RBSOM_ACWMdd.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "89abba81",
   "metadata": {},
   "outputs": [],
   "source": [
    "import pandas as pd\n",
    "import numpy as np\n",
    "import gc\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "from scipy.spatial import distance_matrix, distance\n",
    "from tqdm.notebook import tqdm\n",
    "from collections import defaultdict\n",
    "from IPython.display import clear_output"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "cce40f61",
   "metadata": {},
   "outputs": [],
   "source": [
    "columns =[ \n",
    "'Profile_mean',\n",
    "'Profile_stdev',\n",
    "'Profile_skewness',\n",
    "'Profile_kurtosis',\n",
    "'DM_mean',\n",
    "'DM_stdev',\n",
    "'DM_skewness',\n",
    "'DM_kurtosis',\n",
    "'label'\n",
    "]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "4d82b289",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "<class 'pandas.core.frame.DataFrame'>\n",
      "RangeIndex: 1790 entries, 0 to 1789\n",
      "Data columns (total 8 columns):\n",
      " #   Column            Non-Null Count  Dtype  \n",
      "---  ------            --------------  -----  \n",
      " 0   Profile_mean      1790 non-null   float64\n",
      " 1   Profile_stdev     1790 non-null   float64\n",
      " 2   Profile_skewness  1790 non-null   float64\n",
      " 3   Profile_kurtosis  1790 non-null   float64\n",
      " 4   DM_mean           1790 non-null   float64\n",
      " 5   DM_stdev          1790 non-null   float64\n",
      " 6   DM_skewness       1790 non-null   float64\n",
      " 7   DM_kurtosis       1790 non-null   float64\n",
      "dtypes: float64(8)\n",
      "memory usage: 112.0 KB\n"
     ]
    }
   ],
   "source": [
    "df = pd.read_csv('HTRU2/HTRU_2.csv',header=None,names=columns)\n",
    "df = df.sample(frac=0.1).reset_index(drop=True)\n",
    "labels = df.label\n",
    "df = df.drop(columns='label')\n",
    "df.info()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "ef8124c5",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>Profile_mean</th>\n",
       "      <th>Profile_stdev</th>\n",
       "      <th>Profile_skewness</th>\n",
       "      <th>Profile_kurtosis</th>\n",
       "      <th>DM_mean</th>\n",
       "      <th>DM_stdev</th>\n",
       "      <th>DM_skewness</th>\n",
       "      <th>DM_kurtosis</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>129.062500</td>\n",
       "      <td>47.583274</td>\n",
       "      <td>0.034719</td>\n",
       "      <td>-0.154337</td>\n",
       "      <td>5.727425</td>\n",
       "      <td>26.594803</td>\n",
       "      <td>5.353546</td>\n",
       "      <td>31.453808</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>102.320312</td>\n",
       "      <td>38.602933</td>\n",
       "      <td>0.377556</td>\n",
       "      <td>1.240759</td>\n",
       "      <td>2.185619</td>\n",
       "      <td>17.625631</td>\n",
       "      <td>9.530584</td>\n",
       "      <td>100.988835</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>118.000000</td>\n",
       "      <td>48.656160</td>\n",
       "      <td>0.055570</td>\n",
       "      <td>-0.078968</td>\n",
       "      <td>1.088629</td>\n",
       "      <td>12.336984</td>\n",
       "      <td>14.649034</td>\n",
       "      <td>246.822866</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>130.640625</td>\n",
       "      <td>51.872852</td>\n",
       "      <td>0.145932</td>\n",
       "      <td>0.177934</td>\n",
       "      <td>9.852843</td>\n",
       "      <td>35.871311</td>\n",
       "      <td>3.869401</td>\n",
       "      <td>14.883692</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>102.953125</td>\n",
       "      <td>34.386557</td>\n",
       "      <td>0.646091</td>\n",
       "      <td>2.766828</td>\n",
       "      <td>1.887124</td>\n",
       "      <td>15.579652</td>\n",
       "      <td>10.453287</td>\n",
       "      <td>131.059023</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   Profile_mean  Profile_stdev  Profile_skewness  Profile_kurtosis   DM_mean  \\\n",
       "0    129.062500      47.583274          0.034719         -0.154337  5.727425   \n",
       "1    102.320312      38.602933          0.377556          1.240759  2.185619   \n",
       "2    118.000000      48.656160          0.055570         -0.078968  1.088629   \n",
       "3    130.640625      51.872852          0.145932          0.177934  9.852843   \n",
       "4    102.953125      34.386557          0.646091          2.766828  1.887124   \n",
       "\n",
       "    DM_stdev  DM_skewness  DM_kurtosis  \n",
       "0  26.594803     5.353546    31.453808  \n",
       "1  17.625631     9.530584   100.988835  \n",
       "2  12.336984    14.649034   246.822866  \n",
       "3  35.871311     3.869401    14.883692  \n",
       "4  15.579652    10.453287   131.059023  "
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "01f67d5e",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[ 1.29062500e+02,  4.75832740e+01,  3.47186640e-02, ...,\n",
       "         2.65948033e+01,  5.35354616e+00,  3.14538075e+01],\n",
       "       [ 1.02320312e+02,  3.86029333e+01,  3.77556005e-01, ...,\n",
       "         1.76256311e+01,  9.53058409e+00,  1.00988835e+02],\n",
       "       [ 1.18000000e+02,  4.86561597e+01,  5.55701750e-02, ...,\n",
       "         1.23369840e+01,  1.46490345e+01,  2.46822866e+02],\n",
       "       ...,\n",
       "       [ 1.46156250e+02,  5.70882592e+01, -2.98125307e-01, ...,\n",
       "         1.57747493e+01,  8.80829989e+00,  9.82645936e+01],\n",
       "       [ 9.31015625e+01,  4.47480516e+01,  5.01497565e-01, ...,\n",
       "         2.32835289e+01,  7.46817115e+00,  5.74580838e+01],\n",
       "       [ 1.13398438e+02,  5.17784251e+01,  2.62187455e-01, ...,\n",
