5 Core TMA Dice Calculation.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "bad9132d",
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "from skimage.io import imread, imshow, imsave\n",
    "import os\n",
    "import matplotlib.pyplot as plt\n",
    "from scipy.special import softmax\n",
    "import pandas as pd\n",
    "from skimage.morphology import binary_dilation, disk\n",
    "from skimage.measure import regionprops, label"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "59d5dbb8",
   "metadata": {},
   "outputs": [],
   "source": [
    "methods = [\"Mesmer\", \"Stardist\", \"Cellpose\", \"UnMicst\"]\n",
    "datapath = \"/home/groups/ChangLab/dataset/HMS-TMA-TNP/OHSU-TMA/Segmentations/\"\n",
    "sample_ids = ['Scene 002','Scene 003', 'Scene 017', 'Scene 049', 'Scene 059']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "4ef83ec6",
   "metadata": {},
   "outputs": [],
   "source": [
    "def dice_coeff(true, pred):\n",
    "    true_f = true.flatten()\n",
    "    pred_f = pred.flatten()\n",
    "    union = np.sum(true_f) + np.sum(pred_f)\n",
    "    if union==0:\n",
    "        return 1\n",
    "    intersection = np.sum(true_f * pred_f)\n",
    "    return (2.* intersection/union)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "61926376",
   "metadata": {},
   "outputs": [],
   "source": [
    "from seg_utils import get_mask, get_gt"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "c90335d9",
   "metadata": {},
   "outputs": [],
   "source": [
    "method_gt_table = np.zeros((len(methods), len(sample_ids)))\n",
    "for m,method in enumerate(methods):\n",
    "    for s,sid in enumerate(sample_ids):\n",
    "        method_mask = get_mask(datapath, sid, method)\n",
    "        method_mask = (method_mask > 0).astype('int')\n",
    "        gt = get_gt(sid)\n",
    "        gt = (gt > 0).astype('int')\n",
    "        method_gt_table[m][s] = dice_coeff(gt, method_mask)\n",
    "        "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "8dce5b1b",
   "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>Scene 002</th>\n",
       "      <th>Scene 003</th>\n",
       "      <th>Scene 017</th>\n",
       "      <th>Scene 049</th>\n",
       "      <th>Scene 059</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>Mesmer</th>\n",
       "      <td>0.866914</td>\n",
       "      <td>0.833874</td>\n",
       "      <td>0.867354</td>\n",
       "      <td>0.845348</td>\n",
       "      <td>0.847514</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Stardist</th>\n",
       "      <td>0.871493</td>\n",
       "      <td>0.798544</td>\n",
       "      <td>0.849238</td>\n",
       "      <td>0.836196</td>\n",
       "      <td>0.849104</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Cellpose</th>\n",
       "      <td>0.852206</td>\n",
       "      <td>0.883965</td>\n",
       "      <td>0.844562</td>\n",
       "      <td>0.815968</td>\n",
       "      <td>0.839417</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>UnMicst</th>\n",
       "      <td>0.817395</td>\n",
       "      <td>0.810350</td>\n",
       "      <td>0.799904</td>\n",
       "      <td>0.738577</td>\n",
       "      <td>0.783098</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "          Scene 002  Scene 003  Scene 017  Scene 049  Scene 059\n",
       "Mesmer     0.866914   0.833874   0.867354   0.845348   0.847514\n",
       "Stardist   0.871493   0.798544   0.849238   0.836196   0.849104\n",
       "Cellpose   0.852206   0.883965   0.844562   0.815968   0.839417\n",
       "UnMicst    0.817395   0.810350   0.799904   0.738577   0.783098"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "pd.DataFrame(method_gt_table, index=methods, columns=sample_ids)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "df2cd0e0",
   "metadata": {},
   "source": [
    "# Evaluate methods using ground truth labels"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "928c1a43",
   "metadata": {},
   "outputs": [],
   "source": [
    "dice_table = np.zeros((len(sample_ids), len(methods)))\n",
    "\n",
    "for s,sid in enumerate(sample_ids):\n",
    "    gt = get_gt(sid)\n",
    "    gt = (gt > 0).astype('int')\n",
    "    \n",
    "    for m,method_out in enumerate(methods):\n",
    "        methods_in = [m for m in methods if m != method_out]\n",
    "        m_masks = []\n",
    "        \n",
    "        for method in methods_in:\n",
    "            method_mask = get_mask(datapath, sid, method)\n",
    "            method_mask[method_mask > 0] = 1\n",
    "            m_masks.append(method_mask)\n",
    "            \n",
