6HQ4ePAi1Ws2aonfccUfZJa1kI+dOgVw8T/HMRO1T+lo6Z/G/NK8tZvji5yAKUa88e0oucrkc6urqkM/nMTIywoLYIyMj2LdvH69dXgtGKkFkorq7qAIk/p2ovkQZvpiVigvfxISA/AHN2YuSjKLR3ZXJZDhx4gR//w0bNqCurm5Vyler1nCih4m8MzlIMRLQ3hZyiuKlW1xcRDQaZdVpKt36+vrg8XjgdrvLvgdFJLowYkYqqr+QKK3NZkN3dzfuuusuDAwMlEnUV7JRxme32+H3+wGA35tYtohZNJVu4hqJ+fl5FAoFTE9PIxKJQK1WQ6vVwuPxsF4lXUD6TOhC01oE+jNSFgKWMp5Dhw7h5MmTuO6663DrrbeyTNVaCEIkKLKyellZtomK2nTncrkcy57Rsis67w0bNgAAJiYm+ExJEIM+G1Exnr6/mEGRmnd7ezvOnz+PW265hQWBySFXspVKJahUKtaABZbPU8wa6c9pISCAMnWrSCSCQqGAubk5TsgkEgmv5iBVMQpYFOTE6hZYvs9UgZrNZszNzWFxcRHNzc1wuVyIRqOrkrBbdUap1+tZak28FGKUoBdJb4LKGo1GA4VCgYaGBuh0Ouj1eqjVaqjVauzevRu1tbU4fPhw2epaMdMkZ0GHS4dD2enMzAxqa2uxd+9eTExMsJ5fpa9TBZbXp5IiEr1XkpISSzbxvGlFKgBebrW4uMjyXtFoFKOjozh48CDOnTsHrVbLeom0N4cCGn0/kl0TH3Ba5fHss8/iYx/7GO/+rnTlIGA5i9PpdGXZH2WDlAGSE6P/1+l0iMViSKVSXDpms1nWUCwWi3j33Xdx4sQJtLe3M162MrOk6od2Ea3c99LW1obFxUU8//zz2L17Nyt6U6JQ6UbqUlRaUyAShbdF5XNRcSmdTnPSpFQqIZFIoNFoIJPJEI1GEQqFMDQ0hNnZWSiVyrLvR0LWwDKUIar105+RCtPFixfR3NyMWCzGvusD0aOUSqUs2EsvQhTbpYtGOzPIaCEZ7fCORqNlO1xE4d9rr70WIyMjOHPmDGN19L3oZxA4SxdTTOGj0SjMZjNSqVRZxlnpWY9MJsONN95YtrVSDDSiIKxYbgCA2+1m6TtxjS0teaJGzenTp/HMM88AAMxmM6qrq9lBplIpFAoF2O12Lj/pbEk+bNu2bRgfHy9bRyFe1kq1UqmEhoYGfp1092jVAp0tnS89iJFIBEajETqdjisTCjTi+odSqYRXX30VxWIRN998c5mDpAqI7iJ9pvQzaHmWSqXCnXfeCZlMxnt0aOFbpZtcLse2bds48IgYt9iPWOnAlEolb3ElTJLWctByNsLYp6en4fV6kUqleIX1SpjPbDaXvS5KOPL5PJqbm/GTn/yEN5dKpVLeyX4ptuqdObQMng6BHhQRpAWWozhtViMFbxFvoEshKpgHg0E0NDRg586dePPNN8vKEypjCoUCLBYLOwuxC9bQ0IDBwUEA4Iucy+XKAPVKNNpdIxo9cBQIRJCbnJjdbmd8jRZRibtsCK9VqVRIp9MYHR3FQw89hIWFBej1enaW1GVdt24dPB4PBye6SLlcDg888AAkEglDJqRXSWuHK9UkEgk/kPR7ERISYSN631KpFDqdjjFj8a5RRp/NZhGPx2GxWFBTU4OJiQn84Ac/gMPhKMtSaTkcLS8jyIQy+UKhAKPRCJPJVFYVFAoFrFu37sM8uksyaqoA5Q1cAO/RghVLcaqeRI1baoKRk6OKKh6PY2hoCG+//TYKhQInUeR3CoUCGhoaeM2D+LPy+Ty+8Y1v4PTp0+yMSXv0UjH2VSucU3m9slwpFouMZ4l0FlplQGkzvTAS36S9JT6fD1KpFDabjReoNzQ0wOv1lrX2FxcXWeyTmkoUWdatW4cLFy5AJpPxSt1sNovZ2dmKV4qmnegU7eiMKQtZSR0BAJ1OxyKxNTU1yGazyOVyvINaoVDA4XAgmUzyRaaS/D//8z959474/auqqrB7926+uOSgSWiV9opT6T04OMi4VKWauLpkJZwDLGO95DDJIVLwoAybKhgSoaWVzLOzs4jFYlw69vb28l55MbuirY70PaiM9Pv92L59O7MZSERYpVLh9ddf/7CO7ZJtZfNPbOICy0mT+GcmkwnJZJKxcQq21GCprq6GwWDg/eqUYSsUCnR3dyOfz7OzBJb8j16vh91uZ39Ad7ejowOTk5NIp9NQq9XIZDJIpVKYnp6+5L3pq3KUYikmPkQicCpGX3GZEAAu7yQSCa8QyOfzCAQCHDnFy1hXV8dYhditBJYWA5nNZu7qkrx+Op2GTCbjdDydTqOuru6SU+wPy+hc6SEF8B4MUDzTdDrNWaNMJkMsFuP1tuKGxYGBAZRKS0uexI45APzt3/4tb6oUWQvNzc3o6urihzyVSuGrX/0qcrkcZDIZL+HKZDJ46623Kv5sxUwSANNWVkIGYtCn31MyIDZjCoUCbDYbIpEIQqEQAPAWR5VKhWAwiBMnTvCqAfoecrkcVusyHZJw9V27dkEmk3HpSvSlkydPvqeTW4kmZuHAcvNRDDi0uZXukwhb0O55SrZozcPk5CTUajVTjoClz3BhYQFHjx7lny+uTlm3bh0sFkuZA73lllu4mReLxQAsOeTJycn38Dr/lK3KUdJDSWWXSF0BlrmVUqkU4XAYBoOBO1XUvaZIvLi4iIsXL8JkMkGr1SKRSHBaTJni4uIiXC4XJicn30MzyGazOHr0KN5++20cP34coVCorFtcKBS4Q7ma/b0fplFAET9kKg2pZJNKpVCr1dDpdJx95nI5ZDIZDloOhwPnzp3DwMAAjEYjFApFWcZI5WQ2m8VDDz3EzhJYvvQ333wz3n77bRw6dAj33Xcfd3rpsisUCvzwhz9ELBareEcpVkDkLEXsTCSF63Q6BINBbigS6J/JZJBIJOByuaDRaHDx4kUORoSXk0OQy+UIhUJljRwKLADw4IMPolgsQqVSYfPmzczKIPzSZDLh/PnzUKlUFQ8ZAcsNRhGSEyk/dEfp/V199dX8bynpIlhNpVLhyJEjCIVCMBqNZU6XVj+QswyFQhxICJKSy+Wor6/H73//e5w4cQJf+MIXuMokPnZNTQ2effZZrrwuxVblPUqlEhO5yZFReSLuNxYzD7FkpoObnJxEPp9Ha2srwuEw0uk0v4FkMsmHTFlWe3s7Tp06xd9Ho9HgmWeegdPpxNatW3HTTTchEAi8hzc4NzcHs9kMo9G4mrf5oRmdA2U65NQoi6TM78SJE4zF0ENM2+jMZjNefvll3tdCWDBtWxS5aUqlEr29vQgGg+wgdTodhoaGcP3112NiYgJ+vx9vvvkmdyppedzPf/5z/tmV3igTu68iP4/eC2XgUunSiliDwcDnRk2cYrHIvNxIJAKtVssPJ3XUxUacVLq0pE2k9+RyOdx77734/ve/j6mpKZw5cwahUIgJ2eRQIpEI0uk0QqFQxZ8tAK5w6DwA8BmIAymlUgmHDh1i+K1YXNrfpFarkc/noVQqcfLkSWzcuJF7EbTBUeQJA0sQ0qFDh8r2F1EW/vDDD/OaYLqftNZWIpHg8ccfX/U+olVnlNS9owyE/p8OhrAErVZbttwLAO9OdrvdMBgMjF1Sqp1Opxl3I94U/UyPx4NEIgG5XI4nnngCd955J3MEI5EIWltb+WAWFxcxNTWF5uZmziQq3ah7J5Z9tExJLBNLpRJGR0fLFr/X1NQgl8vBYDDg3LlzcDqdZQuYrFZrWTdXo9Hw9juTyYTHH3+ccbdf/vKXuO+++zA3NweDwcDfn7JZyiwvXryIYDC4Jsj8Im9P5NtRhidSzJLJJNOIFhcXoVQqEY1G4XQ6cfLkSX6/sVgMi4uLUKvVvKWRsDbC4xUKBXp6eiCTyaDRaGCxWPCVr3wFwWAQ1dXVUKvVmJmZQTKZZCpQc3Mz+vr6OGFYC6W3CMcByxAdJVKUbUskEgSDQV5dTQ4sm82itrYW09PTsNvtCIfDDCGJK5FLpRKft0SyNOLr8/lQKi1Noz3yyCPo7u6GXq+Hy+ViyhFBSplMBjMzM7DZbAgEAqviUf5ZS4MJqxFLbxFXq6qqQjgc5khAxORwOAyNRoN8Ps+LyqkRQFzBeDwOpVKJSCTCmVChUIDZbOa9yvv37+eOGf0MtVrNZQpdPOqarQVHSR8mZSiUuVNmQ3+vVqsRjUaZuiJG2IGBAZjNZszMzHCHVaFQIJFIQKfTIZlMcoCi86dMcWZmBnNzc7w+mBoThUIBV155JXcps9ksHnnkkYovt/+YUbYnjsYBy91RYmEQjk53yG63w+fz8UNLVY1SqeSGBGUstI2UKgBiHczMzGBiYoJJ5OS46WcQbn/27FnG3MTnai2Y6BzF5iNVRYTRUhAQ7/bo6CgMBgPm5uY4WdLr9Ugmk7BYLEgkElCr1YjFYkwfojsolUrR3d3NVQFVAMSCGR8f537ISy+9BKfTWQZHXYqtGrijDI14UYRJiJdvcXERer2eS5GamhokEomyPbrk5ACUlYmU6RBWSaVIIpGARqOB3W6HVqtFLpfjsTuPx4O5uTnOoHw+HxwOB6fWa8FRilikyBqgcxW5X3K5HGNjY4jFYnwpiWyfSqVgMpmY3E+Yplqtht1u5+9jtVphMplgs9mQzWaZoE/UF+oWms1mdHZ28uf+2muvIZVKcVASH4pKNir/KHMUhxdEMrQYLKjDPzU1BY1Gw5UOJQfU/KEHM5FIcKOruroatbW10Gq1mJ6e5iYCTaKp1WrE43FcddVVHOyLxWIZ3YocT6XbSvqP2FwRHb5CocD69esxPz/P2GM6nUYymYTT6eSKkuCnbDbLMIler+c+h8lkgkqlgl6vRywWw+DgIEwmE6qrq/lZcTqd0Ol0OH36NPO6f/Ob36C5ublsyu1SbVUZpdi9ozcqluH0gBO9h+g/qVQKNpuNaRjV1dXQaDQIhUJQqVQwGAyQy+WYmJjgvycnOjs7C5lsaRG9TCbjXb00fSKTyZBIJJDNZtnJEA9QHFlbCybSKIBlOpYIktNYVyQSQTAYhMFgQDAYhN1ux/T0NEwmE5eFRqMRSqWSKUOEv1HZ4vf7MTc3x9w9yvQXFxexZcsWuN1u5mkSf/LYsWOMR9FnVOlG8I1IbxNJztTQKRQK0Gq1/JBbLBaMj4/DbrcjEAhw1UJYIpHFa2pqmKpC8BMADvDUwKQuOnXGtVptGaWut7e3bICCXvtaMNHx0P9T8BezSxoDVavVyGazkMvl0Gq1mJ2dhUaj4ekno9HI1Wc6nUY+n4dWq+WSfG5ujjvZNTU1fLa5XA5XXXUVampqsGvXLkQiEW62EWRCZ76as1116U3RgmwlERpAmdSZSAwVyboSiQTJZJIvUDAYxK5du6DRaKDT6XDq1ClMTU1BqVRCoVAgEomUddJuvvlmzM/P43//939x4MABzMzMAADm5uZgt9vLXs9auGwUhCj4AOVNCFJYAYBAIIBEIsHL5K1WK2ZmZmC1WnmMrL29HY2NjZiamuJubTqdxsaNG9Hb24vDhw9jamqKs6N0Os1YaFNTE/7iL/4Cp06dwr/8y79wJnbu3Lmy7iG9xrVwvuIwBLA8HUL3F1hqttB4G0nPmUwmBAIBrm4ymQwsFgtIVmxhYQELCwvw+Xyw2+3YvXs3JicnEYlEUCqVOAABS0yNt956C/fffz82btyI8fFxxu+o800sEara1oLR3RUzNDpfkV5VLBYRDoeZ80sNxdnZWVitVma+tLa2wmw2Y2BgAHa7HSQ/2NbWhldeeQVerxehUIirVnHkt7OzE1dccQV++tOf4vDhw3jppZeQzWbR3d0NjUbD2fpq7+yqHeVKAFSMFlSO5/N5OJ1O+P1+hEIhGAwG5HI5jhAUsdevXw+z2Yxz585BKpXiyJEj0Ol0ePHFF3H11VfD6XSWjSqVSiVs2rQJyWQSTzzxBKampvDkk0/if/7nfxjfpPSbsp21UBYCy69TDEJiA0ckmhNuSyNgo6OjsFgszFN1u92wWCwwmUxwuVyMqxGB/+TJk5idnWWchpyI0WjEt771LWSzWbz++uvYsmVLWRf35ZdfZkrGWgpC9DCJNCDKkkVuJTWvqMlFd4rw8Fwuh9raWgwNDcHhcGDTpk1Ip9Po6OjArl27sGXLFhw+fBizs7PvaXIdO3YMBw4cwP333w+v14tHHnkEDz74IMNUw8PDZfxkqiTWyv1deZbiM0sBie6NwWBAOp2G0+nExMQEHA4H4vE4crkcrFYrrFYrNBoNPB4PXC4Xw0pvvPEG/H4/V0v0WZZKJVgsFvzVX/0Vuru78dhjj+Hv/u7v8Pzzz3OjuK+vDzqd7s8ePFmVoxRLFfoASbWHRg2pHCG+k16vL8MtYrEYdu3ahYGBAQQCAUxNTcHtdmNkZATFYhGnT59GW1sbDAYDl9OLi4vYtm0b/H4/jh8/zl2tf/zHf8Sjjz7KXbBsNgur1fqeyLYWLpvIcxSzdorMZMlkEjt37sSFCxfgcDgYjyWGQWtrKzo7O2EymWCxWOD1euH1etHc3IxXXnkFBw8e5M+F5pmtViu++c1vYmBgAM888wx8Ph/+7d/+DR0dHSwoMjg4yBQwEYRfC0ZnKE49iaO3wDJ3lTQJqGtNTTCNRoOJiQmsW7cODQ0NAICpqSkYjUbU1dVhfn4ev/71rxGNRrnrXSgUcPbsWXz2s5+Fx+OB3+/Hz372MwDAV77yFW70EBVIFOUgWwtNMxHKoCxY5K6So8xmswgEAvD5fHC73fzsp1Ip5PN5tLS0wOPxMGbb0NCAN954A9u3b8ezzz6LgYEBtLS0MOUqlUqhtrYW9957L1599VU8/fTTiEQieOqpp3D8+HFmErz77rtoaWnhLF8M/pdqq84oxchADwx17siIcE6UH61Wy+Csy+VCX18fH2AymcT4+DhPmRgMBrS2tpal0+l0GgcPHkRjYyMcDgdMJhM8Hg+eeuopLk8JK6KGBL3WlSVXJRs1AQCUjbmJJOlUKoUbb7yRqSpms5kbKw6HA+3t7VAqlYjFYjzh0dzcjFOnTuH8+fOc2YvTSzRmSuXf008/DZ1OVzbydfz4cSQSibIJHgLJK91Ex0OZJN0tcSImn8/D5XJhamqKzz4cDkOhUCAcDmPz5s0cvBUKBfR6PQqFAnp6eiCVSjE9Pc38PXIOn/rUpzA7O4vq6mqMjY1h586daGxsLJsCmpubKxOMEF/rWhiUECE1ckJ0tsQPJc7k7bffjlOnTnHPIh6Po1QqweVywe12o7q6Gj09PbBarVCpVKivr8c777yDyclJDlDENgCWRiH7+voYh3zyySdx7NgxxooTiQSmpqYQj8eh0+k4OAGrC0J/1qewsuwiYFu8fLFYDHK5HB6Ph//eaDTybDHhYZFIBCqVCufOnUN/fz9nhMQrzGQyqK6uhtlsRjgchkqlwvz8PF599dUy5SHSuKTRSKA80q0FIwxWpFqJnW8Rj0kkEvB4PACWmj4GgwHXXHMNWlpamMTr9/vhdDpx/vx59Pf3Y3x8HE6nEwBY7DiZTGJ0dBRVVVXYunUrPB4PTCYT0uk0UqkUSqWlVRAU9SkS02e9VozOlrrL1BQAluGkxcVFbhik02kenauqqoLT6QStk6BmptlsZv7u/Pw8lEol81QJ852ZmWF1p61bt6KhoYGrLtIkyGQyiEQiZaWkqHS0Fkwsu4HlFRgipERBure3l3mOqVQKTqcTGzZsgEwmw8LCAlwuF6xWKywWC9599114vV5WpKeGYyAQQC6Xw8DAAOrq6uBwOLBlyxZ0d3ez5gEAzM/Po7a2lktucZDlA+t6A+WMe3JE1GigEkculyOVSsFisWBsbIzxtEKhAL/fzzJS8XgcZrMZf/jDH+B0OmE2m3nckbId4kvV1NQgEAggEolw553+KzoSWqVABGvSzqx0WwkRiKwCmoKSSqUMeG/ZsgVyuRzhcBjr1q3D7t27OTpbrVb4fD4kEgk899xzePbZZ6HX63liymazwev1olAocEdxcHAQH/nIR1AoFJBIJLhhk8lk8Pbbb3PWYzAY+LwJdlkLJma/5Czpv/T/0WiUIYn6+nq+pzabjXE1mUyGZDIJu90OpVKJ7u7usgkao9HI9CJKENLpNAwGAwKBAGevdL6JRAK5XI6zHXKwVH6uFVuZta/kBVPTMB6Pw2g0YmRkBMFgEFdccQWsVivi8Tj3MAKBAGKxGA4fPoy5uTl0dnZiaGiIE4Lp6WlUV1cz9k7NShrQAJayzlQqhTfffJNHI0VsnTLfS7U/K6OkcpZ+0ZgRRWjKLmlU0WazIZFIIBgM8hwtObRAIAC3242mpibk83l0dHTwVAnxAAmQpYe9qqoKqVSKM5tkMgmTycSvY3FxkRZe8cO8FkoYeo3k9EXogB50UnunHStarRbbtm1DMpkKx2f7AAAgAElEQVTkICWRSGCxWGC1WvHCCy9gcXER09PT2Lx5M6qrqxEMBlnTkiLsyZMnOdCRgg1RkcbHxxEOhznzoekTi8VSxo2tVBMHIyjwSCSSsvFLIpETBJRMJhGNRtHW1sbBQ/w6rVaLkydPYm5uDlKpFMPDw1Cr1ZyJA8vjpRMTE5DL5UyDI/xToVBgZGSExYFJkpAUvf/Ufp5KtJVVG/1enJenO/yJT3wCqVQKra2tPM1ECkGU+c/MzCASicBgMCCVSvHQA9F76AwLhQJP2/T19XHmSkHQ6XSywyapR2AJixbVh97PVn3DRQyCsh4qB+mA6IEdGRlBS0sLotEoSqUSU39ov0Y0GkUul4Pdbkcul8PGjRtx9OhRKBQK7N27l0sipVKJO++8E0ajEaFQqEwjEFiGAkTicE1NDWexlIVVuonjnkC5WhOVjVVVVRgeHsb+/fuRTCaxd+9ePm/qYofDYbhcLvziF7/A9PQ0QqEQrr76atTW1uKuu+7CFVdcwRw+nU6HO+64A7fccgtSqRT0ej1PjUgkEly4cKEMi6byki43UT8q3cTgLo7UAuUL2ubn53HgwAHk83ls3LgRpdKS0EUymWS82Gq1YmBgAH6/H5FIBPF4HF1dXejo6GDNAblcjng8jkAggJaWFqam0JkBKJN+A8AOhdR1KDlYCyZyKIHluyvCNQBw4cIF7NixA16vF06nE0qlEnV1ddDpdMhkMojFYli3bh3Gx8d5qu/KK69EJpNBR0cH6uvreShFqVTipptuQmNjI6LRKKLRKDKZDCdLJ0+eZO4x4f90V//YIrn/N1t115tELwloFj9kGlMaHx9HU1MTHnjgAbz88su8PoLKFsJ+Zmdn0dzcDL1ej40bN2J4eJhHl95++21uJNx555147LHHYLFY4HA4uLSh10FkYnFSiDr0IhG4kk3MzEWjTJ04eYcOHWJNvs997nMYHh5GKpXC+vXrYTKZMDk5CY/Hg1/84hfo6+tDLpfDY489BofDgXQ6jWw2i6GhIVZl+trXvoYtW7bAbDYzKE8lTqlUwqlTp+D3+8uyTzpP0elUsokUKLoXYqAnBzg3N4d0Oo0f/OAHuOeee9Db2wu73c5sCr/fD5vNhlQqhfHxcYyOjmLbtm1oaWnBRz7yEbz++utobm5m6CIWi+Hhhx/G2bNnOWMSIRWic9FzJI71kTblWghC5BfE3wPg4E7Z3+nTp3nO+i//8i9ZFSyVSjFFqL6+Hj/60Y9QKpVgMpnwrW99C1NTUwDAPEway33ooYcQCoV4Yo3uLVVFIyMjrDsBLPOSqaIQHef72ar1KGlyY+XoIo1uEcb4/7T3rbFt19f7jx07vsd3O46d5gJtk15CCm2hXEdboGhDsA3GJITG9mqaxKbxCmnqi03a9oqJvWFiE2N7sxU2UXUbY4hLL9BCW9r0mjQhcRO7juNb7PjexJffi+g5+Zq//r82SKyufj4SoutoY3++53s+5zznOc/ZunUr9u/fD4vFIhM3uVwO3d3dQtTt6uoS53355Zdx9uxZXLlyBePj44KdGY1GvPjii6JGzJKd9BQGQ6Wi9xfH0wDcFHtzlOR8Zh2cyy6Xy7KS96GHHsKjjz4q3D/iaLlcDn19fUilUjh27BjC4TC+//3vI5lM4vTp0/jtb3+Lffv2yaRILBaD3W7Hn//8ZywtLUGn08FutwsPsK1tWdFe2SlkxaDT6WAymWTSp5mNl5ASyGdQ0mq1CIVC2LhxI1wuFxYWFvDTn/4Uk5OT8Pv9sFgssNvtWFhYQGdnJ1wuF0KhENLpNO677z48+uijcLlcOHbsmMwpU9j4hRdeEFVtpZQgX1j6NC9Bnjk/c0dHBwYGBm7w6V2fKTcX8CJSZm1utxvpdBqPPvqoTN0QXjMajTL/PTMzg0KhgGQyicceewwHDx7E2NgYfve73zUkaKVSCWfPnsWrr74qa4ddLhcSiYTQv9xutzSPgRUWjF6vh06nk9HT67FV78xRLu9Rlrrt7e2YnZ1Fb28v7rrrLhw5ckRoQzabDYFAAGq1GqlUShosRqMRGzduRDwex8aNGwEsU1zIQ+PtGwgEGnbusGNIYFxJgqcT8rOlUilcuXKl6cfteMMpR9qoXl4sFhGLxfDee+/h17/+tRCbTSYTMpkM4vE4NBqNaHv+8Y9/xNmzZ/Gzn/0MU1NT+Pvf/45Dhw4BgOCNvGzeffddrFu3DiMjIzK2xzPNZrMNwiPMKrlxz2Aw4P3335c55mY1+pHy1/QjioP86U9/gkajwXe+8x3s379ftAYuXrwIYHmpntPpxJkzZ3D69Gn09vaiUqng8OHDCIVCKBQKcsHx8j937pxkrPRVjn+S9cGsnJ+JvtvX14dcLocTJ07cyKO7Lvsif1IZLCuVCtLpNCKRCPbu3YtgMIiRkRHp8jNDJL3v0KFDqFQq2Lt3Lw4ePIgzZ85gamoK27dvx/Hjx1GtViUGhcNh7NmzR85Zr9fD4/GgXq/L1kXl+mGW68x+P/300+tOoL4URsmxKxopJB6PB263G59//jna2tpEWdzlcqGvrw+hUEiIp2q1Gg8//DAuXLggcmk+nw9+v19wMN4eXq8Xfr9fsidmrcQwlCIHyvIql8tBpVIJoNvMRgUkpQAqs7dCoYB8Po/+/n6YTCZ89tln6OnpkbMYGBjAyMgIVCoV/vWvf+HIkSP4+c9/LmOfuVwOY2NjmJycRL2+rCjNP5tKpRAKhTA3NydBgxni+fPnRfIKWAnm/Ofdd9+9KZZfkfdIaIZVBn0xHo/D7/cjEAggFovBYDCIZB0rm6WlJWQyGRGD7ujowOLiIlKpFKampjAxMSHQBkv5RCKBaDSKeDwuPk+/ZQlICIPZrUqlgsPhwKVLl2ROudlNrV7Z0UTjOReLRRQKBdjtdkxPTyOfz0uWrNfrpUrSaDTYt28fzp07h1dffRXvvfeeZI6xWAzpdBoGg0Gk7wqFgpw/LyiujwCAkydPCgQCrCw3Y8X22WefrWpKZ9XCvTwApRAqS+RisYh///vf8Hq9qFQqSCQSQir98MMP4XK5pKO9Z88e/Oc//5GGCzllxOmsVqvgc/x5DKBtbW3Ssfzi6JSS9FooFGC1Wq