naive_bayes_notebook.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "eae851b7-8b70-4a35-80fb-b7c548032b80",
   "metadata": {},
   "source": [
    "# Course 3 - Naive Bayes"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "75105c54-aac1-4011-9a59-00ab4cc31798",
   "metadata": {},
   "source": [
    "## Import librairies and dataset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "id": "6e5db8ef-d13e-459e-9843-d1424f42b23b",
   "metadata": {},
   "outputs": [],
   "source": [
    "#!pip install pandas seaborn matplotlib scikit-learn gensim eli5 tiktoken spacy"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "a685ce16-8bbd-4493-a961-e4004f2526e3",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/gcortal/anaconda3/envs/epita/lib/python3.9/site-packages/tqdm/auto.py:21: TqdmWarning: IProgress not found. Please update jupyter and ipywidgets. See https://ipywidgets.readthedocs.io/en/stable/user_install.html\n",
      "  from .autonotebook import tqdm as notebook_tqdm\n"
     ]
    }
   ],
   "source": [
    "import pandas as pd\n",
    "import seaborn as sns\n",
    "import matplotlib.pyplot as plt\n",
    "from datasets import load_dataset\n",
    "from nltk import word_tokenize          \n",
    "from nltk.stem import WordNetLemmatizer \n",
    "import tiktoken\n",
    "\n",
    "from sklearn.model_selection import train_test_split, cross_val_score, GridSearchCV, StratifiedKFold\n",
    "from sklearn.feature_extraction.text import CountVectorizer, TfidfVectorizer, TfidfTransformer\n",
    "from sklearn.naive_bayes import MultinomialNB\n",
    "from sklearn.pipeline import make_pipeline\n",
    "from sklearn.metrics import classification_report, accuracy_score, ConfusionMatrixDisplay, confusion_matrix\n",
    "from sklearn.utils.class_weight import compute_sample_weight\n",
    "from sklearn.linear_model import LogisticRegression\n",
    "from sklearn.preprocessing import StandardScaler\n",
    "from sklearn.ensemble import VotingClassifier\n",
    "\n",
    "from spacy.lang.fr.stop_words import STOP_WORDS as fr_stop\n",
    "from spacy.lang.en.stop_words import STOP_WORDS as en_stop"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "9592edd5-f84f-4165-b7e9-fdd4ac3947a3",
   "metadata": {},
   "outputs": [],
   "source": [
    "dataset = load_dataset(\"gustavecortal/DreamBank-annotated\")\n",
    "df = dataset[\"train\"].to_pandas()\n",