       "         1.80935576e+01,  1.04315859e+01,  1.13040611e+02]])"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df.values"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "11c988c6",
   "metadata": {},
   "outputs": [],
   "source": [
    "dissimilarity = pd.DataFrame(distance_matrix(df.values, df.values),index=df.index,columns=df.index)\n",
    "norm_diss = np.array(dissimilarity/dissimilarity.iloc[np.argmin(dissimilarity.sum())].sum())\n",
    "E = np.array(df)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "e9572654",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Sigma_0 3.0348542587702925 \n",
      "Sigma_f 1.5174271293851462\n"
     ]
    }
   ],
   "source": [
    "## Inicialização de parâmetros\n",
    "\n",
    "n_iter = 50\n",
    "q = 5\n",
    "n = 1.1\n",
    "x = 2\n",
    "y = 2\n",
    "C = int(x*y)\n",
    "N = E.shape[0]\n",
    "sigma_0 = np.sqrt(-(C+C)/2*np.log(0.1))\n",
    "sigma_f = np.sqrt(-1/2*np.log(0.01))\n",
    "print('Sigma_0', sigma_0, '\\nSigma_f', sigma_f)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "211222ab",
   "metadata": {},
   "outputs": [],
   "source": [
    "def calc_sigma(t):\n",
    "    return sigma_0*pow((sigma_f/sigma_0),t/50)\n",
    "\n",
    "def calc_h(nodes,s,r,sigma):\n",
    "    s=int(s)\n",
    "    r=int(r)\n",
    "    dist = delta[i,j]\n",
    "    return np.exp(-dist/2*np.square(sigma))\n",
    "\n",
    "def D_vr(e,r):\n",
    "    G_r = G[r]\n",
    "    return sum(\n",
    "        [np.power(v[r,i],n)*norm_diss[e,G_r[i]] for i in range(v.shape[1])]\n",
    "    )\n",
    "\n",
    "# def create_delta_matrix(x,y):\n",
    "nodes = []\n",
    "for i in range(1,x+1):\n",
    "    for j in range(1,y+1):\n",
    "        nodes.append([i,j])\n",
    "\n",
    "delta = np.empty((C,C))\n",
    "\n",
    "for i in range(delta.shape[0]):\n",
    "    for j in range(delta.shape[1]):\n",
    "        delta[i,j] = np.square(distance.euclidean(nodes[i],nodes[j]))\n",
    "\n",
    "# return delta"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "e3a9bc30",
   "metadata": {},
   "outputs": [],
   "source": [
    "t = 0\n",
    "sigma = calc_sigma(0)\n",
    "\n",
    "h_matrix = np.empty((C,C))\n",
    "\n",
    "for i in range(h_matrix.shape[0]):\n",
    "    for j in range(h_matrix.shape[1]):\n",
    "        h_matrix[i,j] = calc_h(nodes,i,j,sigma)\n",
    "        \n",
    "G = []\n",
    "for i in range(C):\n",
    "    G.append(np.array(df.sample(5).index))\n",
    "G=np.array(G)\n",
    "\n",
    "v = pd.DataFrame(np.random.rand(C,5))\n",
    "v_row_sum = np.array(v.sum(axis=1))\n",
    "v = np.array(v)\n",
    "\n",
    "for i in range(v.shape[0]):\n",
    "    for j in range(v.shape[1]):\n",
    "        v[i,j] = v[i,j]/v_row_sum[i]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "4289a3bd",
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU times: user 247 ms, sys: 0 ns, total: 247 ms\n",
      "Wall time: 247 ms\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "P = [[] for i in range(C)]\n",
    "clusters_array = np.zeros(N)\n",
    "for k in range(N):\n",
    "    #calculate f_ek\n",
    "    deltas = np.empty(C)\n",
    "    for s in range(C):\n",
    "        val = sum(\n",
    "            [\n",
    "                h_matrix[s,r]*D_vr(k,r) for r in range(C)\n",
    "            ]\n",
    "        )\n",
    "\n",
    "        deltas[s] = val\n",
    "\n",
    "    k_cluster = np.argmin(deltas)\n",
    "    clusters_array[k] = k_cluster\n",
    "    clusters_array = clusters_array.astype(int)\n",
    "    P[k_cluster].append(k)     "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "689cfdeb",
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "001c6baf19fd4f959e9cd16bec400a90",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "  0%|          | 0/149 [00:00<?, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Iter  1\n",
      "Iter  2\n",
      "Iter  3\n",
      "Iter  4\n",
      "Iter  5\n",
      "Iter  6\n",
      "Iter  7\n",
      "Iter  8\n",
      "Iter  9\n",
      "Iter  10\n",
      "Iter  11\n",
      "Iter  12\n",
      "Iter  13\n",
      "Iter  14\n",
      "Iter  15\n",
      "Iter  16\n",
      "Iter  17\n",
      "Iter  18\n",
      "Iter  19\n",
      "Iter  20\n",
      "Iter  21\n",
      "Iter  22\n",
      "Iter  23\n",
      "Iter  24\n",
      "Iter  25\n",
      "Iter  26\n",
      "Iter  27\n",
      "Iter  28\n",
      "Iter  29\n",
      "Iter  30\n",
      "Iter  31\n",
      "Iter  32\n",
      "Iter  33\n",
      "Iter  34\n",
      "Iter  35\n",
      "Iter  36\n",
      "Iter  37\n",
      "Iter  38\n",
      "Iter  39\n",
      "Iter  40\n",
      "Iter  41\n",
      "Iter  42\n",
      "Iter  43\n",
      "Iter  44\n",
      "Iter  45\n",
      "Iter  46\n",
      "Iter  47\n",
      "Iter  48\n",
      "Iter  49\n",
      "Iter  50\n",
      "Iter  51\n",
      "Iter  52\n",
      "Iter  53\n",
      "Iter  54\n",
      "Iter  55\n",
      "Iter  56\n",
      "Iter  57\n",
      "Iter  58\n",
      "Iter  59\n",
      "Iter  60\n",
      "Iter  61\n",
      "Iter  62\n",
      "Iter  63\n",
      "Iter  64\n",
      "Iter  65\n",
      "Iter  66\n",
      "Iter  67\n",
      "Iter  68\n",
      "Iter  69\n",
      "Iter  70\n",
      "Iter  71\n",
      "Iter  72\n",
      "Iter  73\n",
      "Iter  74\n",
      "Iter  75\n",
      "Iter  76\n",
      "Iter  77\n",
      "Iter  78\n",
      "Iter  79\n",
      "Iter  80\n",
      "Iter  81\n",
      "Iter  82\n",
      "Iter  83\n",
      "Iter  84\n",
      "Iter  85\n",
      "Iter  86\n",
      "Iter  87\n",
      "Iter  88\n",
      "Iter  89\n",
      "Iter  90\n",
      "Iter  91\n",
      "Iter  92\n",
      "Iter  93\n",
      "Iter  94\n",