    "        pseudo_gt = (sum(m_masks) >= 3).astype('int')\n",
    "        dice = dice_coeff(gt, pseudo_gt)\n",
    "        dice_table[s,m] = dice"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "928e254c",
   "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>Mesmer dropped</th>\n",
       "      <th>Stardist dropped</th>\n",
       "      <th>Cellpose dropped</th>\n",
       "      <th>UnMicst dropped</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>Scene 002</th>\n",
       "      <td>0.772368</td>\n",
       "      <td>0.777804</td>\n",
       "      <td>0.782760</td>\n",
       "      <td>0.807856</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Scene 003</th>\n",
       "      <td>0.804119</td>\n",
       "      <td>0.821026</td>\n",
       "      <td>0.791103</td>\n",
       "      <td>0.838813</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Scene 017</th>\n",
       "      <td>0.755549</td>\n",
       "      <td>0.769438</td>\n",
       "      <td>0.766917</td>\n",
       "      <td>0.798740</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Scene 049</th>\n",
       "      <td>0.699210</td>\n",
       "      <td>0.706349</td>\n",
       "      <td>0.718214</td>\n",
       "      <td>0.780739</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Scene 059</th>\n",
       "      <td>0.731426</td>\n",
       "      <td>0.737895</td>\n",
       "      <td>0.746556</td>\n",
       "      <td>0.787055</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "           Mesmer dropped  Stardist dropped  Cellpose dropped  UnMicst dropped\n",
       "Scene 002        0.772368          0.777804          0.782760         0.807856\n",
       "Scene 003        0.804119          0.821026          0.791103         0.838813\n",
       "Scene 017        0.755549          0.769438          0.766917         0.798740\n",
       "Scene 049        0.699210          0.706349          0.718214         0.780739\n",
       "Scene 059        0.731426          0.737895          0.746556         0.787055"
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "pd.DataFrame(dice_table, index=sample_ids, columns=[f'{m} dropped' for m in methods])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "1878a7cc",
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "dice_delta_table = np.zeros((len(sample_ids), len(methods)))\n",
    "\n",
    "for i,dice_i in enumerate(dice_table):\n",
    "    mean_dice = np.mean(dice_i)\n",
    "\n",
    "    for j,dice in enumerate(dice_i):\n",
    "        dice_delta_table[i][j] = 100 * (dice - mean_dice) / mean_dice"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "dcf69f83",
   "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>Mesmer dropped</th>\n",
       "      <th>Stardist dropped</th>\n",
       "      <th>Cellpose dropped</th>\n",
       "      <th>UnMicst dropped</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>Scene 002</th>\n",
       "      <td>-1.633849</td>\n",
       "      <td>-0.941568</td>\n",
       "      <td>-0.310388</td>\n",
       "      <td>2.885805</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Scene 003</th>\n",
       "      <td>-1.185376</td>\n",
       "      <td>0.892283</td>\n",
       "      <td>-2.784910</td>\n",
       "      <td>3.078004</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Scene 017</th>\n",
       "      <td>-2.214663</td>\n",
       "      <td>-0.417136</td>\n",
       "      <td>-0.743405</td>\n",
       "      <td>3.375204</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Scene 049</th>\n",
       "      <td>-3.707042</td>\n",
       "      <td>-2.723900</td>\n",
       "      <td>-1.089911</td>\n",
       "      <td>7.520853</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Scene 059</th>\n",
       "      <td>-2.571761</td>\n",
       "      <td>-1.710083</td>\n",
       "      <td>-0.556368</td>\n",
       "      <td>4.838212</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "           Mesmer dropped  Stardist dropped  Cellpose dropped  UnMicst dropped\n",
       "Scene 002       -1.633849         -0.941568         -0.310388         2.885805\n",
       "Scene 003       -1.185376          0.892283         -2.784910         3.078004\n",
       "Scene 017       -2.214663         -0.417136         -0.743405         3.375204\n",
       "Scene 049       -3.707042         -2.723900         -1.089911         7.520853\n",
       "Scene 059       -2.571761         -1.710083         -0.556368         4.838212"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "pd.DataFrame(dice_delta_table, index=sample_ids, columns=[f'{m} dropped' for m in methods])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "1f037fb1",