    "df = df.sample(frac = 1)\n",
    "df[\"report\"] = df[\"report\"].astype(str)\n",
    "df = df.query(\"emotion == 'CO D' or emotion == 'AN D' or emotion == 'SD D' or emotion == 'AP D' or emotion == 'HA D'\")[[\"report\", \"emotion\"]]\n",
    "df.head()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "3a162457-ae00-48c0-bf6d-ab2c38ddec5f",
   "metadata": {},
   "source": [
    "## Preprocessing and visualization"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "59efd75a-6d2a-4709-a2e7-1bc4d355b4ec",
   "metadata": {},
   "source": [
    "Convert code to emotion class according to the annotation guideline : https://dreams.ucsc.edu/Coding/emotions.html"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "e0af0d38-75fb-45f7-9103-4cb0c6c41139",
   "metadata": {},
   "outputs": [],
   "source": [
    "def to_emotion_class(emotion):\n",
    "    \n",
    "    if emotion == \"AP D\":\n",
    "        return \"fear\"\n",
    "    elif emotion == \"CO D\":\n",
    "        return \"confusion\"\n",
    "    elif emotion == \"SD D\":\n",
    "        return \"sadness\"\n",
    "    elif emotion == \"AN D\":\n",
    "        return \"anger\"\n",
    "    else:\n",
    "        return \"happiness\"\n",
    "\n",
    "df[\"emotion\"] = df[\"emotion\"].apply(lambda x: to_emotion_class(x))\n",
    "df[\"report\"] = df[\"report\"].apply(lambda x: x.lower()) # simple preprocessing (lowercase only)\n",
    "df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "3df000fd-33c6-4b49-b53c-c26a005dded3",
   "metadata": {},
   "outputs": [],
   "source": [
    "plt.figure(figsize=(10, 6))\n",
    "sns.countplot(x='emotion', data=df)\n",
    "\n",
    "plt.title('Distribution of emotions') \n",
    "plt.xlabel('Emotion')\n",
    "plt.ylabel('Count')\n",
    "plt.xticks(rotation=45)\n",
    "\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "c46e5ddf-c780-43e6-b5e0-a77d6a7f66c7",
   "metadata": {},
   "source": [
    "## Training and test sets"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "cdc43018-b3db-46f9-9a3d-1695a62307e1",
   "metadata": {},
   "outputs": [],
   "source": [
    "X = df['report']\n",
    "y = df['emotion']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "144dcfd1-a7df-4a8c-a4b8-e1d301afa2a7",
   "metadata": {},
   "outputs": [],
   "source": [
    "X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)\n",