97L8aNNJazfHFYyiwuLsr0B+GLxcVFdHd3o1KpYN26dfB4PDh//jy0Wi1Onz4Nt9uNoaEhaURQco2ydfw5VMKpVqswm82SlU5OTqJcLuPs2bNyznzefE7Hjx9vyMya2ZQTIszw6vW6NBIAYGhoSHDEwcFBlMtlbNu2DXfeeSdyuRz0ej1mZ2dx8OBBWS3Av5v7gwBIFcDMilAVMy4yCcgi+CJ+zvXDxWJR9FtvBlOWt/QpdvH7+voQj8dlxttgMIgYTjKZRCqVArDMKPjmN78p1SiDm5IBQ/k/YGVVh91uh1qthtVqRSQSweLiovw8fiYGSb1ej48//hh2u12e1fXYqgMlb0EeRK1Wg8VigdVqhdFoFLCbCkGUSlKr1bjvvvsQj8fR29uLaDQqwDWDr5IUfPXqVVmLqlSrYSetXl+WwVIqIAMrEy3pdFqwC+V0SzMbv4uyWaLX6xGLxbB582YYjUbMzs6Kcnx7ezump6fx/vvvY8eOHchkMvjoo4+wb98+WfBOrUSPxyOsA6PRKNqAPJtbb70VpVIJdrsdsVgMRqNRyiJiOwCE+0dZu4WFhRt8atdn9A9mlRrN8irYiYkJ+Hw+ZDIZZLNZ5PN5oaWkUinMzc3h7NmzcDqd+OSTT/Dss8/Ks+Fz4joIALLjBWjUuWSSQXZCMpkEACnJ+XcZjUbMzc1Jw+JmYGsAEMYFs2Nmz/V6HWvWrEFnZyfGx8dFwaqtrQ2zs7MYGRnB7t27kUqlMDIygh/+8IcNzRbqGWi1WskAVapl9Stm4L29vchkMvD7/bIl1uv1NsQFfibiyORcfiXNHJVqWSaJBFuC4zqdTl5evljsfvNDut1uvPnmmxgYGMDw8DCCwSDUarWInJJxv7S0hHQ63YA1Mouh49TrywK2SoFeZbbLUgtYySaa3Zg9cpGakndms9lQqVREZ9JoNMqeoM2bN2PDhg3o6+vD888/j1deeQWFQgGxWEzKOXL2uJnSYDDA4XCgv78fZrMZJpNJBB74El+6dAnFYrEBmyQWferUKQCQINns58tzUPoKz5fNmnK5DJ1Oh2g0CpvNJpdWuVzGiy++iD/84Q9Yu3at7IFSngmpU06nU/7xer2i3t3b2ysl9dzcnPg7z5YvKydT2AhhNnUzGMnyzNoByJgiFb02bNgAjUaDo0ePIhwOY3BwEA8++CDGxsbwxhtv4IMPPsDJkydFZ5XnosxSOzo6YDabZS+9yWQS3NnlcmF2dla2NAKNClwajQZjY2MyiPLFIY//zVZND+LtqcStcrmcjBAmEgmhlvT09Ej5/d577+HrX/86/H4/jh492lA2M9oz+pvNZslaenp6RNiXmE25XJYteF/8O9iJJ0CslHxrZqNTJBIJ+U4Eo7meIBgMiqJSLBZDOByGwWDAsWPH8NZbbyEQCMDhcEinnJAGs0ritleuXMGFCxcQi8Wwc+dOIfPmcjk4HA4YDAYcOnRIyP1Kwn5bWxtyuZw05Jo9SNIKhYLMvwMrGYZWq4XZbEYikZBO6OXLl9HZ2QmtVov169fjpZdewtWrV7FmzRphdyibNsByJTM/Py9nnkqlGjZcclWETqeTYMk/r+Qms4piQtDstDaaspvPqS2S79va2nD69GlYrVZR6KcGxNtvv41AIACbzYbDhw+LLB2DG+lqrCDL5TLGx8flZ91zzz0y3cSdW6FQCBaLRXB4YIVHy6lBUgqv11aNFCsfKi2TySAQCEhTh5hDZ2cn7HY7PvjgA9x7771CPKeDdnR0NCjSLC0twel0olgsoqOjAzt27IBOp8Po6KgQfinjRjVkpaPxgPl7SjpIs5vy1gNWurOcjy0WixgeHhZalcViwdzcHI4cOYIjR47giSeewLPPPotSqYRcLieZ4/z8vJQgnEGmFmBvby8MBgO2b98uAaBer8NisYgKNbCiDs7ykLDIam7kG2nKyRxltm6xWKTh0t3dDa1Wi/b2dinfbDabXEBbtmyRJWDs8pNMbTKZJNAGg0HMzMwgnU5j165dGBoaaiivqa2qFOdQNmyU1DBldtbMpuwLKBkGpVIJnZ2dyGQyuO222zA7OyuCLJ2dnfjb3/6G+fl5DAwM4IUXXkAoFBLxDF4ihO3MZrPAfT6fT3B3vV4vEBE1JtjQAVZ8F1gRwuYSxK9s1ls5JcIPUalU0NfXJyOKPp8P0WgUnZ2dqFarmJiYwJNPPoloNIoNGzZgcnJSHEHJiWSH7Pjx47DZbCJq6vf7sXv3bkxMTECn00knm7PPdDb+byWRmKWL8uZvVqNTcM6dnz+Xy2F2dhZbt25FtVpFJBKBVquVG/T06dN48sknMTQ0JBdCpVJBV1cXlpaWMD8/L+Tnnp4eucw4AbFt2zbRWiTsYTAYGjJGsgjYuOOFB+CmuIQYKPkdq9Xl/duTk5OoVCq45557kEwmkcvlRB3eYrHgL3/5C3bv3o0LFy4IzaS9vV0W59EPa7UaNm7cKNNlAwMDaGtrQzgclgqII6M6na6haiA1CIBc9jxzZdC8GUz5HpIHOj4+jjvuuAMLCwu4ePEi1q1bJyO18XgcP/rRjzA0NITp6WkJkm63W5It6hzodDoZrc3lcrBYLPjGN74hbA1S6NLpNBwOB4BGHVeVSoVSqSR9jdVWQ6vOKInPMKVVq9Wy8Mfr9aJYLMJoNCIajUKv1+Puu+/G6OgoHn/8cSm5ldkT019G+7Vr1yKfz6Ovrw/1eh2Dg4MIBoNSjihvX+WCIWUJxBlZBhul7mCzGoM8u6IM9tVqFdu2bZONh2fOnIHD4ZAZWJVKhUAggL6+PhFUYPbY3d2NhYUFeeHi8bgseuIkxAMPPIBcLifntLi4KNgasBIISa9hE4TPn53FZjdmMgCkG+31elGr1TAxMQGPxwMAomj+0UcfYdu2bTh8+DAeeOABedE4vkloh+IsiURCdq1rtVpMTU1hx44dwj1mR53BgJc4fZSZDql3XN96M0w+qVQq6dIr56lHR0exbds21Go1ZDIZJBIJUYq32WzQ6XQYHBzExMSEcE3ZvW5rW5a448VGf1Sr1ejq6oLRaJR3hOtOGPz4TJTTemTKAI0+fb226kBpMBjkAbN5Uq8vS9kHg0HBZW6//XZ89NFHqFareOyxx3D48GEAK7JRSmGLtWvXIp1OC/VFrVbD5XLJDVUsFoVGpEzzv6gAws9TKpVkyoU4X7M7HCkjNJbBbrcbdrtduIs+nw8OhwPz8/PIZrO46667cPvtt8ufYXnJxtrw8DBGRkZQKpWQzWYxMzMj0nXf+973UC6XUSwWpTnGwKfRaER7kY5GxRaSfnnJ3QzZurIByYuTikF+vx/AMnXn/PnzGBoagsFgwNjYGHbv3i0XtF6vF1jo6tWrwjFlYIvFYrhy5Qrq9WUJNuWFQ7hDmRgod0+xjORFx/Pnn29m4/my803/6+rqkoA4Pz8Pp9MpQTCfz+Oee+5BMBgU/zEYDA30IaPR2DD5l81mxfeeeeYZJBIJLC4uSkOHVYOyOcZnrtFoZOMr4Rdlw+hatmqkmJJFDEp6vR6nTp3CpUuXZM4SAE6dOoVNmzbBZrPhwIED4iC8YWu15SVier0eR48exYkTJxAKhaQsYiD95JNPJPIrKR6FQkG6i5Tt52Ex62XzaefOnbDb7av9qv9VI6WBQZ1A9qlTp2CxWORFValUImhBnignTPjQme1wWqGrqwtutxtWq1U64LzQSqWSnF2lUhFCNRWd6Oh8blNTUzLlpNfrZda5mY2BXolLMVOhz1arVWSzWQwMDGBsbAzJZBJDQ0MNVCLiuJyq2b59u+CQVqtVLpXe3l489dRTMgyhZIiw1CaEwXFgJX7KzP7YsWPo7e29sYd3HUbfZUAi62RyclLgGpvNhoWFBalo4vF4Q4bNZiOnZvjuMtgRZzcYDLJnh0kQGTflchm1Wg1+vx8ej0fEMHj+wWAQBoMBwHJQ3rp1K7q7u6/rO64qUHJ1rMVikYdKOXYu9GJHWqvVIhqNCsZgMplk1SoxL2AZpwwEAhgcHBSFY2atRqMRmzZtwu9//3scPHhQMB7iPdPT0xgdHUUymURHR4d0z3w+n2S9Pp8PJ06cQCwWW81X/a8bJzd6enqg0Whk09zmzZsRCAQEZ9y8eTNmZ2dlPtloNOLQoUMN8+10uPb2doRCIZw5c0YIzh0dHXC73ZiamsKPf/xjvPjiiwiHww3PhuNler0e69atw/DwsBDbuT1Qo9Ggu7sbr732WtNnlBRGYCCjYrnNZmtQopmdnZWxTb/fj0qlgitXrshlouRPms1mBINBHDp0CBcvXhSIw2g0Ynp6Gi+//DJsNht27NiBQqEgupTcJ9Xf349169YJ3lmpVJBMJoVQHQ6H8atf/Qp33HHHjT6+axpLZgYhtVqNy5cv44477hA61ZUrV7BlyxZ0dHTg7Nmz8Pl84qvlcrlh1zcvjampKSnl6/U6Ojo6RCn+lVdewTvvvIPZ2dmGaba2tuVNrbVaDU6nE4FAAF6vVwjm5Cn39vbiueeeE/m8a9mqAiV5XolEQkBnNgXy+Tw6OztFgsrpdKKrq0uc6urVqygUCg0itFyjqvy1SqWSdj9H+p5//nns2rULr732mpQr5FZaLBZUKhVMT0/jzJkziEQimJmZQbVaRSAQwOTkJMLhcNPTLFKpFA4ePAiv1wsAwgszmUyipcetiYODg3jiiSdQqVQwPj4Ok8mE0dFR4WKy0RCJRCQzVdKmMpkM2traZO/x008/jb1798otTIHlWCyGmZkZxONxqNXLoqvpdBqlUgldXV148803MTo62vSCI5zDdrvdgg2uWbMGmUwGDocDly9fxvnz52Gz2US1PJFIoFqtYnh4GADkHNm5ZtdVydnlEjJixwcOHMAvf/lLbN26FRqNRkpE4nVcXHbrrbfC5XLJypRsNouf/OQnKJVKQkxvZiNvVymM4fF4MDg4iHw+j2g0ivn5eQQCAVSrVSHt8/+jj1NImhmnUryE/pvJZCSrDAaDeOONN3Du3DloNBqhX3HvPOG8arWKdDqNZDKJUqmEtWvX4plnnsHHH3983Q2dVUUPqsvwVgWWFZ0J7icSCczNzcFqtSKZTCIYDOLq1asIBAJCrSD2BkAIzKT8MBXnDU8+FrHIH/zgB3jrrbeka8XMizxLErOtVqtgUUzPmz1QarVaHD16VKaSSEPp6emRDGfdunXo6emBw+GQstFmsyEUCmF6elpwWWK5wWBQxI45RUUM2Gw2i1CtVqvFvn378O6778Lj8QjozbIol8vJiz03NweVSoXnnnsOFy5ckKGCZrZqtSpdVfpTKBRq2Aa4YcMGWffL8bpUKoXDhw8LRskLoVAoIJFICJ2IXEHOhCv3fAPASy+9BL/fLxUYqx2qQqVSKbS1tcm0U19fH+bn5xsoWs1ukUhEymyO2ppMJlQqFdxyyy0i2ELqGSEglUqFCxcuCC+SO68mJyeFCaNUKyPXVTna/M4770CtVkvGyfhESIr8S6/Xi7a2Nnz729/GyMiIiC5fj61aZo07dmncIEdHYjD0er3w+XxSZsfjcfT390tWyS9CYJcvOBsSvEWUhNF6vY4tW7b8P0otTPtLpZJkPzabDfF4HLVa7UuvqPxvml6vl7IWgIwokqxsNBrl4iFeyNUCXq8XoVBIsDPlqCfpRsDKCgRiw4Qw6HC/+c1v4PP5BBbhZ+Bl9Pbbb2NpaQlWq1XUbWw2W9MzCgjk05fq9Tq6u7tRKBSECA5A/p1MJgXmGRgYEBER5Q5udlSVI72sdJR4JMv1Tz/9FFarVZIKwhXE7+LxOBYXF0WTUcnYaHar1ZZ3OCk5wMyuKVQ8MTEhXX0mS8FgEN3d3Th+/Lj4OLFKjoEqx3l5psQrWSmpVCr89a9/hcFggNPplM+knMY6fPgw2tvb4fV6Zd8O9RKux76UKAZLZ44Q8YGmUils3rwZxWIR2WwW8/Pz8Hg8iEQi6OzsFCCc9B1iP3x5gcZ9yzwcZXcqEAjg/PnzDWogJK2TXkTMgrQllkLNbDxDvmRKIj23/fHc3W43HA4HfD4fyuUykskknE4nEomEOEg4HMbc3Byy2aywBwiK09GUwZLNs7m5Ofk8dHytVotgMIjZ2VnU63U88sgj8lI3O5sAWGkCcmqEpRk7q2azGdFoVC5Yt9uN2dnZhubV4uIiDAaDNL2IgylffD4zVjv8fbVaLVJsvMSIp2m1WszPzyMYDGJpaQldXV1SPra3t98UXW++13yHeSFzHLRSqSAQCCCdTsuuIY6HTkxM4O6770alUpFmTzgcxuLiouCTDIY8L8YG5XgiRZepLcGGo1qtxrlz5wRX37VrlyihK2lu17Iv1fVWEps5ncAXjgP95IqNjIwInWVmZgYej0cUW7hagFkPM0kGBWYqSofjodHZlMTfrq4umM1mtLe3I5PJiHwbnbOZjd1Cjg0CaMgq0uk0br31VqHlJBIJWRbvdrvx3e9+VzitWq0WY2Nj0oxYXFyUzIdORO1L4joMjK+//jqAleBCbKhUKsmOnIGBAeFvrsbZbpQpyfFkDzBrISZOOhp3uPCs+/v70d3djWw2i7m5OVnxwLKZNB6+mBzxVTYm+FIzg+X7wpefQVKlUqGzs1P0D8rlsmRWzWysJglxMbixw83RWGLEbFoNDAzA6/Xi/vvvF79Uq9WCrZP8z0xS2cFWVjr8PXKAGagJb6jVy0pZXq9XLqv+/v5VJVCrzij5g3kgzOoMBoPsBVGr1bDb7UIo7+/vh9frFeGFaDSKTCYjLytJosobWZlNKru5tdqykO/AwADWrl2L9evXy3jTwsKCzD/zJUilUk0fJIEVvTwGSt6a6XQa+XweXq9XGirVahUmkwkTExPYvn07/H4/gsEgnE4nenp6EIlEkMlksH79ekQiEbl4eBkxs+T5K0vw119/HTt37sTw8DBcLpcQf8nn7O/vR2dnJ4AVFe5mN5VKJYu/lC8dswq+0OTq8aUiU+DEiRMwmUxSFXE6jPgZ/UuJDytn5PnvcDgMlWpZGyEUCsnWTI/Hg4GBgYZpEavVKiV+sxvLW+WiOQbLbDYLl8slVDSOIGYyGfh8PlGqInWNu3I2bdok2bQSZuOlo8Qo6dNvvPEG7rrrLvj9ftjtdrS1La+eGRwcxMzMDG655RZZS5PJZOB0Or+aZg4xROrvMZIbjUb5UiwdWZbt2rULBw4cQKlUQjAYRL1eFxa/2WyWbpmy3KZ98UvQwU0mEw4ePIhPP/0Uk5OT6Orqwvbt23H58mXodDo4HA4RbuB+nmbPeohFMpgBKw5itVpFdIT/rV6vh8lkEhWmjRs3CqF5bGwMtVoNt912mzRmmJXzUiJ9iB1EluMkZjudTjz44IN47rnnsH37doyOjuLy5ct4+umnsWbNGtjt9gah2mY2TowAK+U3v6ter0c+n28IatVqFR9++CH6+/uFZaBSqWA2m5FKpbCwsIAtW7Y0aE0qJ8T4/HgBMbuhkv/w8DDuv/9+OJ1OlEol2cFuNBrR3d0tOo0MvDeDcVhB2ag1Go1wuVzIZDKSGXMkU6vVIplMyrAJ5ftisZiwXZS7uJl9K1WbmL0z05yZmUE0GoVWq0VfXx927NgBn8+HYDCIUqmEr33ta9BoNKIOtbCwcN2bD74URqmc/6VKDUFYZpqcMz558iQeeeQRuR0ZKD0ejyjjKJ0JWNGUpNMpmwXKCQaHw4GlpSUhrNvtduzcuROTk5Oy6ZHZb7M3HMjnYjOKog12ux3FYhE6na5hOVW5XMamTZuQTqcF49XpdBgfH0cikcC3vvUt7N+/XxyXLzNxTpYtOp2uofutUqkwNTWFVFRWRwoAAAQPSURBVCqFeDwOAOjo6MDc3BzWr1+PPXv2QKPRyDx/Nptt+kuIWZlyDI40n1Kp1DAKy2C5bds2XLp0CalUSjZ6lkolXLx4EVu3bsXU1JQwNFgJ8UXmS63k9rGjGwqFkEwmkc/nsXHjRni9XuTzeRw7dgw7d+4U2MjlcqFYLMJsNt+wc7teo0gNz7BarcLtdosICy94JiyZTAbd3d0S8Mj7JSF8z549+Oc//wlgRXBDKRCt9DeyZvgM/H6/bLusVquYnJyEVqvFQw89BGD5PYhGozAajasSRV6VhxsMBhlxo7Px19xLnM/nJTsClhWEEomEjHYxiB4/fhyRSERSZN7oShBcqUSiBHWBZc4aMRFmRo8//jjefvttIVZT9FY5H96sxkZCIpGQi6NQKMhqBpYetVoNp06dQjqdFhK92+0WLqrRaERnZyfefPNNmSQBVnaFE1cG0JBVUnBWo9HgH//4h2TlJ0+eRCQSwVNPPYVf/OIXiMfjGB8fx4YNGxCJRG6KeWR+X/oOmQFms7lhzDWTyWBpaQkWi0Uy+0AgIIvXfD4f7r33XkxNTSGfzwt+S1xdySfl+6FU6Sfnkpni3NwcFhYWkE6nsXfvXtRqNYTDYcHVATT92QIQTjMxPyZD5EaqVCrBFLkXPpvNYnZ2VrL7fD6PQCCAhYUF7Nu3T2AfYIXxovx5hIz4b8aCS5cuQa/XI5fL4fPPP4fH48GePXtEL2Fubg4bNmzAxMSEdMivx1YVQdjt4w4MtVrdMNZFQJaseE5AOBwOZLNZOBwOxONxOJ1OeDweefG/2N1WZqy8hehoDJzt7e0iHEBn379/P0wmkzQ1stksOjo60NHR0fQ4pU6nQzabxejoKIDli4FZxeLiomTjJISTq7dlyxZZ33vmzBmcP39eJkCI7SpfNuVmPKUgA0vwQqGAYDAIo9GIhYUFJJNJCRKXL1+G1WpFf38/0uk0wuEwdu7c2fQZJYMihxuq1arg2iTeU5EmEokgm83CbrfD4XDA6/XKXP2lS5dE2YnVDgMBM3Y2NWjKHTm12vL20d7eXpTLZRndu/POOxGJRGQ+mvizTqcTulgzm0azvOiL8nEApNrgeVFZf3p6WsZw+/v7UalURGOS4ixszgKQ5IkYOy87ZVwgdUulUuGzzz7Dww8/jFpteR2wzWZDNBoVXnZ3d7ckbX19fdfdzFGtJoCoVKoEgJnVHWPTWE+9Xnff6A/x/7PW2X51dpOfLdA636/SrutsVxUoW9aylrXs/6I1d83Uspa1rGVNYK1A2bKWtaxl17BWoGxZy1rWsmtYK1C2rGUta9k1rBUoW9aylrXsGtYKlC1rWctadg1rBcqWtaxlLbuGtQJly1rWspZdw1qBsmUta1nLrmH/A/IFn5h+AeLqAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<Figure size 432x288 with 16 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# coding: utf-8\n",
    "import sys, os\n",
    "sys.path.append(os.pardir)  # 为导入父目录中的文件而进行的设置\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "# from simple_convnet import SimpleConvNet\n",
    "from matplotlib.image import imread\n",
    "from common.layers import Convolution\n",
    "\n",
    "def filter_show(filters, nx=4, show_num=16):\n",
    "    \"\"\"\n",
    "    c.f. https://gist.github.com/aidiary/07d530d5e08011832b12#file-draw_weight-py\n",
    "    \"\"\"\n",
    "    FN, C, FH, FW = filters.shape\n",
    "    ny = int(np.ceil(show_num / nx))\n",
    "\n",
    "    fig = plt.figure()\n",
    "    fig.subplots_adjust(left=0, right=1, bottom=0, top=1, hspace=0.05, wspace=0.05)\n",
    "\n",
    "    for i in range(show_num):\n",
    "        ax = fig.add_subplot(4, 4, i+1, xticks=[], yticks=[])\n",
    "        ax.imshow(filters[i, 0], cmap=plt.cm.gray_r, interpolation='nearest')\n",
    "\n",
    "\n",
    "network = SimpleConvNet(input_dim=(1,28,28), \n",
    "                        conv_param = {'filter_num':30, 'filter_size':5, 'pad':0, 'stride':1},\n",
    "                        hidden_size=100, output_size=10, weight_init_std=0.01)\n",
    "\n",
    "# 学习后的权重\n",
    "network.load_params(\"params.pkl\")\n",
    "\n",
    "filter_show(network.params['W1'], 16)\n",
    "\n",
    "img = imread('dataset/lena_gray.png')\n",
    "img = img.reshape(1, 1, *img.shape)\n",
    "\n",
    "fig = plt.figure()\n",
    "\n",
    "w_idx = 1\n",
    "\n",
    "for i in range(16):\n",
    "    w = network.params['W1'][i]\n",
    "    b = 0  # network.params['b1'][i]\n",
    "\n",
    "    w = w.reshape(1, *w.shape)\n",
    "    #b = b.reshape(1, *b.shape)\n",
    "    conv_layer = Convolution(w, b) \n",
    "    out = conv_layer.forward(img)\n",
    "    out = out.reshape(out.shape[2], out.shape[3])\n",
    "    \n",
    "    ax = fig.add_subplot(4, 4, i+1, xticks=[], yticks=[])\n",
    "    ax.imshow(out, cmap=plt.cm.gray_r, interpolation='nearest')\n",
    "\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "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": "3.7.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}