   "metadata": {},
   "outputs": [],
   "source": [
    "weights  = np.zeros(len(methods))\n",
    "avg_deltas  = np.zeros(len(methods))\n",
    "delta_table = dice_delta_table\n",
    "\n",
    "for m, method in enumerate(methods):\n",
    "    avg_deltas[m] = -np.mean(dice_delta_table[:,m])\n",
    "\n",
    "weights = softmax(avg_deltas)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "id": "c8c7843e",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([0.62874652, 0.17438593, 0.19601406, 0.00085349])"
      ]
     },
     "execution_count": 19,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "weights"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "id": "41c8d2ce",
   "metadata": {},
   "outputs": [],
   "source": [
    "def get_dice_scores(sample_ids, methods, weights):\n",
    "    dices = np.zeros(len(sample_ids))\n",
    "\n",
    "    for s,sid in enumerate(sample_ids):\n",
    "        gt = get_gt(sid)\n",
    "        gt = (gt > 0).astype('int')\n",
    "        m_masks = []\n",
    "\n",
    "        for m,method in enumerate(methods):\n",
    "            method_mask = get_mask(datapath, sid, method).astype('float')\n",
    "            method_mask[method_mask > 0] = weights[m]\n",
    "            m_masks.append(method_mask)\n",
    "\n",
    "        pseudo_gt = (sum(m_masks) >= 0.75).astype('int')\n",
    "        dices[s] = dice_coeff(gt, pseudo_gt)\n",
    "        \n",
    "    return dices"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "id": "ff3c36af",
   "metadata": {},
   "outputs": [],
   "source": [
    "equal_weights = np.ones(len(methods)) * (1/len(methods))\n",
    "dices = get_dice_scores(sample_ids, methods, equal_weights)\n",
    "w_dices = get_dice_scores(sample_ids, methods, weights)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "id": "6f57ef9c",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAiEAAAH/CAYAAACByw1FAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjYuMCwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy89olMNAAAACXBIWXMAAA9hAAAPYQGoP6dpAABH/UlEQVR4nO3de1TVVcL/8c8B5apg3iANxcwoDcFLsLykVCilj2U1Zd4gGnWcpFImZ9BEskapfjNEU5o1g5mNPvJU6jhppGFQXhIDzRwVU2vwMS5SBokjKuf7+6Of59cJUEBge3m/1vqu8eyz9z5778XqfGZ/L8dmWZYlAACAZuZiegAAAODqRAgBAABGEEIAAIARhBAAAGAEIQQAABhBCAEAAEYQQgAAgBGEEAAAYAQhBAAAGEEIAQAARlwSIWThwoUKDAyUh4eHwsPDlZOTU2vdM2fO6Nlnn1X37t3l4eGhkJAQZWRkXFSfAACg+RkPIenp6YqPj1dSUpLy8vIUEhKiqKgolZSU1Fh/zpw5ev311/XKK69o7969mjp1qu677z7t3LmzwX0CAIDmZzP9A3bh4eG69dZb9eqrr0qS7Ha7AgIC9PjjjyshIaFa/U6dOunpp5/WtGnTHGUPPPCAPD099fe//71BfQIAgObXwuSHnz59Wrm5uZo1a5ajzMXFRZGRkdq2bVuNbSorK+Xh4eFU5unpqc2bN19Un5WVlY7Xdrtd33//vdq1ayebzdbg+QEAcLWxLEs//vijOnXqJBeX859wMRpCSktLVVVVJT8/P6dyPz8/7d+/v8Y2UVFRSklJ0ZAhQ9S9e3dlZmZq1apVqqqqanCfycnJmjdvXiPMCAAASNKRI0d03XXXnbeO0RDSEC+//LImT56sm266STabTd27d1dsbKyWLFnS4D5nzZql+Ph4x+uysjJ16dJFR44ckY+PT2MMGwCAq0J5ebkCAgLUunXrC9Y1GkLat28vV1dXFRcXO5UXFxfL39+/xjYdOnTQmjVrdOrUKX333Xfq1KmTEhISdP311ze4T3d3d7m7u1cr9/HxIYQAANAAdbmcwejdMW5uburXr58yMzMdZXa7XZmZmRowYMB523p4eKhz5846e/as3nvvPd17770X3ScAAGg+xk/HxMfHKyYmRv3791dYWJhSU1NVUVGh2NhYSVJ0dLQ6d+6s5ORkSdL27dt19OhRhYaG6ujRo3rmmWdkt9v1+9//vs59AgAA84yHkDFjxujYsWOaO3euioqKFBoaqoyMDMeFpQUFBU5X1546dUpz5szR4cOH1apVK40YMUJvv/222rRpU+c+AQCAecafE3IpKi8vl6+vr8rKyrgmBMAVqaqqSmfOnDE9DFyGWrZsKVdX11rfr893qPGdEABA87EsS0VFRfrhhx9MDwWXsTZt2sjf3/+in6VFCAGAq8i5ANKxY0d5eXnxQEbUi2VZOnnypONnUK699tqL6o8QAgBXiaqqKkcAadeunenh4DLl6ekpSSopKVHHjh3Pe2rmQoz/gB0AoHmcuwbEy8vL8EhwuTv3N3Sx1xURQgDgKsMpGFysxvobIoQAAAAjuCYEAKDi78+q7IS92T7Pt5WL/Npenl9BNptNq1ev1ujRo5vl8wIDAzV9+nRNnz69TvW/+eYbdevWTTt37lRoaGiTju1iXZ5/AQCARlP8/VnFPPOtTp9tvs90ayG99UynyzaINKcdO3bI29u7UftcunSppk+fbvxWbU7HAMBVruyEvVkDiCSdPqtm3Xm5nHXo0OGKvZiYEAIAuOTZ7XYlJyerW7du8vT0VEhIiN59912nOuvXr9eNN94oT09P3X777Vq6dKlsNpvj/+0/88wz1U5PpKamKjAw0PF6x44dGjZsmNq3by9fX18NHTpUeXl5dR7n+++/rzZt2qiqqkqStGvXLtlsNiUkJDjqTJo0SRMmTHC83rx5s2677TZ5enoqICBATzzxhCoqKhzvBwYGKjU11fF6//79Gjx4sDw8PNSzZ0999NFHstlsWrNmjdNYDh8+rNtvv11eXl4KCQnRtm3bJElZWVmKjY1VWVmZbDabbDabnnnmGUnSokWL1KNHD3l4eMjPz0+/+tWv6jz3hiCEAAAuecnJyVq2bJkWL16sf/3rX5oxY4YmTJig7OxsSdKRI0d0//33a9SoUdq1a5cmTZrk9MVfVz/++KNiYmK0efNmffbZZ+rRo4dGjBihH3/8sU7tb7vtNv3444/auXOnJCk7O1vt27dXVlaWo052drYiIiIkSYcOHdJdd92lBx54QLt371Z6ero2b96suLi4GvuvqqrS6NGj5eXlpe3bt+uNN97Q008/XWPdp59+Wk899ZR27dqlG2+8UWPHjtXZs2c1cOBApaamysfHR4WFhSosLNRTTz2lzz//XE888YSeffZZ5efnKyMjQ0OGDKn74jUAJ+MAAJe0yspKLViwQB999JEGDBggSbr++uu1efNmvf766xo6dKhee+01de/eXX/+858lSUFBQfryyy/1wgsv1Ouz7rjjDqfXb7zxhtq0aaPs7Gz913/91wXb+/r6KjQ0VFlZWerfv7+ysrI0Y8YMzZs3TydOnFBZWZkOHjyooUOHSvopXI0fP95x0WmPHj30l7/8xTEnDw8Pp/43btyoQ4cOKSsrS/7+/