    "sample_weights = compute_sample_weight(class_weight='balanced', y=y_train)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "6f8f1353-cd7b-4e0b-afc7-6184e1a86888",
   "metadata": {},
   "source": [
    "## Tokenization functions"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "c76949c3-2986-4f74-91a1-8782ed3af58e",
   "metadata": {},
   "outputs": [],
   "source": [
    "def lemma_tokenize(doc):\n",
    "    wnl = WordNetLemmatizer()\n",
    "    return [wnl.lemmatize(t) for t in word_tokenize(doc)]\n",
    "\n",
    "def char_tokenize(doc):\n",
    "    return [char for char in doc]\n",
    "\n",
    "def byte_tokenize(doc):\n",
    "    tokens = doc.encode(\"utf-8\")\n",
    "    tokens = list(map(int, tokens))\n",
    "    return [str(token) for token in tokens]\n",
    "\n",
    "def gpt_tokenize(doc):\n",
    "    enc = tiktoken.encoding_for_model(\"gpt-4\")\n",
    "    tokens = enc.encode(doc)\n",
    "    return [str(token) for token in tokens]"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "7ac26f4c-aaf1-4312-8420-b029d3925ab0",
   "metadata": {},
   "source": [
    "## Training: Bag of Words (CountVectorizer) and Naive Bayes (MultinomialNB)\n",
    "\n",
    "https://scikit-learn.org/stable/tutorial/text_analytics/working_with_text_data.html\n",
    "\n",
    "https://scikit-learn.org/stable/modules/generated/sklearn.naive_bayes.MultinomialNB.html\n",
    "\n",
    "https://scikit-learn.org/stable/modules/generated/sklearn.feature_extraction.text.CountVectorizer.html"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "3f3b13ac-1ff0-4fd7-af59-39884f7ad83e",
   "metadata": {},
   "outputs": [],
   "source": [
    "model = make_pipeline(CountVectorizer(ngram_range = (1,1)), MultinomialNB())\n",
    "#model = make_pipeline(CountVectorizer(ngram_range = (1,1), stop_words = en_stop), MultinomialNB())\n",
    "#model = make_pipeline(CountVectorizer(tokenizer=word_tokenize, ngram_range = (1,1)), MultinomialNB())\n",
    "#model = make_pipeline(CountVectorizer(tokenizer=lemma_tokenize, ngram_range = (1,1)), MultinomialNB())\n",
    "#model = make_pipeline(CountVectorizer(tokenizer=char_tokenize, ngram_range = (1,1)), MultinomialNB())\n",
    "#model = make_pipeline(CountVectorizer(tokenizer=char_tokenize, ngram_range = (2,2)), MultinomialNB())\n",
    "#model = make_pipeline(CountVectorizer(tokenizer=char_tokenize, ngram_range = (3,3)), MultinomialNB())\n",