      "Iter  95\n",
      "Iter  96\n",
      "Iter  97\n",
      "Iter  98\n",
      "Iter  99\n",
      "Iter  100\n",
      "Iter  101\n",
      "Iter  102\n",
      "Iter  103\n",
      "Iter  104\n",
      "Iter  105\n",
      "Iter  106\n",
      "Iter  107\n",
      "Iter  108\n",
      "Iter  109\n",
      "Iter  110\n",
      "Iter  111\n",
      "Iter  112\n",
      "Iter  113\n",
      "Iter  114\n",
      "Iter  115\n",
      "Iter  116\n",
      "Iter  117\n",
      "Iter  118\n",
      "Iter  119\n",
      "Iter  120\n",
      "Iter  121\n",
      "Iter  122\n",
      "Iter  123\n",
      "Iter  124\n",
      "Iter  125\n",
      "Iter  126\n",
      "Iter  127\n",
      "Iter  128\n",
      "Iter  129\n",
      "Iter  130\n",
      "Iter  131\n",
      "Iter  132\n",
      "Iter  133\n",
      "Iter  134\n",
      "Iter  135\n",
      "Iter  136\n",
      "Iter  137\n",
      "Iter  138\n",
      "Iter  139\n",
      "Iter  140\n",
      "Iter  141\n",
      "Iter  142\n",
      "Iter  143\n",
      "Iter  144\n",
      "Iter  145\n",
      "Iter  146\n",
      "Iter  147\n",
      "Iter  148\n",
      "Iter  149\n"
     ]
    }
   ],
   "source": [
    "# %%time\n",
    "v_t = v.copy()\n",
    "for t in tqdm(range(1,150)):#range(1,n_iter+1):\n",
    "    gc.collect()\n",
    "#     clear_output(wait=True)\n",
    "    print('Iter ',t)\n",
    "    sigma_t = calc_sigma(t)\n",
    "    h_t = np.empty((C,C))\n",
    "    \n",
    "    for i in range(C):\n",
    "        for j in range(C):\n",
    "            h_t[i,j] = calc_h(nodes,i,j,sigma_t)\n",
    "    \n",
    "    ## step 1: compute Gr\n",
    "    for r in range(C):\n",
    "        g = []\n",
    "\n",
    "        for h in range(N):\n",
    "            hfec = np.array([h_t[x,r] for x in clusters_array])\n",
    "            dist = norm_diss[h]\n",
    "            val = (dist*hfec).sum()\n",
    "\n",
    "            g.append((h,val))\n",
    "            \n",
    "        g = sorted(g, key= lambda x: x[1])[:q]\n",
    "        \n",
    "        G[r] = np.array([pair[0] for pair in g])\n",
    "    \n",
    "    \n",
    "#     print('Iter ',t, '- Weighting')\n",
    "    ## step 2: weighting\n",
    "    for r in range(v_t.shape[0]): #for all lines of v\n",
    "        for e_idx in range(v_t.shape[1]): #for all columns of v\n",
    "            base_sum = []\n",
    "\n",
    "            upper = sum([h_t[clusters_array[k],r]*norm_diss[k,G[r,e_idx]] for k in range(N)])\n",
    "            \n",
    "            for element in G[r]:\n",
    "                lower = np.sum(\n",
    "                    [h_t[clusters_array[k],r]*norm_diss[k,element] for k in range(N)]\n",
    "                )\n",
    "\n",
    "                base_sum.append(\n",
    "                    pow(upper/lower,1/(n-1))\n",
    "                )\n",
    "            \n",
    "            v_t[r][e_idx] = 1/sum(base_sum)\n",
    "            \n",
    "#     print('Iter ',t, '- Assignment')\n",
    "    ## step 3: assignment\n",
    "    P = [[] for i in range(C)]\n",
    "#     elements_clusters = defaultdict()\n",
    "    \n",
    "    for k in range(N):\n",
    "        #calculate f_ek\n",
    "        deltas = np.empty(C)\n",
    "        for s in range(C):          \n",
    "            val = sum(\n",
    "                [\n",
    "                    h_t[s,r]*D_vr(k,r) for r in range(C)\n",
    "                ]\n",
    "            )\n",
    "\n",
    "            deltas[s] = val\n",
    "\n",
    "        k_cluster = np.argmin(deltas)\n",
    "#         elements_clusters[k] = {'cluster': k_cluster, 'element': E[k]}\n",
    "        clusters_array[k] = k_cluster\n",
    "\n",
    "        P[k_cluster].append(k)     "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "18e28b61",
   "metadata": {},
   "outputs": [],
   "source": [
    "results = defaultdict()\n",
    "for idx,r in enumerate(P):\n",
    "    r_labels = [labels.iloc[val] for val in r]\n",
    "    positives = sum(r_labels)\n",
    "    negatives = abs(len(r_labels) - positives)\n",
    "    \n",
    "    results[idx] = {\n",
    "        'positives': positives,\n",
    "        'negatives': negatives\n",
    "    }"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "5a2bc154",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAxkAAAMWCAYAAACdtUsqAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjUuMywgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy/NK7nSAAAACXBIWXMAAA9hAAAPYQGoP6dpAABJ30lEQVR4nO3dd7jedWH//9d9VvYgJIFAEmZAhkEBAwgqDnDh3qNoHThbKmqH31r5WautijiqtM4i4h64RURAEdk77BkIgUyyk5Mzfn/cOFBGxuc+73s8Htd1rkMoufMCe5L7eT6rNjw8PBwAAICKdJUeAAAAtBeRAQAAVEpkAAAAlRIZAABApUQGAABQKZEBAABUSmQAAACVEhkAAEClRAYAAFApkQEAAFRKZAAAAJUSGQAAQKVEBgAAUCmRAQAAVEpkAAAAlRIZAABApUQGAABQKZEBAABUSmQAAACVEhkAAEClRAYAAFApkQEAAFRKZAAAAJUSGQAAQKVEBgAAUCmRAQAAVEpkAAAAlRIZAABApUQGAABQKZEBAABUSmQAAACVEhkAAEClRAYAAFApkQEAAFRKZAAAAJUSGQAAQKVEBgAAUCmRAQAAVEpkAAAAlRIZAABApUQGAABQKZEBAABUSmQAAACVEhkAACNloD8ZHi69Ahqup/QAAICWs255svKuZO3SZN2yBz4v/dOPN65KNq5J+tck/Wv/9NfDg/Wf392X9IxOekZtxucxydgpyXa7JpN3SbbbJZk8O+kdU/Q/ATyS2vCwnAYAeEir70uW3JAsufHBn9ctLTysloyf/mfR8RefJ85Mun0vmXJEBgDAyoV/HRNLb0zWryi9bOvUuuuxMfMJyaxD6h/T9026nCnPyBAZAEDnWXpLcts5ye3nJXec37oxsSVGTUpmHvSn6Jj5hGTU+NKraFMiAwBof2uWJLedW/+4/bz69RSdrtad7LDvA9FxaDJrXv3oB1RAZAAA7ad/bXLH7/4UFouvS+Itz6OauHOy1zOTxzw32e0pSXdv6UW0KJEBALSHFXck13wnueXXyd2XJEObSi9qbaMmJXOOSvY5JtnzKKdWsUVEBgDQujauTuafkVz1jeTOC+JoRYP0jE72fEay/0uSvZ/t9rk8KpEBALSWoaH6dRVXfSO5/sfJpnWlF3WWvgn106ke+9Jk96e6VS4PSWQAAK1h6S3JlacnV387WXV36TUkydip9diYd1yy/R6l19BERAYA0LzW359c+736UYu7Lym9hodT60r2elZy2DuSXY8ovYYmIDIAgOZz3/zkd59O5v8gGdxYeg1bYsbj6rGx34vcnaqDiQwAoHnceUFy/snJzb8svYRtNWGnZN6bk4P/NhmzXek1jDCRAQCUNTyc3Pjz5HefTO66qPQaqtY7Lnncq5ND3+a6jQ4iMgCAMoaHk+t+mJz3Xw88LI+29sfrNt6Z7Hp46TU