pKk+fPna9iwYdXG8tRTT2nkyJGSpHnz5qlXr146ePCgbrrpJvn6+spmszn6kH76wVhvb2/913/9l1q3bq2uXbuqT58+dV+8BmAnBABwSTt48KBOnjypYcOGqVWrVo5j2bJlOnTokCRp3759Cg8Pd2p3LrDUR3FxsSZPnqwePXrI19dXPj4+OnHihAoKCurcx9ChQ5WVlSXLsvTpp5/q/vvv180336zNmzcrOztbnTp1Uo8ePSRJX3zxhZYuXeo0r6ioKNntdn399dfV+s7Pz1dAQIBTeAgLC6txHL1793b8+9zj1c89br0mw4YNU9euXXX99ddr4sSJWr58uU6ePFnneTcEOyEAgEvaiRMnJEnr1q1T586dnd5zd3evcz8uLi765Q/H//KJnzExMfruu+/08ssvq2vXrnJ3d9eAAQN0+vTpOn9ORESElixZoi+++EItW7bUTTfdpIiICGVlZen48eOOXZBzc/vNb36jJ554olo/Xbp0qfNn1qRly5aOf597uJjdXvvFwK1bt1ZeXp6ysrK0YcMGzZ07V88884x27NihNm3aXNRYakMIAQBc0nr27Cl3d3cVFBQ4fYH/3M0336y1a9c6lX322WdOrzt06KCioiJZluX4Ut61a5dTnS1btmjRokUaMWKEpJ+uNSktLa3XeM9dF/LSSy85xhsREaHnn39ex48f1+9+9ztH3b59+2rv3r264YYb6tR3UFCQjhw5ouLiYvn5+Un66WLa+nJzc3NcPPtzLVq0UGRkpCIjI5WUlKQ2bdpo06ZNuv/+++v9GXXB6RgAwCWtdevWeuqppzRjxgy99dZbOnTokPLy8vTKK6/orbfekiRNnTpVX331lWbOnKn8/HytWLFCS5cudeonIiJCx44d04svvqhDhw5p4cKF+uCDD5zq9OjRQ2+//bb27dun7du3a/z48Y4fbKura665Rr1799by5csdF6AOGTJEeXl5OnDggFOQ+sMf/qCtW7cqLi5Ou3bt0ldffaV//OMftV6YOmzYMHXv3l0xMTHavXu3tmzZojlz5kiq36PUAwMDdeLECWVmZqq0tFQnT57U+++/r7/85S/atWuX/v3vf2vZsmWy2+0KCgqq1/zrgxACALjkPffcc0pMTFRycrJuvvlm3XXXXVq3bp26desm6adTF++9957WrFmjkJAQLV68WAsWLHDq4+abb9aiRYu0cOFChYSEKCcnR0899ZRTnbS0NB0/flx9+/bVxIkT9cQTT6hjx471Hu/QoUNVVVXlCCFt27ZVz5495e/v7/Sl3rt3b2VnZ+vAgQO67bbb1KdPH82dO1edOnWqsV9XV1etWbNGJ06c0K233qpJkyY57o755UWs5zNw4EBNnTpVY8aMUYcOHfTiiy+qTZs2WrVqle644w7dfPPNWrx4sf77v/9bvXr1qvf868pm/fIEGVReXi5fX1+VlZXJx8fH9HAAoFGcOnVKX3/9tbp16+b0hXWlPjE1KytLt99+u44fP95k1zRcCrZs2aLBgwfr4MGD6t69e7N8Zm1/S1L9vkO5JgQArnJ+bVvorWc68dsxl4nVq1erVatW6tGjhw4ePKgnn3xSgwYNarYA0pj4CwAAyK9tC/m1NT0K1MWPP/6oP/zhDyooKFD79u0VGRnpeD7K5YYQAgC4IkVERFS7JfdKEB0drejoaNPDaBRcmAoAAIwghAAAACMIIQAAwAhCCAAAMIIQAgAAjCCEAAAAI7hFFwAg/XBMOlnefJ/n5SO16XBRXURERCg0NFSpqamSfvo9lOnTp2v69OmSpKKiIk2cOFFbt25Vy5Yt9cMPP9RY1tR+Oc66sNlsWr16tUaPHt1k47oUEEIA4Gr3wzHp1WnS2TMXrttYWrSU4hZedBD5uR07dsjb29vx+qWXXlJhYaF27dolX1/fWsua2qpVq9SyZctG7fNKeSQ9IQQArnYny5s3gEg/fd7J8kYNIR06OPd16NAh9evXTz169DhvWVNr25ZH0daGa0IAAJe8iooKRUdHq1WrVrr22mtrfEx5YGCg06mZ9957T8uWLZPNZtMjjzxSY9kv7dmzRy4uLjp27Jgk6fvvv5eLi4sefvhhR50//vGPGjx4sFObu+++W61atZKfn58mTpyo0tJSx/sRERGOU0SSVFhYqJEjR8rT01PdunXTihUrnMZ+Tmlpqe677z55eXmpR48eWrt2rSTpm2++0e233y5Juuaaa5zm8u677yo4OFienp5q166dIiMjVVFRUed1bm6EEADAJW/mzJnKzs7WP/7xD23YsEFZWVnKy8urtf6OHTt011136aGHHlJhYaFefvnlGst+qVevXmrXrp2ys7MlSZ9++qnTa0nKzs5WRESEJOmHH37QHXfcoT59+ujzzz9XRkaGiouL9dBDD9U6tujoaH377bfKysrSe++9pzfeeEMlJSXV6s2bN08PPfSQdu/erREjRmj8+PH6/vvvFRAQoPfee0+SlJ+f75hLYWGhxo4dq0cffVT79u1TVlaW7r///kv60fWEEADAJe3EiRNKS0vTn/70J915550KDg7WW2+9pbNnz9bapkOHDnJ3d5enp6f8/f3l6+tbY9kv2Ww2DRkyRFlZWZJ+uvYiNjZWlZWV2r9/v86cOaOtW7dq6NChkqRXX31Vffr00YIFC3TTTTepT58+WrJkiT7++GMdOHCgWv/79+/XRx99pL/+9a8KDw9X37599be//U3/+c9/qtV95JFHNHbsWN1www1asGCBTpw4oZycHLm6ujpO8XTs2NExl8LCQp09e1b333+/AgMDFRwcrMcee0ytWrVqyLI3C0IIAOCSdujQIZ0+fVrh4eGOsrZt2yooKKhJPm/o0KGOEJKdna077rjDEUx27NihM2fOaNCgQZKkL774Qh9//LFatWrlOG666SbHuH8pPz9fLVq0UN++fR1lN9xwg6655ppqdXv37u34t7e3t3x8fGrcMTknJCTEEdIefPBB/fWvf9Xx48cbtAbNhRACAMDPREREaO/evfrqq6+0d+9eDR48WBEREcrKylJ2drb69+8vLy8vST/t0owaNUq7du1yOr766isNGTLkosbxyztqbDab7HZ7rfVdXV21ceNGffDBB+rZs6deeeUVBQUF6euvv76ocTQl4yFk4cKFCgwMlIeHh8LDw5WTk3Pe+qmpqQoKCpKnp6cCAgI0Y8YMnTp1yvF+VVWVEhMT1a1bN3l6eqp79+567rnnLulzYgCA2nXv3l0tW7bU9u3bHWXHjx+v8XRHYwgODtY111yjP/7xjwoNDVWrVq0UERGh7OxsZWVlOa4HkaS+ffvqX//6lwIDA3XDDTc4HT+/XficoKAgnT17Vjt37nSUHTx4sN47Fm5ubpJ++s77OZvNpkGDBmnevHnauXOn3NzctHr16nr13ZyMhpD09HTFx8crKSlJeXl5CgkJUVRUVK3bTStWrFBCQoKSkpK0b98+paWlKT09XbNnz3bUeeGFF/Taa6/p1Vdf1b59+/TCCy/oxRdf1CuvvNJc0wIANKJWrVrp17/+tWbOnKlNmzZpz549euSRR+Ti0jRfYeeuC1m+fLkjcPTu3VuVlZXKzMx0XA8iSdOmTdP333+vsWPHaseOHTp06JA+/PBDxcbGVgsIknTTTTcpMjJSU6ZMUU5Ojnbu3KkpU6bI09NTNputzmPs2rWrbDab3n//fR07dkwnTpzQ9u3btWDBAn3++ecqKCjQqlWrdOzYMd18880XvSZNxWgISUlJ0eTJkxUbG6uePXtq8eLF8vLy0pIlS2qsv3XrVg0aNEjjxo1TYGCghg8frrFjxzrtnmzdulX33nuvRo4cqcDAQP3qV7/S8OHDL7jDAgC4dP2f//N/dNttt2nUqFGKjIzU4MGD1a9fvyb7vKFDh6qqqsoRQlxcXDRkyBDHTsM5nTp10pYtW1RVVaXhw4crODhY06dPV5s2bWoNScuWLZOfn5+GDBmi++67T5MnT1br1q3l4eFR5/F17txZ8+bNU0JCgvz8/BQXFycfHx998sknGjFihG688UbNmTNHf/7zn3X33Xdf1Fo0JZtl6DzF6dOn5eXlpXfffdfpsbQxMTH64Ycf9I9//KNamxUrVuixxx7Thg0bFBYWpsOHD2vkyJGaOHGiYzdkwYIFeuONN7RhwwbdeOON+uKLLzR8+HClpKRo/PjxNY6lsrJSlZWVjtfl5eUKCAhQWVmZfHx8GnfiAGDIqVOn9PXXX6tbt27OX3hXyBNTL1f/+7//q4CAAH300Ue68847TQ+nTmr9W9JP36G+vr51+g419sTU0tJSVVVVyc/Pz6ncz89P+/fvr7HNuHHjVFpaqsGDB8uyLJ09e1ZTp051Oh2TkJCg8vJy3XTTTXJ1dVVVVZXmz59fawCRpOTkZM2bN69xJgYAl5s2HX4KBJfZb8dcrjZt2qQTJ04oODhYhYWF+v3vf6/AwMCLvpD1cnRZPbY9KytLCxYs0KJFixQeHq6DBw/qySef1HPPPafExERJ0v/8z/9o+fLlWrFihXr16qVdu3Zp+vTp6tSpk2JiYmrsd9asWYqPj3e8PrcTAgBXjTYdrtpQ0NzOnDmj2bNn6/Dhw2rdurUGDhyo5cuXN/rvy1wOjIWQ9u3by9