    "#model = make_pipeline(CountVectorizer(tokenizer=byte_tokenize, ngram_range = (1,1)), MultinomialNB())\n",
    "#model = make_pipeline(CountVectorizer(tokenizer=byte_tokenize, ngram_range = (2,2)), MultinomialNB())\n",
    "#model = make_pipeline(CountVectorizer(tokenizer=byte_tokenize, ngram_range = (3,3)), MultinomialNB())\n",
    "#model = make_pipeline(CountVectorizer(tokenizer=gpt_tokenize, ngram_range = (1,1)), MultinomialNB())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "fd8370cd-706b-46bb-9b23-f2837eaa64b7",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Features dimension: (4520, 21298)\n",
      "Classification Report:\n",
      "               precision    recall  f1-score   support\n",
      "\n",
      "       anger       0.78      0.27      0.41       186\n",
      "   confusion       0.41      0.31      0.35       203\n",
      "        fear       0.50      0.87      0.63       431\n",
      "   happiness       0.49      0.42      0.45       184\n",
      "     sadness       0.56      0.04      0.07       126\n",
      "\n",
      "    accuracy                           0.50      1130\n",
      "   macro avg       0.55      0.38      0.38      1130\n",
      "weighted avg       0.54      0.50      0.45      1130\n",
      "\n"
     ]
    }
   ],
   "source": [
    "model.fit(X_train, y_train)\n",
    "#model.fit(X_train, y_train, multinomialnb__sample_weight=sample_weights)\n",
    "\n",
    "y_pred = model.predict(X_test)\n",
    "\n",
    "print(f\"Features dimension: {len(X_train), len(model[0].vocabulary_)}\")\n",
    "print(\"Classification Report:\\n\", classification_report(y_test, y_pred))"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "8a44084f-57db-4211-916e-07eab0d94b1a",
   "metadata": {},
   "source": [
    "## Plot confusion matrix"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "a56408c8-d136-426f-8eb9-4e8252e4eaa2",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAjgAAAGwCAYAAACkfh/eAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjguMywgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy/H5lhTAAAACXBIWXMAAA9hAAAPYQGoP6dpAAB3oUlEQVR4nO3dd1hTZxsG8DvsGZbKUESQrYhbqYoLxdlatVprFXfdVauidYHW0s89q621ov20al2te+8tihNREAQFBEW2rOR8f/AZG0EFEwiE+3dd57o857zn5EkMyZN3igRBEEBERESkRjRUHQARERGRsjHBISIiIrXDBIeIiIjUDhMcIiIiUjtMcIiIiEjtMMEhIiIitcMEh4iIiNSOlqoDoJKTSqWIi4uDsbExRCKRqsMhIqISEAQB6enpsLGxgYZG6dUzZGdnIzc3Vyn30tHRgZ6enlLuVVaY4FRAcXFxsLW1VXUYRESkgNjYWNSoUaNU7p2dnQ17OyMkJEqUcj8rKytERUVVqCSHCU4FZGxsDADwdv8WWpq6Ko6mfBM9jlN1CBVCqo+rqkOoEEyO3Vd1CBWCJCNL1SGUa/lCHs4Je2Wf5aUhNzcXCYkSPA6pBbGxYrVEaelS2DWKRm5uLhMcKl2vm6W0NHWZ4HyASKSj6hAqBC3tivOhpUpafD8Vi0iUp+oQyj8BZdLFwMhYBCNjxR5HiorZFYIJDhERkZqSCFJIFFxxUiJIlRNMGWOCQ0REpKakECCFYhmOoterCoeJExERkdphDQ4REZGakkIKRRuYFL+DajDBISIiUlMSQYBEUKyJSdHrVYVNVERERKR2WINDRESkpipzJ2MmOERERGpKCgGSSprgsImKiIiI1A5rcIiIiNQUm6iIiIhI7XAUFREREZEaYQ0OERGRmpL+f1P0HhURExwiIiI1JVHCKCpFr1cVJjhERERqSiJACauJKyeWssY+OERERKR2mOAQERGpKamStuJas2YN6tWrB7FYDLFYDC8vLxw8eFB2vk2bNhCJRHLbyJEj5e4RExODrl27wsDAANWqVcOUKVOQn59f4ufOJioiIiI1JYUIEogUvkdx1ahRAz/99BOcnJwgCAI2btyIzz77DDdu3ECdOnUAAMOHD8fcuXNl1xgYGMj+LZFI0LVrV1hZWeHChQuIj4/HwIEDoa2tjR9//LFEcTPBISIiIqXo3r273P78+fOxZs0aXLp0SZbgGBgYwMrKqsjrjxw5gnv37uHYsWOwtLRE/fr1MW/ePPj7+yMgIAA6OjrFjoVNVERERGpKKihnA4C0tDS5LScn572PLZFIsHXrVmRmZsLLy0t2fPPmzahSpQrq1q2L6dOnIysrS3bu4sWL8PDwgKWlpeyYr68v0tLScPfu3RI9d9bgEBERqSmJEpqoXl9va2srd3zOnDkICAgoVP727dvw8vJCdnY2jIyMsHv3bri7uwMAvvrqK9jZ2cHGxga3bt2Cv78/wsPDsWvXLgBAQkKCXHIDQLafkJBQoriZ4BAREdEHxcbGQiwWy/Z1dXWLLOfi4oLQ0FCkpqZix44d8PPzw+nTp+Hu7o4RI0bIynl4eMDa2hrt27dHZGQkateurdR4meAQERGpKWXW4LweGfUhOjo6cHR0BAA0atQIV69exfLly/HLL78UKtusWTMAQEREBGrXrg0rKytcuXJFrsyzZ88A4J39dt6FfXCIiIjUlFQQKWVTKAap9J39dUJDQwEA1tbWAAAvLy/cvn0biYmJsjJHjx6FWCyWNXMVF2twiIiISCmmT5+Ozp07o2bNmkhPT8eWLVtw6tQpHD58GJGRkdiyZQu6dOkCCwsL3Lp1CxMnToS3tzfq1asHAOjYsSPc3d0xYMAALFiwAAkJCZg5cybGjBnzziaxd2GCQ0REpKaU2URVHImJiRg4cCDi4+NhYmKCevXq4fDhw+jQoQNiY2Nx7NgxLFu2DJmZmbC1tUWvXr0wc+ZM2fWamprYt28fRo0aBS8vLxgaGsLPz09u3pziYoJDRESkpiTQgETB3iiSEpRdv379O8/Z2tri9OnTH7yHnZ0dDhw4UIJHLRoTHCIiIjUlKKEPjaDg9arCTsZERESkdliDQ8XWv/9tfP21/EySsbHGGDGiKwCgc+cItGnzGI6OL2FgkI/evXsiM7P402qri7qNU9Fr6BM41smARbVczBvjhovHq8iVsXXIwuDJUfBokgpNTQExkQaYP94NSfF6Koq69Hk6xOGrdjfhavscVUyyMG19R5y9bf+vEgKGdb6G7s3vw1g/B7eirLDor1Z48txEVmJgh+v4xD0GTtVfIE+igU7TB5f9EyljH3o/TQwKR4fPE+WuuXbWDLOH1y3rUMuVbgOS0HVgEixr5AIAHj/Qx+ZlVrh20uQDV6qXsu6DU54wwaESiY42wffft5HtSyRvKgF1dSW4ds0a165ZY8iQWyqIrnzQ05cg6r4hjuy0xKxVYYXOW9m+wsItN3FkhxX+u9IOWRmasHPMQm6Oeleo6uvmIyLOAvsvuyJo6JFC5/u3v4ne3nfww+a2iH9hjOFdrmLJyP34+qc+yM0v+KjS1pTgZKgD7kRbolvz+2X9FFTiQ+8nALh2xgxLv3eW7eflVswvJGVKitfG70HV8TRKFyIAHb54gYD1jzCmkyseP9BXdXhlRiJoQCIo2AdHUFIwZYwJTjmSl5cHbW1tVYfxXhKJCC9fFv3hsGePCwDAw+NZWYZU7lw7a45rZ83fed5vQjSunTbH74ve1F4kxKr/B+6lsJq