0mMgAAEbe9T9Jzv3P5L5rSi+hhD2eljzzw8n0fUovoUFEBgAwcm78eXLuR5JFV5VeQmm17uTAY5On/WsybmrpNVRMZAAAjbfo6uSn707uvrj0EprNqInJk95dv2ajZ1TpNVREZAAAjbNxdfLr/0gu/nwyPFh6Dc1su12TZ/x/yX4vLL2ECogMAKAxrv1+cub7ktWLSi+hlcx+YvLM/0h2PrD0EraByAAAqrXs1uRn70lu/XXpJbSsWnLAK5On/1sycafSY9gKIgMAqMbAxvqD9M4/ORnYUHoN7aB3bHL4PyRPOsHTw1uMyAAAtt2tv05++p5k+a2ll9COZjwuefEXkml7lV7CZhIZAMDWW7UoOfNfkvk/KL2EdtczJjn635N5by69hM0gMgCArXPVN5OfvTfZuKr0EjrJnkclL/hsMmGH0kt4BCIDANgym9bX4+KK00ovoVON3T553qeTfY4pvYSHITIAgM239Jbk28cmi+eXXgLJ41+bPOu/klHjSy/hL4gMAGDzXPPd5MfHJ/1rSi+BP9lut+TFn09mzSu9hD8jMgCARzawMfnFPyeXfrn0Enhote7kSe9OnvJPSXdP6TVEZAAAj2T5bcm3X5fce3XpJfDodj44ecVpHuDXBEQGAPDQ5p+R/Ojv3D2K1jJ+x+SVX09mHlR6SUcTGQDAgw30J7/8f8nFny+9BLZOz+j6bW4f+9LSSzqWyAAA/mTN4uQbr0wWXlZ6CWy7J707edr7k1qt9JKOIzIAgLoVdySnvah+HQa0i8cck7z4C0nf2NJLOorIAACS++Ynp704WXNv6SVQvZ0PSl797WTc1NJLOobIAIBOt+DC5OsvTzasLL0EGmfK7slrvptsv0fpJR1BZABAJ7vpl/UneA+sL70EGm/s1PoRDXeeariu0gMAgEKu+lbyzVcJDDrHuqXJqcckN/689JK2JzIAoBNdeEryg7ckQwOll8DI2rQu+eZrkmu+W3pJW3O6FAB0mrP/Pfntx0uvgLK6epKXnZrsc0zpJW1JZABApxgaSn56QnLZV0ovgebQ3Ze88hvJnGeUXtJ2RAYAdILBgeT7b0rm/6D0EmguPWOS13w72e3JpZe0FddkAEC7Gx5OfvROgQEPZWB98o1XJQsuKr2krYgMAGh3Z74vueobpVdA8+pfk5z+suSeK0ovaRsiAwDa2XkfSy78XOkV0Pw2rkxOe1Fy3/zSS9qCyACAdnXJF5NzPlR6BbSO9SuSr74wWXpz6SUtT2QAQDu69vvJz95begW0nrWLk1Ofn6y4o/SSliYyAKDd3Pn75AdvTYaHSi+B1rT6nuTU5yUrF5Ze0rJEBgC0k6W3JN98dTK4sfQSaG33L0i+/opk0/rSS1qSyACAdrF2WXL6S5P1y0svgfZw3zXJj/6+9IqWJDIAoB1s2pB845XJittLL4H2cs23kwtPKb2i5YgMAGgHP3hLcvfFpVdAe/rlvyZ3/K70ipYiMgCg1f3+c8l1Z5ReAe1raCD5zuuTVYtKL2kZIgMAWtk9Vya/+kDpFdD+1i5Ovv03yUB/6SUtQWQAQKvauDr57t8mg970wIi4+5Lk5/9YekVLEBkA0Kp+8q5k+W2lV0BnuewryeWnlV7R9EQGALSiK76WXPOd0iugM/3sPcnCy0uvaGoiAwBazZKbkp85ZQOKGdiQfPvY+rNpeEgiAwBayaYNyXffkGxaW3oJdLaVd9WviRoeLr2kKYkMAGglv/x/9acQA+Xdfl5y6ZdLr2hKIgMAWsX1P04u+WLpFcCf+9WJyap7Sq9oOiIDAFrB/XclP3xn6RXAX9q4KvnJCaVXNB2RAQDNbmgo+d6bkg33l14CPJSbfp5c+73SK5qKyACAZnf5qcldF5ZeATySn/9Tsm556RVNQ2QAQDNbtzw5+4OlVwCPZu2S5Mz/V3pF0xAZANDMfv3vyXrfHYWWcNXXk1vOLr2iKYgMAGhW91yZXPZ/pVcAW+In/5D0e46NyACAZjQ8nPzsPcnwUOklwJa4f0Hy6w+VXlGcyACAZnTl6cndl5ReAWyNi/4nufuy0iuKEhkA0Gw2rKw/4AtoTcNDyY/emQxuKr2kGJEBAM3mnA/X71QDtK7F1yWXfLH0imJEBgA0k/vmJxd/ofQKoAq/+XiycXXpFUWIDABoJj99TzI8WHoFUIV1S5MLPlN6RREiAwCaxdXfThZcUHoFUKXffzZZ03mnP4oMAGgGA/0u9oZ21L8m+c1HS68YcSIDAJrBVd9IVi0svQJohEu/kiy/vfSKESUyAKC0oaHkgk+XXgE0ytCm5LzOOpohMgCgtOt/mCy7pfQKoJGu+XZHHc0QGQBQ2vknl14ANNrQQPLbj5deMWJEBgCUdMvZyaKrSq8ARsJV30pW3Fl6xYgQGQBQkqMY0DmGNiXnf6L0ihEhMgCglLsvTe74bekVwEi68uvJ/XeVXtFwIgMASnEUAzrPYH9y8f+WXtFwIgMASlhyY3LDT0uvAEq48hv1B3C2MZEBACWc/8kkw6VXACWsW5rc8OPSKxpKZADASLv/ruSa75ReAZR02amlFzSUyACAkXbhKfW7zACd6/bfJMtvK72iYUQGAIykgf7kqm+UXgEUN9zWRzNEBgCMpJt+kaxfXnoF0Ayu/Hoy2J5HNUUGAIykK08vvQBoFmsXJzf+rPSKhhAZADBS1ixObvlV6RVAM7ns/0ovaAiRAQAj5epvJUMDpVcAzeTWc5IVd5ZeUTmRAQAj5UoXfAN/aTi5/KulR1ROZADASLhvfrJ4fukVQDO68vRksL2OcooMABgJ136/9AKgWa1elNz8y9IrKiUyAGAkzBcZwCO44aelF1RKZABAo91zZVs/2ReowC2/SoaHS6+ojMgAgEZzFAN4NGvuTRZdVXpFZUQGADTa/B+UXgC0gpvPKr2gMiIDABpp0VXJ/QtKrwBaQRtd/C0yAKCRbv9N6QVAq1h4abJueekVlRAZANBIt/+29AKgVQwPJbecXXpFJUQGADTK0GCy4MLSK4BW0ianTIkMAGiURVclG1eWXgG0klvPToaGSq/YZiIDABrlDqdKAVto3bJk4WWlV2wzkQEAjXLH+aUXAK2oDU6ZEhkA0AhDg8mdvy+9AmhFIgMAeEj3XJn0ry69AmhFi65K1iwpvWKbiAwAaATXYwBbbTi554rSI7aJyACARhAZwLa49+rSC7aJyACAqg0OJAsuKr0CaGX3XVt6wTYRGQBQtUVXuh4D2Db3XlN6wTYRGQBQtQXuKgVso+W3Jf1rS6/YaiIDAKq25IbSC4BWNzyU3De/9IqtJjIAoGpLbym9AGgHLXzxt8gAgKotExlABVr4ugyRAQBVWr8iWbe09AqgHdzbuneYEhkAUCWnSgFVWXxdMjRUesVWERkAUCWnSgFV2bSuZX9PERkAUKVlN5deALSTFr34W2QAQJVa9LuOQJNq0Yu/RQYAVMk1GUCVlt9WesFWERkAUJXh4WT5raVXAO1k9aLSC7aKyACAqqy8KxnYUHoF0E5WiQwA6GxLXfQNVGzNffWjpC1GZABAVZY5VQqo2NCmZO2S0iu2mMgAgKq04BsBoAW04HUZPaUHAEDb2Li69AKgTQx3j8rg2KnpHzUl/WvWZXLpQVtIZABAVUQG8AiGR03IwOhp2TBqStb2bpeVXZOzIpOzeHhiFm2akLs3jcsdG8bllnVjs2htX7K2/vM+ev/MvLzs9C0mMgCgKv0iAzrJcK07w2OmpH/09lnft33W9EzO/bXJWZZJuW9wYu7ZND539Y/P7evH5JZ1Y7J6Q0+ycst/nWVr+7dp55FHHpn9998/SXLaaaelt7c3b3vb2/LBD34wtVptm1774YgMAKiKIxnQ8oZ7RmdwzNRsHLV91vVNyequyVlRm5wlwxNz7+DELOwflzs3js9t68bk9vWjMri+8Zc4r1i3bZGRJKeeemre+MY35uKLL86ll16a4447LrNnz86b3/zmChb+NZEBAFURGdCUhkdNyqYxU7Ohb0rW9k7Jyq7JWZ5JWTw8MfdsmpC7+8fnjg1jc8u6sVm8pjdZU3rxg61av2mbX2PWrFk5+eSTU6vVsvfee+eaa67JySefLDIAoOltbLJ3JtCmhrt6MjRmSjaNnpp1vVOypme73F+bnKWZ+MfTlBb0j8tt68bmtnVjs3ZD11adptQs1mwc2ObXOPTQQx90atRhhx2Wk046KYODg+nu7t7m1/9LIgMAquJIBmy14d6xD5ymNCVre7fPqu7JWZFJWTo8MYsG60cb7twwLretG5M71o/O8LrGXEvQjNZWEBkjTWQAQFVc+A1/NJxahkdPzsCYqVnfNyVre+p3U1qWSVk8VD9N6a5N9Yuib103NstW9ya+hB7Suv7BbX6Niy666EE/vvDCCzNnzpyGHMVIRAYAVMfpUrS54a7eDI2dmv4/XBTdXb+b0tLhiblvaGIW9o/Pgo3jc9v6Mbl9/eis39Cd3F96devbMDC0za+xYMGCnHDCCXnLW96Syy+/PJ/5zGdy0kknVbDuoYkMAKhC/9pkeNu/2wgjbbhvXAbGTMvGBy6KXtU9OcszOUuGJ+TegYm5q39c7twwLreuG5O714/qqNOUmsXGTdv+e8uxxx6b9evXZ968eenu7s7xxx+f4447roJ1D01kAEAVHMWgSQzXujI8ertsGj01G/q2y+reKVlZm5RlmZzFgxNzz8D4LOgfn9vXjc0t68dk5aqeZFXp1TyS/sFtP5LR29ubT37ykznllFMqWPToRAYAVMFF3zTQcHdfBsdOS/+oKVnfOyWrurfLij+cpjQ4IQs3jc+dGx+4m9L6Mdm03tGGdjI8XHrBlhMZAFCFgfWlF9BihkdNyMDoadkwakrW9tYvil6RyVn8F6cp3bJuTO5ZOypZW3oxpQy1YGWIDACoQndf6QW0mNrG1enduDq9uS0Tkuz4SP/w6BEaRXMatUeSy7f6p5977rmVTdlcjX8OOgB0ApEBNEpX6x0XEBkAUIUe32oGGqQFI6P1FgNABY488sjMnTs3o0ePzhe/+MX09fXlrW99a0488cSte8GeUZXuA/ijrsY8MK+RHMkAoGOdeuqpGTduXC666KJ89KMfzQc/+MGcddZZW/diTpcCGqUFj2SIDAA61ty5c/OBD3wgc+bMybHHHpuDDz44Z5999ta9mCMZQKOIDABoHXPnzn3Qj2fMmJHFixdv3Yt19yZdvRWsAvgL3a33e4vIAKBj9fY++A/uWq2WoaFteLLu6InbuAjgIYzZrvSCLSYyAKAqo0QG0ADjdyi9YIuJDACoiiMZQCNMeMRHNTYlkQEAVRk9qfQCoB214JGM1rtUHQAqcO655/7V3zvjjDO27UWdLgU0giMZANDBRk8uvQBoRy14JENkAEBVJs4ovQBoR45kAEAHm7J76QVAu6l1JeOmlV6xxUQGAFRFZABVGzct6eouvWKLiQwAqMqUPUovANpNC16PkYgMAKjO+GnuMAVUqwWvx0hEBgBUa7tdSy8A2okjGQCA6zKASjmSAQBke9dlABWaNKv0gq0iMgCgSo5kAFWaMbf0gq0iMgCgSiIDqEpXbzJ9v9IrtorIAIAquY0tUJXpj0l6+kqv2CoiAwCqNGGHpG986RVAO5hxQOkFW01kAEDVttut9AKgHcx4XOkFW01kAEDVpj+m9AKgHYgMAOCPZh9aegHQ6mrdyY77l16x1UQGAFRtl8NLLwBa3dS9kt4xpVdsNZEBAFWb9phk7PalVwCtrIUv+k5EBgBUr1ZLZh9WegXQykQGAPBXdnli6QVAK9vpcaUXbBORAQCN4EgGsNVqyY5zS4/YJiIDABphxgFJ34TSK4BWNHWvZFRrP9RTZABAI3R1J7PmlV4BtKI9n1F6wTYTGQDQKK7LALbG3s8qvWCbiQwAaBTPywC21KhJyezW/waFyACARtn5oKRndOkVQCvZ8+lJd0/pFdtMZABAo/T0JTsfXHoF0Er2av1TpRKRAQCNtduTSy8AWkWtO5lzVOkVlRAZANBI+72o9AKgVcyal4ydUnpFJUQGADTStL2SHR9begXQCtrkVKlEZABA4+3/0tILgFYgMgCAzbb/i5PUSq8Amtl2uybTH1N6RWVEBgA02uTZnv4NPLI2OoqRiAwAGBlOmQIeicgAALbYfi+s354S4C+NnpTsekTpFZUSGQAwEsZP98wM4KE99mVJd2/pFZUSGQAwUh7rlCngITz+b0ovqJzIAICRss/zku6+0iuAZrLj3GSnx5VeUTmRAQAjZfSkZM+jSq8AmsmBx5Ze0BAiAwBG0mNfUnoB0Cx6Rtevx2hDIgMARtLez0lGTy69AmgG+zw/GTO59IqGEBkAMJJ6xyQHvb70CqAZPOFNpRc0jMgAgJE277ikq6f0CqCkGY9LZh9SekXDiAwAGGmTdk72fWHpFUBJh7yl9IKGEhkAUMJhby+9AChl7NRk//a+CYTIAIASdj4omdW+p0oAj+Cg1yU9o0qvaCiRAQClHPaO0guAkdbV09YXfP+ByACAUh7zvGTqXqVXACPpsS9PJu5UekXDiQwAKKWrKzniXaVXACOluy956r+UXjEiRAYAlPTYlyeTZ5deAYyEg17fMV/vIgMASuruSQ4/vvQKoNF6xyVPfm/pFSNGZABAaY//m2T8jqVXAI10yFuS8dNLrxgxIgMASusZlRz+96VXAI0yenLHHbEUGQDQDJ7w5mTKHqVXAI1w+PHJmMmlV4wokQEAzaCnL3nmh0uvAKo2fofkkLeWXjHiRAYANIu9n5XseVTpFUCVnvzepG9s6RUjTmQAQDN51keSrt7SK4AqTN6lftvaDiQyAKCZTJ1TvwsN0PqO/JekuzO/aSAyAKDZPOWfknGdc6tLaEvT9knmvqL0imJEBgA0m9ETk2d8oPQKYFs86yNJV+e+1e7cf3MAaGaPe02y04GlVwBb48Bjkz2eWnpFUSIDAJpRrZY8+6NJaqWXAFti4s7J0f9RekVxIgMAmtWsJyQHvLL0CmBLHHNy/ZTHDicyAKCZPePEpG986RXA5pj7imSvZ5Ze0RREBgA0swk7Jk97f+kVwKMZNz151n+WXtE0RAYANLtD3uJJ4NDsnntSMnZK6RVNQ2QAQLOr1ZIXnuLZGdCs9n1Bsu/zS69oKiIDAFrB+GnJi06Ju01BkxkzJXnOSaVXNB2RAQCtYs9nJIe+vfQK4M89+7/q3wTgQUQGALSSZ5yY7PjY0iuAJNnr2cncl5de0ZREBgC0kp6+5CVfTnrHll4CnW3s1PozMXhIIgMAWs20vdwqE0rq6kle/tVk4ozSS5qWyACAVnTQ6+p3tAFG3jM/kux6eOkVTU1kAECret6nkokzS6+AzvL41yaHHFd6RdMTGQDQqsZsl7zkC0nNH+cwInY+OHnuJ0qvaAl+VwKAVrbLE5Onvb/0Cmh/43dIXvG1pGdU6SUtQWQAQKt70gnJQa8vvQLaV3df8vLTXOi9BUQGALSD534imfPM0iugPT37o8nsQ0qvaCkiAwDaQVd38rKvJDs9vvQSaC8HvT45+G9Lr2g5IgMA2kXfuOTV304m71J6CbSHWYcmz/5Y6RUtSWQAQDsZPz157ffqd54Ctt6EneoP3OvpK72kJYkMAGg3U+ckr/pm0jO69BJoTWO2S1773WTCDqWXtCyRAQDtaPahyYv+1zM0YEv1TUhe871kh/1KL2lpfucBgHa13wuToz9UegW0jp4xyau/mcw8qPSSlicyAKCdHfaO5JC3lV4Bza+7L3nFacmuR5Re0hZEBgC0u2d+ONn/JaVXQPOqdScv/kIy56jSS9qGyACAdtfVVX8DdcCrSi+B5lPrTl78+frphVRGZABAJ+jqTl7wueTA15VeAs2jqyd5yReTx7609JK2IzIAoFN0dSXP+1Qy77jSS6C8rp7kJV9K9n9x6SVtSWQAQCep1ZLnfCx54t+XXgLldPUmL/s/p0g1UG14eHi49AgAoIDzPpac4xa3dJie0clLv5w85rmll7Q1kQEAneyy/0t+ckIyPFh6CTTeuOnJK7+ezHpC6SVtT2QAQKe7/ifJ996YDGwovQQaZ/p+9QftTZ5deklHEBkAQHLnBck3XplsWFl6CVRvzjOTl34pGTWh9JKOITIAgLr75idfe2my+p7SS6A6h74jOfpD9burMWJEBgDwJ2uW1E+duv280ktg23T1Js/9eHLQ60sv6UgiAwB4sKGh5Jz/SH57UhJvE2hBoycnL/9qsvtTSi/pWCIDAHhoN52ZfP+4ZMP9pZfA5puyR/LqbydT9yy9pKOJDADg4a24M/n2scmiK0svgUe365PqRzDGTim9pOOJDADgkQ1sTH7+T8llXym9BB5GLTnkrcnR/55095YeQ0QGALC5rvpm8pN3JZvWlV4CfzJxZvLCzya7H1l6CX9GZAAAm+++65Jv/02y7JbSSyCZ+8rkOR9NRk8qvYS/IDIAgC2zcXXyw3ck1/2w9BI61djtk2M+mez7/NJLeBgiAwDYOld9Kznzfcm6paWX0En2enby/E8n46eXXsIjEBkAwNZbtzw569+SK74Wz9SgofomJM/6cHLgsaWXsBlEBgCw7e74Xf2i8KU3ll5CO9rl8OSFpyTb7VJ6CZtJZAAA1RjoT373yfqTwgc2lF5DO+gelTz9/cmh70i6ukqvYQuIDACgWsturR/VuP280ktoZY85Jjnqg8n2e5RewlYQGQBAY1z1zQcuDF9WegmtZOeDkqP/I9nlsNJL2AYiAwBonHXLk7Pen1xxelwYziOavEvy9H9L9n9JUquVXsM2EhkAQOPddXFyzn8kt51begnNZvTk5MnvSea9JenpK72GiogMAGDkLLgoOfcjyW3nlF5Cad19yRPelDz5vcnYKaXXUDGRAQCMvLsursfGrb8uvYQS9n1B8owTkym7l15Cg4gMAKCcuy55IDbOLr2EkbDLEckzPpDMmld6CQ0mMgCA8u6+tB4bt/yq9BKqVutO9n1+8sS/q985io4gMgCA5nH3ZQ/Exlmll7Ct+sYnj/+b5NC3eVJ3BxIZAEDzueeK5JIvJtd+P9m0rvQatsTk2cnBb0gOen0yZrvSayhEZAAAzWvDquTqbyWX/V9y37Wl1/Bwal3JHk+v3y1qztFJV1fpRRQmMgCA1nDXJcllX0nmn5FsWlt6DUkyZkry+NfWj1xM2a30GpqIyAAAWkv/2uT6nyRXf7P+cL/hodKLOsuY7ZK9n5Ps8/xkj6cmPaNKL6IJiQwAoHWtvje55jvJVd90OlUjjd8hecwxyT7PS3Z9UtLdU3oRTU5kAADtYenNyS1n15+5ccfvnFK1rSbNrkfFPs9LZh3iOgu2iMgAANrPQH+y4Pf1J4rf+uvk3muSeMvzqLbfs34a1L7PT3Z6fOk1tDCRAQC0vzVLktvOqR/puO2cZM19pRc1gVoydU4y8wnJzIOTXQ5Ppu1dehRtQmQAAJ1leLh+/catv07uOD+5b36yamHpVY03enI9Jv4QFTsfnIyZXHoVbUpkAABsWJksvr4eHIuvTxZfV/9Yv6L0sq1T60522PeBoHjgY/s9k1qt9DI6hMgAAHg4q+/96/BYcmMTPIW8lkzYMZk0M5k0K5k8q/550qz635uyW9I3rvBGOpnIAADYUhtWJeuXJ+uWP/B5RbJu2V/8vT/7vG75A3e7qtWPJtS6HvjrrvqPu/uS3jH1j54HPveOrX+esOOfhcQDUTFx56Snr/R/BXhYIgMAAKiUGx4DAACVEhkAAEClRAYAAFApkQEAAFRKZAAAAJUSGQAAQKVEBgAAUCmRAQAAVEpkAAAAlRIZAABApUQGAABQKZEBAABUSmQAAACVEhkAAEClRAYAAFApkQEAAFRKZAAAAJUSGQAAQKVEBgAAUCmRAQAAVEpkAAAAlRIZAABApUQGAABQKZEBAABUSmQAAACVEhkAAEClRAYAAFApkQEAAFRKZAAAAJUSGQAAQKVEBgAAUCmRAQAAVEpkAAAAlRIZAABApUQGAABQKZEBAABUSmQAAACVEhkAAEClRAYAAFApkQEAAFRKZAAAAJUSGQAAQKVEBgAAUCmRAQAAVEpkAAAAlRIZAABApUQGAABQKZEBAABUSmQAAACVEhkAAEClRAYAAFApkQEAAFRKZAAAAJUSGQAAQKVEBgAAUCmRAQAAVEpkAAAAlRIZAABApUQGAABQKZEBAABUSmQAAACVEhkAAEClRAYAAFApkQEAAFRKZAAAAJUSGQAAQKVEBgAAUCmRAQAAVEpkAAAAlRIZAABApUQGAABQKZEBAABUSmQAAACVEhkAAEClRAYAAFApkQEAAFRKZAAAAJUSGQAAQKVEBgAAUCmRAQAAVEpkAAAAlRIZAABApUQGAABQKZEBAABUSmQAAACVEhkAAEClRAYAAFApkQEAAFRKZAAAAJUSGQAAQKVEBgAAUCmRAQAAVEpkAAAAlRIZAABApUQGAABQKZEBAABUSmQAAACVEhkAAEClRAYAAFApkQEAAFRKZAAAAJUSGQAAQKVEBgAAUCmRAQAAVEpkAAAAlRIZAABApUQGAABQKZEBAABUSmQAAACVEhkAAEClRAYAAFApkQEAAFRKZAAAAJUSGQAAQKVEBgAAUCmRAQAAVEpkAAAAlRIZAABApUQGAABQKZEBAABUSmQAAACVEhkAAEClRAYAAFApkQEAAFRKZAAAAJUSGQAAQKVEBgAAUCmRAQAAVEpkAAAAlRIZAABApUQGAABQKZEBAABUSmQAAACVEhkAAEClRAYAAFApkQEAAFRKZAAAAJUSGQAAQKVEBgAAUCmRAQAAVEpkAAAAlRIZAABApUQGAABQKZEBAABUSmQAAACVEhkAAEClRAYAAFApkQEAAFRKZAAAAJUSGQAAQKVEBgAAUCmRAQAAVEpkAAAAlRIZAABApUQGAABQKZEBAABUSmQAAACVEhkAAEClRAYAAFApkQEAAFRKZAAAAJUSGQAAQKVEBgAAUCmRAQAAVEpkAAAAlRIZAABApUQGAABQKZEBAABUSmQAAACVEhkAAEClRAYAAFApkQEAAFRKZAAAAJUSGQAAQKVEBgAAUKme0gNgm2zakKxdnKxdkqxZkqxfkQwNJMNDf/Ex/OAfd/cmY7dPxk9Pxk1Lxk1Pxk5JarXS/0YAAC1PZNDchoeTZbcm916d3H9ncv9dycq7kpV3JysXJhtXVvdrdfXUw2Pc9GT8tHp8TJ6d7LB/suNjkym7ixAAgM1QGx4eHi49Av5o+e3JPVf86WPR1dWGxLboG59M37ceHH/42GG/pHdM6WUAAE1FZFDOpg3J7b9JFvz+gaC4sn66UyupdSfb75HMOiTZ42nJ7kfWT7sCAOhgIoORtWZJctMvkht/ntx2TrJpXelF1ap1JTvOrQfHXs9MZs5LutxfAQDoLCKDxlt8Q3Ljz+phsfDS+oXXnWLs1Hps7P3senj0jSu9CACg4UQGjXHfdcmVp9fjYvltpdc0h95xyf4vSg58fTLrCaXXAAA0jMigOoObkut+mFzypWTBBaXXNLfp+yYHvi454BXJmO1KrwEAqJTIYNvdf1dy2VeSy0+