XVVcXFxU7lxcXF8vf3r7FNYmKiJk6cqEmTJkn66QrmiooKTZkyRU8//bRcXFw0c+ZMJSQkOB6xGxwcrH//+99KTk6uNYS4u7vL3d29EWcHAEDNoqKiFBUVZXoYlwRjF6a6ubmpX79+yszMdJTZ7XZlZmZqwIABNbY5efJktQt9XF1dJclxC25tdc53bzUAAGh+Rk/HxMfHKyYmRv3791dYWJhSU1NVUVGh2NhYST89X79z585KTk6WJI0aNUopKSnq06eP43RMYmKiRo0a5Qgjo0aN0vz589WlSxf16tVLO3fuVEpKih599FFj8wQAANUZDSFjxozRsWPHNHfuXBUVFSk0NFQZGRmOi1ULCgqcdjXmzJkjm82mOXPm6OjRo+rQoYMjdJzzyiuvKDExUY899phKSkrUqVMn/eY3v9HcuXObfX4AcCliZxgXq7H+hozdonspq8/tRQBwubDb7frqq6/k6uqqDh06yM3NrV4PyAIsy9Lp06d17NgxVVVVqUePHtUugbgsbtEFADQvFxcXdevWTYWFhfr2229NDweXMS8vL3Xp0uWin1pLCAGAq4ibm5u6dOmis2fP1vhYceBCXF1d1aJFi0bZRSOEAMBVxmazqWXLllflcylwaTH+K7oAAODqRAgBAABGEEIAAIARhBAAAGAEIQQAABhBCAEAAEYQQgAAgBGEEAAAYAQhBAAAGEEIAQAARhBCAACAEYQQAABgBCEEAAAYQQgBAABGEEIAAIARhBAAAGAEIQQAABhBCAEAAEYQQgAAgBGEEAAAYAQhBAAAGEEIAQAARhBCAACAEYQQAABgBCEEAAAYQQgBAABGEEIAAIARhBAAAGAEIQQAABhBCAEAAEYQQgAAgBGEEAAAYITxELJw4UIFBgbKw8ND4eHhysnJOW/91NRUBQUFydPTUwEBAZoxY4ZOnTrlVOfo0aOaMGGC2rVrJ09PTwUHB+vzzz9vymkAAIB6amHyw9PT0xUfH6/FixcrPDxcqampioqKUn5+vjp27Fit/ooVK5SQkKAlS5Zo4MCBOnDggB555BHZbDalpKRIko4fP65Bgwbp9ttv1wcffKAOHTroq6++0jXXXNPc0wMAAOdhsyzLMvXh4eHhuvXWW/Xqq69Kkux2uwICAvT4448rISGhWv24uDjt27dPmZmZjrLf/e532r59uzZv3ixJSkhI0JYtW/Tpp582eFzl5eXy9fVVWVmZfHx8GtwPAABXm/p8hxo7HXP69Gnl5uYqMjLy/w/GxUWRkZHatm1bjW0GDhyo3Nxcxymbw4cPa/369RoxYoSjztq1a9W/f389+OCD6tixo/r06aO//vWvTTsZAABQb8ZOx5SWlqqqqkp+fn5O5X5+ftq/f3+NbcaNG6fS0lINHjxYlmXp7Nmzmjp1qmbPnu2oc/jwYb322muKj4/X7NmztWPHDj3xxBNyc3NTTExMjf1WVlaqsrLS8bq8vLwRZggAAM7H+IWp9ZGVlaUFCxZo0aJFysvL06pVq7Ru3To999xzjjp2u119+/bVggUL1KdPH02ZMkWTJ0/W4sWLa+03OTlZvr6+jiMgIKA5pgMAwFXNWAhp3769XF1dVVxc7FReXFwsf3//GtskJiZq4sSJmjRpkoKDg3XfffdpwYIFSk5Olt1ulyRde+216tmzp1O7m2++WQUFBbWOZdasWSorK3McR44cucjZAQCACzEWQtzc3NSvXz+ni0ztdrsyMzM1YMCAGtucPHlSLi7OQ3Z1dZUknbu+dtCgQcrPz3eqc+DAAXXt2rXWsbi7u8vHx8fpAAAATcvoLbrx8fGKiYlR//79FRYWptTUVFVUVCg2NlaSFB0drc6dOys5OVmSNGrUKKWkpKhPnz4KDw/XwYMHlZiYqFGjRjnCyIwZMzRw4EAtWLBADz30kHJycvTGG2/ojTfeMDbPJvXMfY3Y1+rG6wsAgAswGkLGjBmjY8eOae7cuSoqKlJoaKgyMjIcF6sWFBQ47XzMmTNHNptNc+bM0dGjR9WhQweNGjVK8+fPd9S59dZbtXr1as2aNUvPPvusunXrptTUVI0fP77Z5wcAAGpn9Dkhl6rL6jkh7IQAAC4hl8VzQgAAwNWNEAIAAIwghAAAACMIIQAAwAijd8cAzemOx2p/YF19bVrUpdH6AoCrFSEEwOWLu8OAyxqnYwAAgBGEEAAAYAQhBAAAGEEIAQAARhBCAACAEYQQAABgBCEEAAAYQQgBAABGEEIAAIARPDEVuBg8sRMAGoydEAAAYAQhBAAAGEEIAQAARhBCAACAEYQQAABgBHfHNKM7HitotL42LerSaH0BAGACOyEAAMAIQggAADCCEAIAAIwghAAAACMIIQAAwAjujgEA4HJzhfxuFTshAADACEIIAAAwghACAACMIIQAAAAjCCEAAMAIQggAADCCEAIAAIwghAAAACMuiRCycOFCBQYGysPDQ+Hh4crJyTlv/dTUVAUFBcnT01MBAQGaMWOGTp06VWPd559/XjabTdOnT2+CkQMAgIYyHkLS09MVHx+vpKQk5eXlKSQkRFFRUSopKamx/ooVK5SQkKCkpCTt27dPaWlpSk9P1+zZs6vV3bFjh15//XX17t27qacBAADqyXgISUlJ0eTJkxUbG6uePXtq8eLF8vLy0pIlS2qsv3XrVg0aNEjjxo1TYGCghg8frrFjx1bbPTlx4oTGjx+vv/71r7rmmmuaYyoAAKAejIaQ06dPKzc3V5GRkY4yFxcXRUZGatu2bTW2GThwoHJzcx2h4/Dhw1q/fr1GjBjhVG/atGkaOXKkU9+1qaysVHl5udMBAACaltEfsCstLVVVVZX8/Pycyv38/LR///4a24wbN06lpaUaPHiwLMvS2bNnNXXqVKfTMStXrlReXp527NhRp3EkJydr3rx5DZ8IAACoN+OnY+orKytLCxYs0KJFi5SXl6dVq1Zp3bp1eu655yRJR44c0ZNPPqnly5fLw8OjTn3OmjVLZWVljuPIkSNNOQUAACDDOyHt27eXq6uriouLncqLi4vl7+9fY5vExERNnDhRkyZNkiQFBweroqJCU6ZM0dNPP63c3FyVlJSob9++jjZVVVX65JNP9Oqrr6qyslKurq5Ofbq7u8vd3b2RZwcAAM7H6E6Im5ub+vXrp8zMTEeZ3W5XZmamBgwYUGObkydPysXFedjnQoVlWbrzzjv15ZdfateuXY6jf//+Gj9+vHbt2lUtgAAAADOM7oRIUnx8vGJiYtS/f3+FhYUpNTVVFRUVio2NlSRFR0erc+fOSk5OliSNGjVKKSkp6tOnj8LDw3Xw4EElJiZq1KhRcnV1VevWrXXLLbc4fYa3t7fatWtXrRwAAJhjPISMGTNGx44d09y5c1VUVKTQ0FBlZGQ4LlYtKChw2vmYM2eObDab5syZo6NHj6pDhw4aNWqU5s+fb2oKAACgAYyHEEmKi4tTXFxcje9lZWU5vW7RooWSkpKUlJRU5/5/2QcAADDvsrs7BgAAXBkIIQAAwIhL4nQMgCvfHY8VNFpfmxZ1abS+0EDP3NeIfa1uvL5wWWEnBAAAGEEIAQAARhBCAACAEYQQAABgBCEEAAAYQQgBAABGEEIAAIARPCcEAIAmxDNyasdOCAAAMIIQAgAAjCCEAAAAIwghAADACEIIAAAwghACAACMIIQAAAAjCCEAAMAIQggAADCCEAIAAIwghAAAACMIIQAAwAhCCAAAMIIQAgAAjCCEAAAAIwghAADACEIIAAAwghACAACMIIQAAAAjCCEAAMAIQggAADCCEAIAAIwghAAAACMIIQAAwIhLIoQsXLhQgYGB8vDwUHh4uHJycs5bPzU1VUFBQfL09FRAQIBmzJihU6dOOd5PTk7WrbfeqtatW6tjx44aPXq08vPzm3oaAACgHoyHkPT0dMXHxyspKUl5eXkKCQlRVFSUSkpKaqy/YsUKJSQkKCkpSfv27VNaWprS09M1e/ZsR53s7GxNmzZNn332mTZu3KgzZ85o+PDhqqioaK5pAQCAC2hhegApKSmaPHmyYmNjJUmLFy/WunXrtGTJEiUkJFSrv3XrVg0aNEjjxo2TJAUGBmrs2LHavn27o05GRoZTm6VLl6pjx47Kzc3VkCFDmnA2AACgrozuhJw+fVq5ubmKjIx0lLm4uCgyMlLbtm2rsc3AgQOVm5vrOGVz+PBhrV+/XiNGjKj1c8rKyiRJbdu2rfH9yspKlZeXOx0AAKBpGd0JKS0tVVVVlfz8/JzK/fz8tH///hrbjBs3TqWlpRo8eLAsy9LZs2c1depUp9MxP2e32zV9+nQNGjRIt9xyS411kpOTNW/evIubDAAAqBfj14TUV1ZWlhYsWKBFixYpLy9Pq1at0rp16/Tcc8/VWH/atGnas2ePVq5cWWufs2bNUllZmeM4cuRIUw0fAAD8P0Z3Qtq3by9XV1cVFxc7lRcXF8vf37/GNomJiZo4caImTZokSQoODlZFRYWmTJmip59+Wi4u/z9XxcXF6f3339cnn3yi6667rtZxuLu7y93dvRFmBAAA6sroToibm5v69eunzMxMR5ndbldmZqYGDBhQY5uTJ086BQ1JcnV1lSRZluX437i4OK1evVqbNm1St27dmmgGAACgoYzfHRMfH6+YmBj1799fYWFhSk1NVUVFheNumejoaHXu3FnJycmSpFGjRiklJUV9+vRReHi4Dh48qMTERI0aNcoRRqZNm6YVK1boH//4h1q3bq2ioiJJkq+vrzw9Pc1MFAAAODEeQsaMGaNjx45p7ty5KioqUmhoqDIyMhwXqxYUFDjtfMyZM0c2m01z5szR0aNH1aFDB40aNUrz58931HnttdckSREREU6f9eabb+qRRx5p8jkBAIALMx5CpJ+u3YiLi6vxvaysLKfXLVq0UFJSkpKSkmrt79xpGQAAcOm67O6OAQAAVwZCCAAAMIIQAgAAjCCEAAAAIwghAADACEIIAAAwghACAACMIIQAAAAjCCEAAMAIQggAADDiknhsOwCg6dzxWEGj9bVpUZdG6wtgJwQAABhBCAEAAEYQQgAAgBGEEAAAYAQhBAAAGEEIAQAARhBCAACAEYQQAABgBCEEAAAYQQgBAABGEEIAAIARhBAAAGAEIQQAABhBCAEAAEYQQgAAgBGEEAAAYMRFh5BTp041xjgAAMBVpkEhxG6367nnnlPnzp3VqlUrHT58WJKUmJiotLS0Rh0gAAC4MjUohPzxj3/U0qVL9eKLL8rNzc1Rfsstt+hvf/tbow0OAABcuRoUQpYtW6Y33nhD48ePl6urq6M8JCRE+/fvb7TBAQCAK1eDQsjRo0d1ww03VCu32+06c+bMRQ8KAABc+RoUQnr27KlPP/20Wvm7776rPn36XPSgAADAla9FQxrNnTtXMTExOnr0qOx2u1atWqX8/HwtW7ZM77//fmOPEQAAXIEatBNy77336p///Kc++ugjeXt7a+7cudq3b5/++c9/atiwYY09RgAAcAVq8HNCbrvtNm3cuFElJSU6efKkNm/erOHDhzeor4ULFyowMFAeHh4KDw9XTk7OeeunpqYqKChInp6eCggI0IwZM6o9r6S+fQIAgObVoBCyY8cObd++vVr59u3b9fnnn9err/T0dMXHxyspKUl5eXkKCQlRVFSUSkpKaqy/YsUKJSQkKCkpSfv27VNaWprS09M1e/bsBvcJAACaX4NCyLRp03TkyJFq5UePHtW0adPq1VdKSoomT56s2NhY9ezZU4sXL5aXl5eWLFlSY/2tW7dq0KBBGjdunAIDAzV8+HCNHTvWaaejvn0CAIDm16AQsnfvXvXt27daeZ8+fbR3794693P69Gnl5uYqMjLy/w/IxUWRkZHatm1bjW0GDhyo3NxcR+g4fPiw1q9frxEjRjS4z8rKSpWXlzsdAACgaTUohLi7u6u4uLhaeWFhoVq0qPsNN6WlpaqqqpKfn59TuZ+fn4qKimpsM27cOD377LMaPHiwWrZsqe7duysiIsJxOqYhfSYnJ8vX19dxBAQE1HkOAACgYRoUQoYPH65Zs2aprKzMUfbDDz9o9uzZTX53TFZWlhYsWKBFixYpLy9Pq1at0rp16/Tcc881uM9zczl31HSqCQAANK4GPSfkT3/6k4YMGaKuXbs6Hk62a9cu+fn56e23365zP+3bt5erq2u1XZXi4mL5+/vX2CYxMVETJ07UpEmTJEnBwcGqqKjQlClT9PTTTzeoT3d3d7m7u9d53AAA4OI1aCekc+fO2r17t1588UX17NlT/fr108svv6wvv/yyXqcy3Nzc1K9fP2VmZjrK7Ha7MjMzNWDAgBrbnDx5Ui4uzsM+9/s1lmU1qE8AAND8GrQTIkne3t6aMmXKRQ8gPj5eMTEx6t+/v8LCwpSamqqKigrFxsZKkqKjo9W5c2clJydLkkaNGqWUlBT16dNH4eHhOnjwoBITEzVq1ChHGLlQnwAAwLw6h5C1a9fq7rvvVsuWLbV27drz1r3nnnvqPIAxY8bo2LFjmjt3roqKihQaGqqMjAzHhaUFBQVOOx9z5syRzWbTnDlzdPToUXXo0EGjRo3S/Pnz69wnAAAwr84hZPTo0SoqKlLHjh01evToWuvZbDZVVVXVaxBxcXGKi4ur8b2srCyn1y1atFBSUpKSkpIa3CcAADCvziHEbrfX+G8AAICGqPc1IXa7XUuXLtWqVav0zTffyGaz6frrr9cDDzygiRMnymazNcU4AQDAFaZed8dYlqV77rlHkyZN0tGjRxUcHKxevXrpm2++0SOPPKL77ruvqcYJAACuMPXaCVm6dKk++eQTZWZm6vbbb3d6b9OmTRo9erSWLVum6OjoRh0kAAC48tRrJ+S///u/NXv27GoBRJLuuOMOJSQkaPny5Y02OAAAcOWqVwjZvXu37rrrrlrfv/vuu/XFF19c9KAAAMCVr14h5Pvvvz/vszb8/Px0/Pjxix4UAAC48tUrhFRVVZ33V3JdXV119uzZix4UAAC48tXrwlTLsvTII4/U+mNvlZWVjTIoAABw5atXCImJiblgHe6MAQAAdVGvEPLmm2821TgAAMBVpl7XhAAAADQWQggAADCCEAIAAIwghAAAACMIIQAAwAhCCAAAMIIQAgAAjCCEAAAAIwghAADACEIIAAAwghACAACMIIQAAAAjCCEAAMAIQggAADCCEAIAAIwghAAAACMIIQAAwAhCCAAAMIIQAgAAjCCEAAAAIwghAADACEIIAAAwghACAACMIIQAAAAjLokQsnDhQgUGBsrDw0Ph4eHKycmptW5ERIRsNlu1Y+TIkY46J06cUFxcnK677jp5enqqZ8+eWrx4cXNMBQAA1JHxEJKenq74+HglJSUpLy9PISEhioqKUklJSY31V61apcLCQsexZ88eubq66sEHH3TUiY+PV0ZGhv7+979r3759mj59uuLi4rR27drmmhYAALgA4yEkJSVFkydPVmxsrGPHwsvLS0uWLKmxftu2beXv7+84Nm7cKC8vL6cQsnXrVsXExCgiIkKBgYGaMmWKQkJCzrvDAgAAmpfREHL69Gnl5uYqMjLSUebi4qLIyEht27atTn2kpaXp4Ycflre3t6Ns4MCBWrt2rY4ePSrLsvTxxx/rwIEDGj58eI19VFZWqry83OkAAABNy2gIKS0tVVVVlfz8/JzK/fz8VFRUdMH2OTk52rNnjyZNmuRU/sorr6hnz5667rrr5ObmprvuuksLFy7UkCFDauwnOTlZvr6+jiMgIKDhkwIAAHVi/HTMxUhLS1NwcLDCwsKcyl955RV99tlnWrt2rXJzc/XnP/9Z06ZN00cffVRjP7NmzVJZWZnjOHLkSHMMHwCAq1oLkx/evn17ubq6qri42Km8uLhY/v7+521bUVGhlStX6tlnn3Uq/89//qPZs2dr9erVjjtmevfurV27dulPf/qT06mfc9zd3eXu7n6RswEAAPVhdCfEzc1N/fr1U2ZmpqPMbrcrMzNTAwYMOG/bd955R5WVlZowYYJT+ZkzZ3TmzBm5uDhPzdXVVXa7vfEGDwAALorRnRDpp9tpY2Ji1L9/f4WFhSk1NVUVFRWKjY2VJEVHR6tz585KTk52apeWlqbRo0erXbt2TuU+Pj4aOnSoZs6cKU9PT3Xt2lXZ2dlatmyZUlJSmm1eAADg/IyHkDFjxujYsWOaO3euioqKFBoaqoyMDMfFqgUFBdV2NfLz87V582Zt2LChxj5XrlypWbNmafz48fr+++/VtWtXzZ8/X1OnTm3y+QAAgLoxHkIkKS4uTnFxcTW+l5WVVa0sKChIlmXV2p+/v7/efPPNxhoeAABoApf13TEAAODyRQgBAABGEEIAAIARhBAAAGAEIQQAABhBCAEAAEYQQgAAgBGEEAAAYAQhBAAAGEEIAQAARhBCAACAEYQQAABgBCEEAAAYQQgBAABGEEIAAIARhBAAAGAEIQQAABhBCAEAAEYQQgAAgBGEEAAAYAQhBAAAGEEIAQAARhBCAACAEYQQAABgBCEEAAAYQQgBAABGEEIAAIARhBAAAGAEIQQAABhBCAEAAEYQQgAAgBGEEAAAYAQhBAAAGEEIAQAARhBCAACAEZdECFm4cKECAwPl4eGh8PBw5eTk1Fo3IiJCNput2jFy5Einevv27dM999wjX19feXt769Zbb1VBQUFTTwUAANSR8RCSnp6u+Ph4JSUlKS8vTyEhIYqKilJJSUmN9VetWqXCwkLHsWfPHrm6uurBBx901Dl06JAGDx6sm266SVlZWdq9e7cSExPl4eHRXNMCAAAX0ML0AFJSUjR58mTFxsZKkhYvXqx169ZpyZIlSkhIqFa/bdu2Tq9XrlwpLy8vpxDy9NNPa8SIEXrxxRcdZd27d2+iGQAAgIYwuhNy+vRp5ebmKjIy0lHm4uKiyMhIbdu2rU59pKWl6eGHH5a3t7ckyW63a926dbrxxhsVFRWljh07Kjw8XGvWrGmKKQAAgAYyGkJKS0tVVVUlPz8/p3I/Pz8VFRVdsH1OTo727NmjSZMmOcpKSkp04sQJPf/887rrrru0YcMG3Xfffbr//vuVnZ1dYz+VlZUqLy93OgAAQNMyfjrmYqSlpSk4OFhhYWGOMrvdLkm69957NWPGDElSaGiotm7dqsWLF2vo0KHV+klOTta8efOaZ9AAAECS4Z2Q9u3by9XVVcXFxU7lxcXF8vf3P2/biooKrVy5Ur/+9a+r9dmiRQv17NnTqfzmm2+u9e6YWbNmqayszHEcOXKkAbMBAAD1YTSEuLm5qV+/fsrMzHSU2e12ZWZmasCAAedt+84776iyslITJkyo1uett96q/Px8p/IDBw6oa9euNfbl7u4uHx8fpwMAADQt46dj4uPjFRMTo/79+yssLEypqamqqKhw3C0THR2tzp07Kzk52aldWlqaRo8erXbt2lXrc+bMmRozZoyGDBmi22+/XRkZGfrnP/+prKys5pgSAACoA+MhZMyYMTp27Jjmzp2roqIihYaGKiMjw3GxakFBgVxcnDds8vPztXnzZm3YsKHGPu+77z4tXrxYycnJeuKJJxQUFKT33ntPgwcPbvL5AACAujEeQiQpLi5OcXFxNb5X0+5FUFCQLMs6b5+PPvqoHn300cYYHgAAaALGn5gKAACuToQQAABgBCEEAAAYQQgBAABGEEIAAIARhBAAAGAEIQQAABhBCAEAAEYQQgAAgBGEEAAAYAQhBAAAGEEIAQAARhBCAACAEYQQAABgBCEEAAAYQQgBAABGEEIAAIARhBAAAGAEIQQAABhBCAEAAEYQQgAAgBGEEAAAYAQhBAAAGEEIAQAARhBCAACAEYQQAABgBCEEAAAYQQgBAABGEEIAAIARhBAAAGAEIQQAABhBCAEAAEYQQgAAgBGEEAAAYAQhBAAAGEEIAQAARlwSIWThwoUKDAyUh4eHwsPDlZOTU2vdiIgI2Wy2asfIkSNrrD916lTZbDalpqY20egBAEBDGA8h6enpio+PV1JSkvLy8hQSEqKoqCiVlJTUWH/VqlUqLCx0HHv27JGrq6sefPDBanVXr16tzz77TJ06dWrqaQAAgHoyHkJSUlI0efJkxcbGqmfPnlq8eLG8vLy0ZMmSGuu3bdtW/v7+jmPjxo3y8vKqFkKOHj2qxx9/XMuXL1fLli2bYyoAAKAejIaQ06dPKzc3V5GRkY4yFxcXRUZGatu2bXXqIy0tTQ8//LC8vb0dZXa7XRMnTtTMmTPVq1evC/ZRWVmp8vJypwMAADQtoyGktLRUVVVV8vPzcyr38/NTUVHRBdvn5ORoz549mjRpklP5Cy+8oBYtWuiJJ56o0ziSk5Pl6+vrOAICAuo+CQAA0CDGT8dcjLS0NAUHByssLMxRlpubq5dffllLly6VzWarUz+zZs1SWVmZ4zhy5EhTDRkAAPw/RkNI+/bt5erqquLiYqfy4uJi+fv7n7dtRUWFVq5cqV//+tdO5Z9++qlKSkrUpUsXtWjRQi1atNC///1v/e53v1NgYGCNfbm7u8vHx8fpAAAATctoCHFzc1O/fv2UmZnpKLPb7crMzNSAAQPO2/add95RZWWlJkyY4FQ+ceJE7d69W7t27XIcnTp10syZM/Xhhx82yTwAAED9tTA9gPj4eMXExKh///4KCwtTamqqKioqFBsbK0mKjo5W586dlZyc7NQuLS1No0ePVrt27ZzK27VrV62sZcuW8vf3V1BQUNNOBgAA1JnxEDJmzBgdO3ZMc+fOVVFRkUJDQ5WRkeG4WLWgoEAuLs4bNvn5+dq8ebM2bNhgYsgAAKARGA8hkhQXF6e4uLga38vKyqpWFhQUJMuy6tz/N99808CRAQCApnJZ3x0DAAAuX4QQAABgBCEEAAAYQQgBAABGEEIAAIARhBAAAGAEIQQAABhBCAEAAEYQQgAAgBGEEAAAYAQhBAAAGEEIAQAARhBCAACAEYQQAABgBCEEAAAYQQgBAABGEEIAAIARhBAAAGAEIQQAABhBCAEAAEYQQgAAgBGEEAAAYAQhBAAAGEEIAQAARhBCAACAEYQQAABgBCEEAAAYQQgBAABGEEIAAIARhBAAAGAEIQQAABhBCAEAAEYQQgAAgBGEEAAAYAQhBAAAGHFJhJCFCxcqMDBQHh4eCg8PV05OTq11IyIiZLPZqh0jR46UJJ05c0Z/+MMfFBwcLG9vb3Xq1EnR0dH69ttvm2s6AACgDoyHkPT0dMXHxyspKUl5eXkKCQlRVFSUSkpKaqy/atUqFRYWOo49e/bI1dVVDz74oCTp5MmTysvLU2JiovLy8rRq1Srl5+frnnvuac5pAQCAC2hhegApKSmaPHmyYmNjJUmLFy/WunXrtGTJEiUkJFSr37ZtW6fXK1eulJeXlyOE+Pr6auPGjU51Xn31VYWFhamgoEBdunRpopkAAID6MLoTcvr0aeXm5ioyMtJR5uLiosjISG3btq1OfaSlpenhhx+Wt7d3rXXKyspks9nUpk2bGt+vrKxUeXm50wEAAJqW0RBSWlqqqqoq+fn5OZX7+fmpqKjogu1zcnK0Z88eTZo0qdY6p06d0h/+8AeNHTtWPj4+NdZJTk6Wr6+v4wgICKjfRAAAQL0ZvybkYqSlpSk4OFhhYWE1vn/mzBk99NBDsixLr732Wq39zJo1S2VlZY7jyJEjTTVkAADw/xi9JqR9+/ZydXVVcXGxU3lxcbH8/f3P27aiokIrV67Us88+W+P75wLIv//9b23atKnWXRBJcnd3l7u7e/0nAAAAGszoToibm5v69eunzMxMR5ndbldmZqYGDBhw3rbvvPOOKisrNWHChGrvnQsgX331lT766CO1a9eu0ccOAAAujvG7Y+Lj4xUTE6P+/fsrLCxMqampqqiocNwtEx0drc6dOys5OdmpXVpamkaPHl0tYJw5c0a/+tWvlJeXp/fff19VVVWO60vatm0rNze35pkYAAA4L+MhZMyYMTp27Jjmzp2roqIihYaGKiMjw3GxakFBgVxcnDds8vPztXnzZm3YsKFaf0ePHtXatWslSaGhoU7vffzxx4qIiGiSeQAAgPoxHkIkKS4uTnFxcTW+l5WVVa0sKChIlmXVWD8wMLDW9wAAwKXjsr47BgAAXL4IIQAAwAhCCAAAMIIQAgAAjCCEAAAAIwghAADACEIIAAAwghACAACMIIQAAAAjCCEAAMAIQggAADCCEAIAAIwghAAAACMIIQAAwAhCCAAAMIIQAgAAjCCEAAAAIwghAADACEIIAAAwghACAACMIIQAAAAjCCEAAMAIQggAADCCEAIAAIwghAAAACMIIQAAwAhCCAAAMIIQAgAAjCCEAAAAIwghAADACEIIAAAwghACAACMIIQAAAAjCCEAAMAIQggAADDikgghCxcuVGBgoDw8PBQeHq6cnJxa60ZERMhms1U7Ro4c6ahjWZbmzp2ra6+9Vp6enoqMjNRXX33VHFMBAAB1ZDyEpKenKz4+XklJScrLy1NISIiioqJUUlJSY/1Vq1apsLDQcezZs0eurq568MEHHXVefPFF/eUvf9HixYu1fft2eXt7KyoqSqdOnWquaQEAgAswHkJSUlI0efJkxcbGqmfPnlq8eLG8vLy0ZMmSGuu3bdtW/v7+jmPjxo3y8vJyhBDLspSamqo5c+bo3nvvVe/evbVs2TJ9++23WrNmTTPODAAAnE8Lkx9++vRp5ebmatasWY4yFxcXRUZGatu2bXXqIy0tTQ8//LC8vb0lSV9//bWKiooUGRnpqOPr66vw8HBt27ZNDz/8cLU+KisrVVlZ6XhdVlYmSSovL2/QvGpz9vSPjdaXY2yVZxqtTzXyfC81rL9ZrL85rL1ZV9v6nxujZVkXrmwZdPToUUuStXXrVqfymTNnWmFhYRdsv337dkuStX37dkfZli1bLEnWt99+61T3wQcftB566KEa+0lKSrIkcXBwcHBwcDTSceTIkQt+jxvdCblYaWlpCg4OVlhY2EX1M2vWLMXHxzte2+12ff/992rXrp1sNtvFDrNJlJeXKyAgQEeOHJGPj4/p4Vx1WH+zWH9zWHuzLof1tyxLP/74ozp16nTBukZDSPv27eXq6qri4mKn8uLiYvn7+5+3bUVFhVauXKlnn33Wqfxcu+LiYl177bVOfYaGhtbYl7u7u9zd3Z3K2rRpU8dZmOXj43PJ/iFeDVh/s1h/c1h7sy719ff19a1TPaMXprq5ualfv37KzMx0lNntdmVmZmrAgAHnbfvOO++osrJSEyZMcCrv1q2b/P39nfosLy/X9u3bL9gnAABoPsZPx8THxysmJkb9+/dXWFiYUlNTVVFRodjYWElSdHS0OnfurOTkZKd2aWlpGj16tNq1a+dUbrPZNH36dP3xj39Ujx491K1bNyUmJqpTp04aPXp0c00LAABcgPEQMmbMGB07dkxz585VUVGRQkNDlZGRIT8/P0lSQUGBXFycN2zy8/O1efNmbdiwocY+f//736uiokJTpkzRDz/8oMGDBysjI0MeHh5NPp/m4u7urqSkpGqnkdA8WH+zWH9zWHuzrrT1t1lWXe6hAQAAaFzGH1YGAACuToQQAABgBCEEAAAYQQgBAABGEELq6dixY/rtb3+rLl26yN3dXf7+/oqKitKWLVtMD+2CCgoKNHLkSHl5ealjx46aOXOmzp4961QnKytLffv2lbu7u2644QYtXbrU6f3k5GTdeuutat26tTp27KjRo0crPz+/2eZwta//a6+9pt69ezseVDRgwAB98MEHzTL+K3ntCwsLNW7cON14441ycXHR9OnTq/UREREhm81W7Rg5cmSzzOFKXv+f27Jli1q0aFHt4ZI//vijpk+frq5du8rT01MDBw7Ujh07mmH0V/baZ2Vl1fh3XVRU5KjTlGtv/Bbdy80DDzyg06dP66233tL111+v4uJiZWZm6rvvvjM9tPOqqqrSyJEj5e/vr61bt6qwsFDR0dFq2bKlFixYIOmnH/8bOXKkpk6dquXLlyszM1OTJk3Stddeq6ioKElSdna2pk2bpltvvVVnz57V7NmzNXz4cO3du9fxI4JN6Wpf/+uuu07PP/+8evToIcuy9NZbb+nee+/Vzp071atXryadw5W89pWVlerQoYPmzJmjl156qcZ+Vq1apdOnTztef/fddwoJCXH8gndTu5LX/5wffvhB0dHRuvPOO6s9SXvSpEnas2eP3n77bXXq1El///vfFRkZqb1796pz585NOoerYe3z8/OdnsDasWNHx7+bdO0v+OsycDh+/LglycrKyrpgvSlTplgdO3a03N3drV69eln//Oc/He9/+umn1uDBgy0PDw/ruuuusx5//HHrxIkTjve7du1qzZ8/34qNjbVatWplBQQEWK+//rrTZxQUFFgPPvig5evra11zzTXWPffcY3399de1jmn9+vWWi4uLVVRU5Ch77bXXLB8fH6uystKyLMv6/e9/b/Xq1cup3ZgxY6yoqKha+y0pKbEkWdnZ2eddk8bA+tfsmmuusf72t7+dt87FutLX/ueGDh1qPfnkkxdYEct66aWXrNatWzuNv6lcLes/ZswYa86cOVZSUpIVEhLiKD958qTl6upqvf/++071+/btaz399NPnXZOLdaWv/ccff2xJso4fP15jH0299pyOqYdWrVqpVatWWrNmjSorK2usY7fbdffdd2vLli36+9//rr179+r555+Xq6urJOnQoUO666679MADD2j37t1KT0/X5s2bFRcX59TPn//8Z/Xv3187d+7UY489pt/+9reO0x5nzpxRVFSUWrdurU8//VRbtmxRq1atdNdddzn9P7Wf27Ztm4KDgx0PgZOkqKgolZeX61//+pejTmRkpFO7qKgobdu2rdY1KSsrkyS1bdv2fEvXKFh/Z1VVVVq5cqUqKiqa/CcJrvS1b4i0tDQ9/PDDzbIDeDWs/5tvvqnDhw8rKSmpWh9nz55VVVVVtQdOenp6avPmzXVYwYa7GtZekkJDQ3Xttddq2LBhTqeZmnztLzrGXGXeffdd65prrrE8PDysgQMHWrNmzbK++OILx/sffvih5eLiYuXn59fY/te//rU1ZcoUp7JPP/3UcnFxsf7zn/9YlvVTIp4wYYLjfbvdbnXs2NF67bXXLMuyrLffftsKCgqy7Ha7o05lZaXl6elpffjhhzV+7uTJk63hw4c7lVVUVFiSrPXr11uWZVk9evSwFixY4FRn3bp1liTr5MmT1fqsqqqyRo4caQ0aNKjGz2wKrL9l7d692/L29rZcXV0tX19fa926dTUvViO7ktf+5+qyE7J9+3ZLkrV9+/bz1mtMV/L6HzhwwOrYsaNj7L/cCbEsyxowYIA1dOhQ6+jRo9bZs2ett99+23JxcbFuvPHGWtessVzJa79//35r8eLF1ueff25t2bLFio2NtVq0aGHl5uY62jTl2rMTUk8PPPCAvv32W61du1Z33XWX40LCcxcQ7tq1S9ddd51uvPHGGtt/8cUXWrp0qSNdt2rVSlFRUbLb7fr6668d9Xr37u34t81mk7+/v0pKShx9HDx4UK1bt3b00bZtW506dUqHDh1qusn/wrRp07Rnzx6tXLmy2T6T9ZeCgoK0a9cubd++Xb/97W8VExOjvXv3Nvnnsvb/X1pamoKDgxUWFtZsn3mlrn9VVZXGjRunefPm1Tp2SXr77bdlWZY6d+4sd3d3/eUvf9HYsWOr/axHU7hS11766b8nv/nNb9SvXz8NHDhQS5Ys0cCBA52ujWrKtefC1Abw8PDQsGHDNGzYMCUmJmrSpElKSkrSI488Ik9Pz/O2PXHihH7zm9/oiSeeqPZely5dHP9u2bKl03s2m012u93RR79+/bR8+fJqfXTo0KHGz/X391dOTo5T2bkLv/z9/R3/+8uLwYqLi+Xj41NtXnFxcXr//ff1ySef6LrrrqvxM5vK1b7+bm5uuuGGGyRJ/fr1044dO/Tyyy/r9ddfr3nSjehKXfv6qKio0MqVK/Xss8/Wu+3FuhLX/8cff9Tnn3+unTt3Ok5P2O12WZalFi1aaMOGDbrjjjvUvXt3ZWdnq6KiQuXl5br22ms1ZswYXX/99eedd2O5Ete+NmFhYU6nWppy7QkhjaBnz55as2aNpJ+S7P/+7//qwIEDNabivn37au/evY4vkYbo27ev0tPT1bFjR6ermc9nwIABmj9/vkpKShxXPW/cuFE+Pj7q2bOno8769eud2m3cuNHpegPLsvT4449r9erVysrKUrdu3Ro8j8ZyNa1/Tex2e63nqpvalbL29fHOO++osrJSEyZMqHfbxnYlrH/Lli315ZdfOrVZtGiRNm3apHfffbfaf2O8vb3l7e2t48eP68MPP9SLL77Y4PlcjCth7Wuza9cuXXvttdXKm2TtL/qEzlWktLTUuv322623337b+uKLL6zDhw9b//M//2P5+flZjz76qKNeRESEdcstt1gbNmywDh8+bK1fv9764IMPLMuyrC+++MLy9PS0pk2bZu3cudM6cOCAtWbNGmvatGmO9l27drVeeuklp88OCQmxkpKSLMv66Zxejx49rIiICOuTTz6xDh8+bH388cfW448/bh05cqTGsZ89e9a65ZZbrOHDh1u7du2yMjIyrA4dOlizZs1y1Dl8+LDl5eVlzZw509q3b5+1cOFCy9XV1crIyHDU+e1vf2v5+vpaWVlZVmFhoeOo6ZqRxsb6W1ZCQoKVnZ1tff3119bu3buthIQEy2azWRs2bLjY5T2vK33tLcuydu7cae3cudPq16+fNW7cOGvnzp3Wv/71r2r9DR482BozZkxDlrHBrob1/7margnJyMiwPvjgA+vw4cPWhg0brJCQECs8PNw6ffp0PVay/q70tX/ppZesNWvWWF999ZX15ZdfWk8++aTl4uJiffTRR446Tbn2hJB6OHXqlJWQkGD17dvX8vX1tby8vKygoCBrzpw5Tl/C3333nRUbG2u1a9fO8vDwsG655Ran25tycnKsYcOGWa1atbK8vb2t3r17W/Pnz3e8f6E/RsuyrMLCQis6Otpq37695e7ubl1//fXW5MmTrbKyslrH/80331h333235enpabVv39763e9+Z505c8apzscff2yFhoZabm5u1vXXX2+9+eabTu9LqvH4Zb2mwPpb1qOPPmp17drVcnNzszp06GDdeeedTR5ALOvqWPua/q67du3qVGf//v2WpGZZ85+7Gtb/52oKIenp6db1119vubm5Wf7+/ta0adOsH3744QIrd/Gu9LV/4YUXrO7du1seHh5W27ZtrYiICGvTpk1OfTTl2tssy7Iufj8FAACgfrg7BgAAGEEIAQAARhBCAACAEYQQAABgBCEEAAAYQQgBAABGEEIAAIARhBAAAGAEIQQAABhBCAEAAEYQQgAAgBGEEAAAYMT/Bc47oC9bHeuMAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<Figure size 600x600 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "xlabels = [s.split('-')[-1].strip() for s in sample_ids]\n",
    "fig, ax = plt.subplots(1, figsize=(6,6))\n",
    "\n",
    "ax.set(ylim=[0.70,0.9])\n",
    "\n",
    "ax.bar(np.arange(5)-0.11, dices, color='royalblue', width=0.2)\n",
    "ax.bar(np.arange(5)+0.11, w_dices, color='coral', width=0.2)\n",
    "\n",
    "plt.setp(ax, xticks=np.arange(5), xticklabels=xlabels)\n",
    "\n",
    "labels = ['equal weights', 'diff weights']\n",
    "handles = [plt.Rectangle((0,0),1,1, color=c) for c in ['royalblue','coral']]\n",
    "\n",
    "plt.yticks(np.arange(0.7, 0.9, 0.02))\n",
    "plt.legend(handles, labels)\n",
    "plt.ylabel('Dice')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "9212cba5",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python [conda env:vit-mesmer] *",
   "language": "python",
   "name": "conda-env-vit-mesmer-py"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.8.12"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}