rP7OCzdczOtnn+clBr0t2PaL0GgCASogMts7wcHLr2fWjFjedmQwPll7U+rbfs3504+C/TUZNKL0GAGCriQy2zOCm5IqvJRd82rUWjTJmSnL43yfz3pL0jS29BgBgi4kMNs/QYHLVN5LzPlp/8jaNN25acsS7koPfmPSOLr0GAGCziQwe2fBwcu33knM+nCy/tfSazjRhRnLECclBr096+kqvAQB4VCKDh3f7b5Oz3l9/GjflTZyZPPk9yeNfm3T3ll4DAPCwRAZ/bcmNyVkfSG76eeklPJQpuyfHnJzsfmTpJQAAD0lk8CcD/clvPpqcf3IyNFB6DY/mgFcnz/yPZOyU0ksAAB5EZFC38PLkh+9IFl9XeglbYty05Fn/mTz2paWXAAD8kcjodAMbk3M/kvzu05510crmHJ089xPJ5FmllwAAiIyOdvelyRlvT5beWHoJVegbnzztX+vP1+jqKr0GAOhgIqMTbdqQnPOh5Pefc/SiHe18cPLizyfb71F6CQDQoURGp1l0dfLdNyTLbi69hEYaNTF5/qeT/V5UegkA0IFERie55rvJD9+ZDKwvvYSRMu+45Oj/8BA/AGBEiYxOMDSUnH1i8rtPlV5CCTsdmLzitGTSzNJLAIAOITLa3fr7k++9MbnlV6WXUNK4acnLv5rs8sTSSwCADiAy2tniG5JvvjpZfmvpJTSDrt7kWR9J5r259BIAoM2JjHZ1w0+T778l6V9degnN5sDX1Z+p0d1TegkA0KZERjs676PJOR9O4n9aHsZjjkle+hUXhAMADSEy2s0v3pdc+NnSK2gFezw9eeXpSe+Y0ksAgDYjMtrJz/4xufh/S6+glexyePLqbyWjJpReAgC0EZHRDoaHk5+ekFz65dJLaEU7HZi89nvJ2CmllwAAbUJktLrh4eTHf59c/tXSS2hl0/dLjj0jGT+99BIAoA2IjFY2NJT86J3JlaeXXkI72H7P5NgfemgfALDNREarGhpMznhbcvW3Si+hnUyanbzh50IDANgmXaUHsBWGhpLvHycwqN7KBcnpL082rCq9BABoYSKjFZ35vuTa75ZeQbtaPD/59rHJ4EDpJQBAixIZrebC/0kuOqX0CtrdbeckP/mH0isAgBYlMlrJDT9NzvyX0ivoFFeclvzmY6VXAAAtyIXfreKeK5OvPDvZtK70EjpKLXnxF5K5Lys9BABoIY5ktII1i5NvvkZgUMBw8sO3J3f8rvQQAKCFiIxmN9CffOtvklV3l15CpxrsT771mmTpzaWXAAAtQmQ0u5+9J7nrwtIr6HTrVyTfem2yaX3pJQBACxAZzeyKryWXn1p6BdQtuSH55ftLrwAAWoDIaFYr7kh+/s+lV8CDXfKF5KYzS68AAJqcyGhGQ0PJGW9P+leXXgJ/7YfvqN+MAADgYYiMZvT7/07udDcfmtTaJfUIdvdrAOBhiIxms/j65NcfKr0CHtktZyUX/W/pFQBAkxIZzWRwU/L945LBjaWXwKP71QeS+64rvQIAaEIio5mc+5/JvVeXXgGbZ2BD8r03JZs2lF4CADQZkdEs7rokOf/k0itgyyyen/zmY6VXALS1I488Mu985zvzzne+M5MmTcrUqVPz/ve/P8OujaOJiYxmMNCfnPHWZHiw9BLYchd8Jll2a+kVAG3t1FNPTU9PTy6++OJ86lOfyic+8Yl88YtfLD0LHlZtWAaXd8F/J7/8f6VXwNbb86jktd8tvQKgLR155JFZvHhx5s+fn1qtliT553/+5/zoRz/Kdde5No7m5EhGaetXON2E1nfLWckNPyu9AqBtHXrooX8MjCQ57LDDcvPNN2dw0FkQNCeRUdp5H0s23F96BWy7X/xzMuDOaACAyChr+W3JJV8ovQKqcf+dnp0B0CAXXXTRg3584YUXZs6cOenu7i60CB6ZyCjpVycmg/2lV0B1fvvxZN3y0isA2s6CBQtywgkn5MYbb8w3vvGNfOYzn8nxxx9fehY8rJ7SAzrWXRcn1/2w9Aqo1oaVyXn/lTz7v0ovAWgaQ0PDWdM/kI2bhrJh02A2Dgxl48BgNmwaSlctmTSmN5PH9mXSmN50d9Ue8jWOPfbYrF+/PvPmzUt3d3eOP/74HHfccSP8bwKbT2SUcqa7SdGmLvlScshbkim7l14CMOL6B4Zy472rc+09K3PtwpW59p5VuWHRqmwcGHrUn1urJduN7cvsKWOz+7Rx2WuHCdlnxsQMDg2nt7c3n/zkJ3PKKaeMwL8FbDuRUcL8HyR3X1x6BTTG0Kbkd59OnvfJ0ksAGm7F2v78Yv69uXLB/blm4crcvHh1Ng1u3dMBhoeT5Wv7s3xtf6686/4//v377lyRe7rvyY4/vyFP3GP7HLbH9untdsY7zc1zMkba8HDy2UOSpTeWXgKN0zM6edf8ZNzU0ksAKrdpcCi/vmFxvnfZ3Tn3xiXpH3z0oxTb4t6v/3P6pu+eKc+onx41ZVxfjpk7Iy8+cGYeN2tyQ39t2FoiY6Td/Kvk9JeUXgGN95R/Sp76vtIrACoz/56V+c6ld+dHV92T5Wub48Yte0wblxcfODMvfPzO2XnymNJz4I9Exkj72kuSW35VegU03tjt60czev2hB7S2yxesyKfPvjnn3rik9JSH1d1Vy8sPnpV3H71Xpo4fVXoOiIwRteSm5LPzkvhPTod47knJE95UegXAVrliwYp84qyb8tubl5aestnGj+rJ247cI288YreM7vUMDcoRGSPpJyckl36p9AoYOVN2T955WdLlAkWgdSxevSH/+bMb8oMrF6ZV3yXtPHlM/vFZe+f5B+yUWu2hb4sLjSQyRsr6+5NP7JtsWlt6CYysl5+W7Pv80isAHtXw8HBOv2hB/usXN2T1hoHScyrxuFmT8/5j9slBu0wpPYUOIzJGyu8+lZz1b6VXwMibOS9501mlVwA8oqVrNuaEb1+V39zUvNddbIvXHbZL/vWYfd36lhEjMkbC0GDyqQOSlXeVXgJlvPGsZNa80isAHtIFtyzNP3zryixevbH0lIaat+uUfPY1B2baBBeG03hydiRc/2OBQWe78vTSCwAe0mfOvjmv/dJFbR8YSXLxHcvzvM+cnysWrCg9hQ4gMkbCpV8uvQDKuu5HyWB7nN8MtIehoeG87wfX5KSzbspQB53Tce+qDXnF5y/Mty5ZUHoKbU5kNNrq+5I7flt6BZS1fnly2zmlVwAkSTYODObtp1+er1/UmW+0+weG8k/fuyYfO/OG0lNoYyKj0a47IxkeKr0Cyrv2e6UXAGT1hk153Zcvzi/m31t6SnGfPefWfOTn15eeQZvqKT2g7XljBXU3/DQZ2Jj0uOAQKGPZmo35my9dnOsWrSo9pWn873m3ZWBwOO8/Zt/SU2gzjmQ00v13JXddXHoFNIeNq5Kbf1l6BdChBgaH8rbTLxcYD+FL59+ej595Y+kZtBmR0UjX/zhJB11NBo/m2u+XXgB0qA/99PpcfPvy0jOa1n+fc0u+cXFnXqNCY4iMRrrxZ6UXQHO56RdJv6feAyPrB1fcnf+74I7SM5re+8+4NhfcurT0DNqEyGiUdcuTOy8ovQKay6Z1yY0/L70C6CDXLlyZf/n+NaVntISBoeG84/TLs2jl+tJTaAMio1Fu/mUyPFh6BTQfkQGMkBVr+/PWr12WDZvc5XFzrVi3KSd866oMddLDQ2gIkdEoTpWCh7bg96UXAB1gcGg4f/eNK3L3Ct+V31K/v21ZPv/b20rPoMWJjEZxqhQ8tFULkxV3ll4BtLnPnnNLzr/F9QVb6xO/vCnXLlxZegYtTGQ0wvLbk7VLSq+A5uVoBtBAi1auzynn3lp6RkvrHxzKv55xbYaHnTbF1hEZjXD3JaUXQHO783elFwBt7GNn3pj1m1wXua2uvOv+/Oiqe0rPoEWJjEbwAD54ZHc6kgE0xrULV+YHVywsPaNtfPQXN2aDYGMriIxGuFtkwCNadnOyximFQPU+9NPr4gyf6iy8f32+dP7tpWfQgkRG1frXJffNL70Cmp/rMoCKnXXdfbnwNk/1rtrnzrklS1ZvLD2DFiMyqnbPFcnQQOkV0PxEBlCxT599c+kJbWlt/2A+e84tpWfQYkRG1ZwqBZvHbZ6BCl16x/Jc45arDfODKxamf8BDDdl8IqNqd7mzFGyWxdcnQ/7AAqrxlQvuKD2hra1cvylnX39f6Rm0EJFRtXuuKL0AWsPgxuT+O0qvANrAopXrc+a195ae0fa+d7m7drH5REaVNm1IVi8qvQJax1Ln+ALb7gdXLMzAkFtKNdp5Ny3OsjUuAGfziIwqrbw7id/kYLMtvan0AqANnHm177CPhE2Dwx7Ox2YTGVW6/87SC6AlDNe6MzBxdpas3VR6CtDqlt2aH6w7Nufs+e0cP+u2jOt2rVcjfd8pU2ymntID2srKu0ovgKYx3N2XTRNmZdWYWbmvZ6fcOTw912+clivWTsllK8dn/eLuHH7n9jm99FCgtd34s3RtuD+73X1G3pUz8g/jJ2bBtCPzo/6D8z/37Ja1A92lF7aVaxauzE33rc5eO0woPYUmJzKqdP+C0gtgRA33jsvGCbNz/+iZubd7Rm4b2iHXb9g+l63ZLletGpfBtY98sPSu5etHaCnQtm7+5YN+WNu4Krvc/aP8XX6Ud44dn7unPyU/3jQvpyzcLasHvO2pws+uWSQyeFS+2qp0vyMZtJ/hUZOyfsIuWTFq59zTNSO3Du6Q+Rum5JJV2+WG1WOT1Vv/2otWrs/Q0HC6umqb/XOOPPLI7L///kmS0047Lb29vXnb296WD37wg6nVNv91gDYwNJQsvPxh/8+1/jWZdfdP8/b8NG8bOy73THtyfjo4L6cs3CMrNnkLtLWuvtvzSHh0vsKq5EgGLWpo7NSsG79LlvXtnIW1HXPzwPRcs377XLJyUu5cOTpp0J8nmwaHs3j1xuw4afQW/bxTTz01b3zjG3PxxRfn0ksvzXHHHZfZs2fnzW9+c2OGAs1pyfVJ/5rN+kdr/Wuz88Kf57j8PG8ePS73zj4iPxual88t3DPL+nsbPLS9XOuhh2wGkVElkUGTGk4tQ+NnZM24WVnaNzN3ZYfctGlarlq7fS5ZNSmLl/cmy8tsW7Z2yyNj1qxZOfnkk1Or1bL33nvnmmuuycknnywyoNPcfelW/bTaprWZsfDMvDFn5g19Y3LfzCPyi+FD87l79szijYLj0SxevTFLVm/MtAmjSk+hiYmMqgxuStZ4EBDlDHf1ZGDCzKweMzOLe3fOguEdcmP/1FyxdkouWTkxq5f2JEtLr/xrK7biDlOHHnrog06NOuyww3LSSSdlcHAw3d0u8oSOsXDrIuPP1QbWZ8d7zsrrc1Ze1zs6S3Z+Yn6Zw/LfC+fk3o19FYxsT9feszJP3Xt66Rk0MZFRlTWLk2G3zaOxhrtHZdPE2Vk5embu7dkpdwztkBv6p+aKNdvlspUTsnFd692VetlaD3YCttKiqyp9udrAhky/59d5bX6d1/T0ZdlOT8xZtcPy3/fslYUbfNf+z81fKDJ4ZCKjKgMbSi+gTQz3jcuGCbtk5eiZWdQ1I7cNTc91G6bm8tWTc/XqR79jU6tZtX7Lj2RcdNFFD/rxhRdemDlz5jiKAZ2mgTdcqQ32Z+qic/OqnJtXdvVm+W5PzNm1Q/Pfi/bOgvVbdopnO7p+0Tbc9YOOIDKqMuihYmy+odGTs378LlkxemYW1mbk1sFpuXb99rl01Xa5adWYZFXphSNn1YaBLf45CxYsyAknnJC3vOUtufzyy/OZz3wmJ510UgPWAU1roD9ZPzIXk9WGNmX7Refl5TkvL+vqyYpdD8053Yfls4v2yW3rOjM4Fq10C3IemcioyqBTPniwwXHTs27c7Czr2zl313bMLQPTc/W6Kbl45eTcff+o5P7SC5vD6q2IjGOPPTbr16/PvHnz0t3dneOPPz7HHXdcA9YBTWv1oiK/bG1oIFPuPT8vyfl5cVdPVu46L+d2PzGfvXef3Lx2TJFNJSxb2196Ak1OZFRlwBdbpxmudWVw/IysHTsrS/p2zoLsmJv6p+WqdVNy6cpJWbKsN1lWemXzW9e/5ZHR29ubT37ykznllFMasAhoCavL32ylNjSQyfdekBfmgryg1p1Vuxyc3/Qekc/eu09uWDO29LyGWrraN1d5ZCKjKo5ktKX6HZtmZdXYWVnSMyN3DO+YG/un5co1k3PxqklZu8Q1ANtqYGi49ASgFTXZHR1rw4OZdN9FeV4uyjG1rqyefVDO7zsin71v38xfPa70vMqt7R/M+v7BjOnz5yAPTWRUZUBktKrhntHpnzA7K8fMzH09O+X2oem5fuPUXL56Sq5cNb4l79jUSoZEBrA1mvhayNrwUCYuviTPySV5dmpZO/vA/K7viHx28X65etX40vMqs65/QGTwsERGVQadLtXMhvvGZ8OEXXL/6JlZ1D0jtw1Mz/wN2+eyNdvlmtXjMrym9ugvQkNs6ZGMc889tzFDgNZSa41vANUynPGLL8szc1mOTi3rZh2Q349+Uj63eP9cvnJC6XnbZNA3iXgEIqMqjmQ0tVr/moxZNj9jMj8zkhyY5KV/+D+69XlZva9N8tnSK4BWU2u9bw7VMpxxS67MM3JlnpEkrX5jqtpNaf1/CRqlNb4N0Aqa+LAtNLUuh9qBrdAiRzLaWndv6QU0MV+hQFldDqgCW0FklCcyeAS+QqvS1353joARITKArTFqYukFna13bDKqta8pobH86Z7kyCOPzNy5czN69Oh88YtfTF9fX9761rfmxBNP3PwX8YUGW8d3woCtMX6H0gs624QZpRfQ5BzJeMCpp56acePG5aKLLspHP/rRfPCDH8xZZ521+S8wqn1uSQcjyncjga0x0ZvcoibuVHoBTU5kPGDu3Ln5wAc+kDlz5uTYY4/NwQcfnLPPPnvzX8AbJdg646eXXgC0otGT6h+UMXHn0gtociLjAXPnzn3Qj2fMmJHFixdv/guMnlztIOgUE3YsvQBoVdvtVnpB53Ikg0chMh7Q2/vg88JrtVqGhoY2/wXGbJfU3IoTtpjzqoGtNX3f0gs61477l15AkxMZVenqSsZOKb0CWo/IALbWrCeUXtC5ZvpvzyMTGVUaN630AmgxNddkAFtv1iGlF3Sm8Tsmk2eXXkGTExlVGje19AJoLWO3dwtbYOtN2ycZ5eLvETfz4NILaAGek5Hk3HPP/au/d8YZZ2z5C02atc1boKM4VQrYFl1dycyDklt/XXpJZ5k1r/QCWoAjGVWa9pjSC6C1TBAZwDba/cjSCzrPnKNLL6AFiIwqiQzYMuPdvhbYRvu9KEmt9IrOMe0xyfR9Sq+gBYiMKk0XGbBFps4pvQBodZNnu9PRSNrvRaUX0CJERpUmzUr6xpdeAa1jxgGlFwDtYP+XlF7QOUQGm0lkVKlWS6btXXoFtI4Zjyu9AGgH+70wqXlL03DT9/U+h83mK7Jq05ynCJtl4s7JuO1LrwDawYQdk72eVXpF+zv4DaUX0EJERtUUPmwep0oBVTriXaUXtLfRk5PHvbr0ClqIyKiaOy7A5hEZQJVmzUtmP7H0ivZ18BuSvnGlV9BCREbV3MYWNs+Oc0svANqNoxmN0Ts2OewdpVfQYkRG1SbPSibsVHoFND9HMoCq7XV0ssNjS69oPwe/IRk3tfQKWozIaARPH4VHNnZqMmnn0iuAdnT0v5de0F7G75A85R9Lr6AFiYxGEBnwyHZ6XOkFQLva46nJY44pvaJ9HPXBZPSk0itoQSKjEUQGPLI9nl56AdDOnvWRpNdFytts9hOTA15ZegUtSmQ0woQd6g+sAR7a3u5nDzTQ5NnJU/+l9IrW1tWTPPfjpVfQwkRGoziaAQ9t+znJlN1LrwDa3aFvT2YdWnpF63ryPyY77Fd6BS1MZDTK7k8tvQCa017PLL0A6ARd3cnL/i8ZN630ktazx9OSJ7+39ApanMholF0PT7p6S6+A5rOXU6WAETJxRvKSLyU1b3c224Sdkhd/Ieny34xt4/+DGqVvXP3po8CfjJqUzD6s9Aqgk+z+lOSp7yu9ojV09SQv/bJnYlAJkdFIe7qDDjzInk9LuntKrwA6zZPek+zzvNIrmt9zPpbs4htBVENkNNJjX5akVnoFNA+nSgEl1GrJS76c7PmM0kua11EfrD/ZGyoiMhpp8uxk1yNKr4DmUOtK5hxdegXQqXr6kld8LdnFn8t/5UnvSQ4/vvQK2ozIaLTHvab0AmgOuz4pGTul9Aqgk/WOSV79rWTmE0ovaR7z3pI8/f2lV9CGREaj7fv8pG986RVQ3uP/pvQCgGTU+OS13/M8qyQ54l3Js/+r9AralMhotL5xyb4vKL0Cyho9yUWXQPMYPSl5zfeSg/629JIyat3Jcz6ePOPE+vUq0AAiYyQc8KrSC6Csx7486R1degXAn3T3JM/7ZPLMj3TWczRGT05e+91k3ptLL6HNddBXVUG7HlG/CBw61YFOlQKa1GFvT171rWRMB1wzttOByZt/XX+iNzSYyBgJtZqjGXSunQ9OZhxQegXAw9vr6OTtv2/fW9x29SZP/dfkjWcl2+9Reg0dQmSMlANeFc/MoCPNO670AoBHN2HH+gXhz/t0MmpS6TXVmb5f/ejFU97rYaiMKJExUqbsluz97NIrYGSNm57s96LSKwA230GvS955cTL3Fa19rcaoScnT3p8cd24yY27pNXSgFv7qaUFHvKv0AhhZB72u/gAsgFYyYcfkxZ9P3np+svdzSq/ZMj2jkyf+XXL8lcmT3+P3YIqpDQ8PD5ce0VG+/OxkwQWlV0Dj9YxO/v7KZOKM0ksAts3dlyZnfzC5/bzSSx5e96jkgFckT/nnZNLOpdeAyBhxN/0y+frLSq+Axjv0HcmzPlx6BUB1Fl+fXHZqcvU3k/UrSq+p235O/ajxAa9Oxm1feg38kcgo4X+OSO69pvQKaJy+8cnxVyXjppZeAlC9gY3JdT9KrvhqcucFydDAyP76oyYmez0zOfB1yW5PGtlfGzaTyCjhhp8m33x16RXQOE9+b/K0fy29AqDxNq5Obv9Ncuuv6x/Lb2vAL1JLdtw/2e0pyZyjkl0OT7p7G/DrQHVERin/+5Rk0ZWlV0D1Rk9O/uHqZHQb3QISYHOtuLP+5/t985MlNyTLbktW3J70r9m8nz92ajLtMcm0vZPp+9T/eof9krEd8LBA2orIKOWmM5Ovv7z0Cqje0/8tedK7S68AaLgjjzwyc+fOzejRo/PFL34xfX19eetb35oTTzzxr//hTRvqobFx9QOfH4iO0ROTURPqp0CNmph0ufEn7cFTWUrZ65n1JyEvvLT0EqjOuOnJIW8rvQJgxJx66qk54YQTctFFF+X3v/99Xv/61+fwww/PUUcd9eB/sHd0/cO1anQIuVzSsz/a2g/6gb/0pHcnfWNLrwAYMXPnzs0HPvCBzJkzJ8cee2wOPvjgnH322aVnQXHe4ZY086DkoNeXXgHVmDQrOfhvS68AGFFz5z74adozZszI4sWLC62B5iEySnv6B5Jx00qvgG33jBOTnlGlVwCMqN7eB9/lqVarZWhoqNAaaB4io7Qxk5OjP1R6BWybvZ6dPPalpVcAAE1CZDSDA16Z7HJE6RWwdUZNSo75ROkVAEATERnN4rknJV0erEMLOur/SybuVHoFANBEPCejmfzqxOT8k0uvgM2365OS1/04qdVKLwEAmogjGc3kyf+YTJ5degVsnt6xyfM/LTAAgL8iMppJ39jkeZ/27Axaw1Pfl0zZvfQKAKAJeTfbbPZ4av2BZtDMdj4oOfTtpVcAAE1KZDSjI/+lfq47NKOeMckLPpt0dZdeAgA0KZHRjLq6k5d8KRk3vfQS+GvP/0wyfZ/SKwCAJiYymtWEHZKXfMH1GTSXQ9+ezH1Z6RUAQJPzDraZ7X5k/Y5T0Ax2fVJy1L+XXgEAtACR0eye8k/Jbk8pvYJON3Fm8rL/S7p7Si8BAFqAyGh2XV3JS76YjN+h9BI6Vc/o5BWnJeOmll4CALQIkdEKxk9PXvG1+sPPYKQ99xPJzgeWXgEAtBCR0SpmzUtedmrS5XQVRtAT3pQ8/jWlVwAALUZktJK9jk5e8LkktdJL6AS7PzV51n+WXgEAtCCR0WoOeEXyzA+XXkG7mzkveeXpSXdv6SUAQAsSGa3osLcnR7yr9Ara1Q77J6/5TtI3rvQSAKBF1YaHh4dLj2Ar/ejvksu/WnoF7WTK7skbzqzfbAAAYCs5ktHKjvlk8phjSq+gXWy3W/K6HwsMAGCbiYxW1tWdvPTL9Qt0YVtst2vy+p8kk2aWXgIAtAGR0ep6RiWv/nayz/NKL6FVbbdr8vqfCgwAoDIiox309NWfofG415ZeQquZurfAAAAq58LvdjI8nPzyX5Pf/3fpJbSC3Y9MXv7VZPSk0ksAgDYjMtrRBf9dj434n5aHceCxyXNPTro9QR4AqJ7IaFfX/TD5/nHJwIbSS2gqteSo/y85/PjSQwCANiYy2tmCi5JvvipZt6z0EppB79jkxZ93kwAAoOFERrtbcUfyndcn91xRegkljd8xedU3kp0PLL0EAOgAIqMTDPQnv/pAcuHnSi+hhB32T179LXeQAgBGjMjoJDf+PDnjbcn6FaWXMCJqybzj6tdg9I4pPQYA6CAio9OsvDv53puSBb8vvYRGGr9j8sLPJXs+vfQSAKADiYxONDSYnPPh5PxPJMNDpddQtX2enzzvU8nYKaWXAAAdSmR0slvPqd/mdu3i0kuoQt+E5DkfTR736tJLAIAOJzI63dplya/+Lbni9Hh4XwubfVjyov9Jttu19BIAAJHBA+66JPnpCcm9V5dewpboHZc85b3JE49PurpKrwEASCIy+HNDQ8mlX0p+/e/JhpWl1/CIavXTop72/mTijNJjAAAeRGTw19YuTc76t+TKr8cpVE1ol8OTZ3442elxpZcAADwkkcHDW3BR8rN3J/deU3oJSbLdbslRH0z2fX7pJQAAj0hk8MiGBpOrv5Wcf3Ky9KbSazrTqEn16y7mvSXp6Su9BgDgUYkMNs/QUHL9j+rP1lh0Vek1naF3XHLQ65MnvTsZt33pNQAAm01ksOVu/lXy2497anijjJ2aHPLW5Alv9EA9AKAliQy23p0XJL89KbnlV6WXtIft96zHxeNfm/SOKb0GAGCriQy23T1XJhf9T3LdD5NN60qvaTG1ZM7RySHHJXs8PanVSg8CANhmIoPqbFydXPu9+tPD77649JrmNml2st8Lk4P/Npmye+k1AACVEhk0xpKbkmu+k1z73WT5baXXNIdJs+u3n93vxcnMg0qvAQBoGJFB4y28vH6E47ofJivvKr1mZE2alez7gmS/FyUzDy69BgBgRIgMRtayW5Pbzk1u/01yx2+TdctKL6pWrSuZvl+y+1OSfV9YDwvXWQAAHUZkUM7wcHLf/OT28+rRcecFycZVpVdtmb7xyc4HJbMPTWYdksx8QjJ6YulVAABFiQyax9Bg/dSqey5PltxYf8L4khuTtYtLL6urdSUTZ9aPTvwhKnZ8bNLVXXoZAEBTERk0v/UrkqU3/yk8/hAfaxYnm9ZW+2uNmpRsNzvZbtf6x+Rdku12e+CvZyc9fdX+egAAbUhk0NoGB+qnWG24P9mw8oGPVX/66z+cftXVU//o7k26epO+sfVTnUZNeODz+GTizp6wDQBQAZEBAABUqqv0AAAAoL2IDAAAoFIiAwAAqJTIAAAAKiUyAACASokMAACgUiIDAAColMgAAAAqJTIAAIBKiQwAAKBSIgMAAKiUyAAAAColMgAAgEqJDAAAoFIiAwAAqJTIAAAAKiUyAACASokMAACgUiIDAAColMgAAAAqJTIAAIBKiQwAAKBSIgMAAKiUyAAAAColMgAAgEqJDAAAoFIiAwAAqJTIAAAAKiUyAACASokMAACgUiIDAAColMgAAAAqJTIAAIBKiQwAAKBSIgMAAKiUyAAAAColMgAAgEqJDAAAoFIiAwAAqJTIAAAAKiUyAACASokMAACgUiIDAAColMgAAAAqJTIAAIBKiQwAAKBSIgMAAKiUyAAAAColMgAAgEqJDAAAoFIiAwAAqJTIAAAAKiUyAACASokMAACgUiIDAAColMgAAAAqJTIAAIBKiQwAAKBSIgMAAKiUyAAAAColMgAAgEqJDAAAoFIiAwAAqJTIAAAAKiUyAACASokMAACgUiIDAAColMgAAAAqJTIAAIBKiQwAAKBSIgMAAKiUyAAAAColMgAAgEqJDAAAoFIiAwAAqJTIAAAAKiUyAACASokMAACgUiIDAAColMgAAAAq9f8D1k+c5gbAjc0AAAAASUVORK5CYII=\n",
      "text/plain": [
       "<Figure size 1000x1000 with 4 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig,ax = plt.subplots(x,y,figsize=(10,10))\n",
    "ax = ax.flatten()\n",
    "sbt = ['p','n']\n",
    "max_values = max([results[key]['positives']+results[key]['negatives'] for key in results.keys()])\n",
    "for idx, res in enumerate(results):\n",
    "    p = results[idx]['positives']\n",
    "    n = results[idx]['negatives']\n",
    "    if p == 0:\n",
    "        p=None\n",
    "    if n==0:\n",
    "        n=None\n",
    "#     import pdb\n",
    "#     pdb.set_trace()\n",
    "#     ax[int(idx/5)][idx%5].pie([p,n],labels=sbt)\n",
    "    if p and n:\n",
    "        ax[idx].pie([p,n],labels=sbt,radius = (p+n)/max_values)\n",
    "    else:\n",
    "        ax[idx].axis('off')"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "89cfd92c",
   "metadata": {},
   "source": [
    "## old"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "7634aa3d",
   "metadata": {},
   "outputs": [],
   "source": [
    "# %%time\n",
    "# v_t = v.copy()\n",
    "# for t in tqdm(range(1,51)):#range(1,n_iter+1):\n",
    "# #     clear_output(wait=True)\n",
    "#     print('Iter ',t)\n",
    "#     sigma_t = calc_sigma(t)\n",
    "#     h_t = np.empty((C,C))\n",
    "    \n",
    "#     for i in range(C):\n",
    "#         for j in range(C):\n",
    "#             h_t[i,j] = calc_h(nodes,i,j,sigma_t)\n",
    "    \n",
    "#     ## step 1: compute Gr\n",
    "#     for r in tqdm(range(C)):\n",
    "#         g = []\n",
    "        \n",
    "#         for h in range(norm_diss.shape[0]):\n",
    "#             values_array = np.empty((norm_diss.shape[0],norm_diss.shape[0]))\n",
    "#             for k in range(h,norm_diss.shape[0]):\n",
    "#                 rescalc = (\n",
    "#                     h_t[elements_clusters[k]['cluster'],r]\n",
    "#                     *distance.euclidean(elements_clusters[k]['element'],elements_clusters[h]['element'])\n",
    "#                 )\n",
    "#                 values_array[h,k] = rescalc\n",
    "#                 values_array[k,h] = rescalc\n",
    "            \n",
    "#             val = values_array.sum(axis=1)[h]\n",
    "#             g.append((h,val))\n",
    "            \n",
    "#         g = sorted(g, key= lambda x: x[1])[:q]\n",
    "        \n",
    "#         G[r] = np.array([norm_diss[pair[0]] for pair in g])\n",
    "    \n",
    "    \n",
    "# #     print('Iter ',t, '- Weighting')\n",
    "#     ## step 2: weighting\n",
    "#     for r in tqdm(range(v_t.shape[0])): #for all lines of v\n",
    "#         for e_idx in range(v_t.shape[1]): #for all columns of v\n",
    "#             base_sum = []\n",
    "            \n",
    "#             upper = np.sum(\n",
    "#                 [\n",
    "#                     (\n",
    "#                         calc_h(nodes, elements_clusters[k]['cluster'],r, sigma_t)\n",
    "#                         *distance.euclidean(elements_clusters[k]['element'],G[r][e_idx])\n",
    "#                     ) \n",
    "#                     for k in range(norm_diss.shape[0]) \n",
    "#                 ]\n",
    "#             )\n",
    "            \n",
    "#             for element in G[r]:            \n",
    "#                 lower = np.sum(\n",
    "#                     [\n",
    "#                         (\n",
    "#                             calc_h(nodes, elements_clusters[k]['cluster'],r, sigma_t)\n",
    "#                             *distance.euclidean(elements_clusters[k]['element'],element)\n",
    "#                         ) \n",
    "#                         for k in range(norm_diss.shape[0]) \n",
    "#                     ]\n",
    "#                 )\n",
    "\n",
    "#                 base_sum.append(\n",
    "#                     pow(upper/lower,1/(n-1))\n",
    "#                 )\n",
    "            \n",
    "#             v_t[r][e_idx] = 1/sum(base_sum)\n",
    "            \n",
    "# #     print('Iter ',t, '- Assignment')\n",
    "#     ## step 3: assignment\n",
    "#     P = [[] for i in range(C)]\n",
    "#     indices = list(range(norm_diss.shape[0]))\n",
    "#     elements_clusters = defaultdict()\n",
    "    \n",
    "#     for k in tqdm(indices):\n",
    "#         #calculate f_ek\n",
    "#         deltas = np.empty(C)\n",
    "#         for s in range(C):\n",
    "#             val = sum(\n",
    "#                 [\n",
    "#                     calc_h(nodes,s,r,sigma)*D_vr(norm_diss[k],r) for r in range(C)\n",
    "#                 ]\n",
    "#             )\n",
    "\n",
    "#             deltas[s] = val\n",
    "\n",
    "#         k_cluster = np.argmin(deltas)\n",
    "#         elements_clusters[k] = {'cluster': k_cluster, 'element': norm_diss[k]}\n",
    "\n",
    "#         P[k_cluster].append(k)   "
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "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": "3.10.8"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}