4FFbzHWcF9PG+jY1HGuLcnVoAgHmb22LvvD/QyiMax28UTBq2/lATAECXpuFlEHH58KH3EwDk5Wrg5fPKV1v6PpePmcrtBy+ojm4Dn8O1YWalSnAqM/X+yfgOhw4dQsuWLWFqagoLCwt069YNkZGRAIDo6GiIRCLs2rULbdu2hYGBATw9PXHx4kW5e6xbtw62trYwMDDA559/jiVLlsDU1FSuzN9//42GDRtCT08PDg4OCAwMRH5+vuy8SCTCmjVr8Omnn8LQ0BDz588v9eeuqOrV0/Hf/+7B77/vxdSpF1G1aqaqQ6pQRCIBTdq8xNNofcz77Ta2nL+EpdtC4dX+uapDUykbi3RUMcnCtQfVZccys3Vx73E11K1VuRPm4vBomoIt5y/h14PXMGbOQxib5qk6pHJFQ0NA60+ToasvRViIoarDKVNSiCCFhoJbxawRrJQJTmZmJiZNmoRr167h+PHj0NDQwOeffw6pVCorM2PGDEyePBmhoaFwdnZGv379ZMnJ+fPnMXLkSHz77bcIDQ1Fhw4dCiUnZ8+excCBA/Htt9/i3r17+OWXXxAcHFyoXEBAAD7//HPcvn0bQ4YMKTLenJycQqu4qkJ4uAUWL26GmTPbYNWqxrC0zMDChcehr88P0+IytciDgaEEXwyPRchZc8wcWhcXjllgxsow1G2SourwVMbcuGA14eR0+V/Wyen6sBBnFXUJ/V/IWTMs9nfB94M9sGFRLXg0ScXcX+9AQ6OCtisoUS3XV9gTHop9j25gfFAs5g53QMzDylV787oPjqJbRVQpm6h69eolt//777+jatWquHfvHoyMjAAAkydPRteuBZ1nAwMDUadOHURERMDV1RUrV65E586dMXnyZACAs7MzLly4gH379snuGRgYiGnTpsHPzw8A4ODggHnz5mHq1KmYM2eOrNxXX32FwYPf31EyKCgIgYGBij9xBV27ZiP7d3S0KcLDLbBx4160ahWDI0eUu0iauhL9/0vn0gkL7NlYUFvx6L4R3BqkocuXCbhz1VSF0VFFdOZANdm/ox8YIircEL8fuwaPpim4eclMhZGp3pNIXYz2dYWBsRStur7E5KWPMaW3U6VLciqrSlmD8/DhQ/Tr1w8ODg4Qi8WoVasWACAmJkZW5vW00cCbNTJer40RHh6Opk2byt3z7f2bN29i7ty5MDIykm3Dhw9HfHw8srLe/CJt3LjxB+OdPn06UlNTZVtsbGzJnnApyczUwdOnxrCxyVB1KBVG2ktt5OeJEBNhIHc8NtIA1ayLXqulMkhOL3g9zI1fyR03N36FF2kGRV1C75DwRB+pyVqwsctWdSgql5+ngbhoPUTcNsCGn6oj6p4+egxNUnVYZep1J2NFt4qoUtbgdO/eHXZ2dli3bh1sbGwglUpRt25d5Obmysr8u7OvSFRQPffvJqwPycjIQGBgIHr27FnonJ7em6HAhoYfbg/W1dUt8RocZUFPLw/W1hk4fryWqkOpMPLzNPDgjhFq2Mt/kVev9QqJceXv/7isxL0wxvNUAzRyeoqHTwuGQBvo5sLdLhG7z5dsgb3KzsIyB8am+UhOZKfjt4k0BGjrFP9zXB0U9MFRcLFMNlFVDC9evEB4eDjWrVuHVq1aAQDOnTtXonu4uLjg6tWrcsfe3m/YsCHCw8NlS8arg2HDbuDy5ep49swAFhbZ+Prr25BKRTh9umBkjJnZK5iZZctqdGrVSsGrV9pITDRARkbl+fLWM5DApuabBMayRg4cXDOQnqqFpHg97FxfA9OW3Mfta2LcumyKRq1eolnbF/AfWO89d6349HXyUKNqqmzfxjwdTtWfIy1TF89SjLH9jAf8Ol7HkyQTxCUbY3iXa3ieaoCzt2vJrrE0TYfYMAeWZhnQFAlwql7QOftJkgle5ZbvEYgf633vp/RUbXw15jHOH6mCl891YG37CkOmRCM+Rh8h5yp389TgaU9x9aQYSU91oG8kRdseyajnlYEZ/dXnM5ner9IlOGZmZrCwsMCvv/4Ka2trxMTEYNq0aSW6x7hx4+Dt7Y0lS5age/fuOHHiBA4ePCir6QGA2bNno1u3bqhZsyZ69+4NDQ0N3Lx5E3fu3MEPP/yg7KdVJqpUeQV//wsQi3ORmqqLu3erYuJEH6SmFtRIdekSITcR4KJFJwAAixc3xbFjDiqJWRWc6qbjP5tuy/ZHTH8EADi6uxqWTnfBxWNVsCrAEX1GxGLkjEd4EqWP+ePdce+6ek9A5lozCavG7pXtj/+8YGTigSvOmL+lLTYf94S+Th6m9j0DI/1c3Hpkhe9+6SKbAwcAhnW5hi5NH8j2g6fsBACMXdUdNyLe9BFTJ+97P60OcIS9SyZ8eiTC0DgfyUk6uH7eDH8st0N+XsVsVlAW0yr5mLLsMcyr5SErXRNRYfqY0d8R18+KVR1amZIqYS0qKSpmh3WRIAgVM3IFHDt2DOPHj8ejR4/g4uKCFStWoE2bNti9ezfq168Pe3t73LhxA/Xr1wcApKSkwMzMDCdPnkSbNm0AFAwTDwwMRHJyMnx9fdG4cWOsWrUK8fHxssc5fPgw5s6dixs3bkBbWxuurq4YNmwYhg8fDqCg6Wv37t3o0aNHieJPS0uDiYkJ2nlMhZZm5akZ+RiiqKeqDqFCSO3EZqDiMDl0T9UhVAiSDE4f8T75Qh5OSXchNTUVYnHpJFyvvye2hrrDwFhToXtlpUvwZf17pRpvaah0NTgA4OPjg3v35D+o/p3nvZ3zmZqaFjo2fPhwWaLyev/t5ihfX1/4+vq+M45KmFsSEVEZej2XjWL3qJjfVZUywVGGRYsWoUOHDjA0NMTBgwexceNG/Pzzz6oOi4iIiMAE56NduXIFCxYsQHp6OhwcHLBixQoMGzZM1WERERHJSAQRJIKCi20qeL2qMMH5SNu3b1d1CERERO8lUUInY0kFbaKq3N3siYiISC2xBoeIiEhNSQUNSBWciVhaQQfEMMEhIiJSU2yiIiIiIlIjrMEhIiJSU1IoPgqqoq7exQSHiIhITSlnor+K2dhTMaMmIiIieg/W4BAREakpiaABiYKjqBS9XlWY4BAREakpKUSQQtE+OJzJmIiIiMqRylyDUzGjJiIiInoP1uAQERGpKeVM9Fcx60KY4BAREakpqSCCVNF5cCroauIVMy0jIiIieg/W4BAREakpqRKaqCrqRH9McIiIiNSUclYTr5gJTsWMmoiIiOg9WINDRESkpiQQQaLgRH2KXq8qTHCIiIjUFJuoiIiIiNQIa3CIiIjUlASKNzFJlBNKmWOCQ0REpKbYREVERERq5/Vim4puxbVmzRrUq1cPYrEYYrEYXl5eOHjwoOx8dnY2xowZAwsLCxgZGaFXr1549uyZ3D1iYmLQtWtXGBgYoFq1apgyZQry8/NL/NyZ4BAREZFS1KhRAz/99BNCQkJw7do1tGvXDp999hnu3r0LAJg4cSL27t2Lv/76C6dPn0ZcXBx69uwpu14ikaBr167Izc3FhQsXsHHjRgQHB2P27NkljoVNVERERGpKgAhSBfvgCP+/Pi0tTe64rq4udHV15Y51795dbn/+/PlYs2YNLl26hBo1amD9+vXYsmUL2rVrBwDYsGED3NzccOnSJTRv3hxHjhzBvXv3cOzYMVhaWqJ+/fqYN28e/P39ERAQAB0dnWLHzRocIiIiNaXMJipbW1uYmJjItqCgoPc/tkSCrVu3IjMzE15eXggJCUFeXh58fHxkZVxdXVGzZk1cvHgRAHDx4kV4eHjA0tJSVsbX1xdpaWmyWqDiYg0OERERfVBsbCzEYrFs/+3am9du374NLy8vZGdnw8jICLt374a7uztCQ0Oho6MDU1NTufKWlpZISEgAACQkJMglN6/Pvz5XEkxwKjDhfhQEkbaqwyjXpI1cVR1ChZCvWzFnKi1z7/hAp7e81ZRBbxHKbuC1VBBBKij29/36+tcdhz/ExcUFoaGhSE1NxY4dO+Dn54fTp08rFMPHYIJDRESkpiRKWE28pNfr6OjA0dERANCoUSNcvXoVy5cvR9++fZGbm4uUlBS5Wpxnz57BysoKAGBlZYUrV67I3e/1KKvXZYqLfXCIiIio1EilUuTk5KBRo0bQ1tbG8ePHZefCw8MRExMDLy8vAICXlxdu376NxMREWZmjR49CLBbD3d29RI/LGhwiIiI1pcwmquKYPn06OnfujJo1ayI9PR1btmzBqVOncPjwYZiYmGDo0KGYNGkSzM3NIRaLMW7cOHh5eaF58+YAgI4dO8Ld3R0DBgzAggULkJCQgJkzZ2LMmDHv7PPzLkxwiIiI1JQUGpAq2FhTkusTExMxcOBAxMfHw8TEBPXq1cPhw4fRoUMHAMDSpUuhoaGBXr16IScnB76+vvj5559l12tqamLfvn0YNWoUvLy8YGhoCD8/P8ydO7fEcTPBISIiIqVYv379e8/r6elh9erVWL169TvL2NnZ4cCBAwrHwgSHiIhITUkEESQKNlEper2qMMEhIiJSU2XdB6c8YYJDRESkpgQlrCYucDVxIiIiovKBNThERERqSgIRJAoutqno9arCBIeIiEhNSQXF+9BIBSUFU8bYREVERERqhzU4REREakqqhE7Gil6vKkxwiIiI1JQUIkgV7EOj6PWqUjHTMiIiIqL3YA0OERGRmuJMxkRERKR2KnMfnIoZNREREdF7sAaHiIhITUmhhLWoKmgnYyY4REREakpQwigqgQkOERERlSeVeTVx9sEhIiIitcMaHCIiIjVVmUdRMcEhIiJSU2yiIiIiIlIjrMEhIiJSU5V5LSomOERERGqKTVREREREaoQ1OERERGqqMtfgMMEhIiJSU5U5wWETFREREamdSlWDk5WVhQEDBuDo0aNIT0/Hy5cvYWpqqtA9RSIRdu/ejR49eiglxorGwjIXQ6fHonGbVOjqSxEXrYclk+3x8LahqkNTKQvzLAwdcB1NGj6Fro4EcQnGWLzqEzyMtICmphSDvgpFk4ZPYW2ZjswsHdy4ZY31fzRA8ksDVYdeaurbx+Fr75twrZGEquIsTNnoizP37GXn29R5hJ7N78G1ehJMDHPw9bLeeBhfRXZerJ+N4R2uoZlzLCxNM5CSqY/Td2vhlyNNkJmtq4qnVCb6DInCJ+2TUMM+E7k5GggLNcXvyxzx9PGbv7Gxs8LQoFkyzKvmIDtLE/dummDDMic8ia7cf4cA0H3Qc/QelQjzqvl4dE8fP8+sjvBQ9f07e1tlrsGpVAnOxo0bcfbsWVy4cAFVqlSBiYmJwveMj4+HmZmZEqKreIzE+ViyMww3L4ox088ZqcnaqF4rGxmpmqoOTaWMDHOw5MdDuHXHCjPntUdKmi6qW6cjI0MHAKCrmw9HhxfY8pcHHkWbwcgoF6OGXEXg9JMYN7WriqMvPfo6+XgYb4G911yxYODhIs/fjLbCsVu1MaP36ULnq4izUFWciRX7vRD1zAxWZhmY9vkZVBVnYfp/O5bFU1CJuo1TsG9bDTy4K4ampgC/cRGYv/YGvunphZxXBX9rEfeMcWq/FRIT9GAszkP/UY/ww9rrGNKlJaTSivnlpAytP32JEXPisHJaDdy/boDPhydh/pZHGNrKBakvtFUdXpkQoPgwb0E5oZS5SpXgREZGws3NDXXr1lXaPa2srJR2r4rmi1HxSIrXwZIpb36FP4tV31/SxdXn87t4/twQi1d9Ijv2LNFY9u+sLB1MD+wgd83q35pi5YKDqFolE0nP1fNX98XwmrgYXvOd5w/ecAYAWJulFXn+0TNzTPuvr2z/abIJ1hxuisAvj0NTQwqJVD1b3GePbiC3v2R2HWw9dQZObmm4c73gx9WhnTVk5xPj9LFpVW38vOMyqtm8QsKTylNb8baeI57j0BZzHNlmDgBY4V8DTdunwbdfMravslRxdGWjMtfglKtPBKlUigULFsDR0RG6urqoWbMm5s+fDwC4ffs22rVrB319fVhYWGDEiBHIyMiQXTto0CD06NEDixYtgrW1NSwsLDBmzBjk5eUBANq0aYPFixfjzJkzEIlEaNOmDYCCJqY9e/bIxWFqaorg4GAAQG5uLsaOHQtra2vo6enBzs4OQUFBsrJvX69onBVJ8w4peHDLEDN+jsDWkBtYdeAuOn2ZpOqwVK55kyd4EGmOGZNPY9uG7Vi9aB86+zx87zWGBnmQSoHMzMrxq1JZjPRykZmto7bJTVEMjfIBAOlpRb9XdPUl6PBZHOKf6ON5gl5ZhlauaGlL4VQvC9fPvvlxIQgi3DhrDPdGWSqMjMpKuarBmT59OtatW4elS5eiZcuWiI+Px/3795GZmQlfX194eXnh6tWrSExMxLBhwzB27FhZIgIAJ0+ehLW1NU6ePImIiAj07dsX9evXx/Dhw7Fr1y5MmzYNd+7cwa5du6Cjo1OsmFasWIF//vkH27dvR82aNREbG4vY2NgiyyojzqLk5OQgJydHtp+WVvQv3LJmbZuDbl8nYtdvVti62hrO9TIxKvAx8vNEOLazyodvoKasLdPRzTcdu/a6Y+tODzg7PseooVeRl6+BY6dqFyqvrS3B0AHXcepcLWS9Kt77kgATg1cY0j4Ee664qTqUMiMSCfhm6gPcvWGCxxFGcue69onFkIkR0DeQIDbKADO+aYD8/MqT+L1NbC6BphaQkiT/NffyuRZsHXPecZX6qcw1OOUmwUlPT8fy5cuxatUq+Pn5AQBq166Nli1bYt26dcjOzsamTZtgaFhQfb9q1Sp0794d//nPf2BpWVDVaGZmhlWrVkFTUxOurq7o2rUrjh8/juHDh8Pc3BwGBgbQ0dEpUbNSTEwMnJyc0LJlS4hEItjZ2b2z7JYtWxSOsyhBQUEIDAwsdsxlRaQBPLxtgOCFBdXjkXcNUcvlFbp+nVipExyRCHgYaYENmwuaFiKjzFGrZgq6+j4olOBoakoxY/IZQASs/KWZKsKtkAx1c7Fk8EFEJZph3dHGqg6nzIz+/j7samdg8qDCz/nkAWvcuGQB8yo56On3GNMX3sZkv8bIy63cfeIqu8qc4JSb9D4sLAw5OTlo3759kec8PT1lSQMAtGjRAlKpFOHh4bJjderUgabmmz9ma2trJCYmKhTXoEGDEBoaChcXF4wfPx5Hjhx573MojTinT5+O1NRU2fauGqSylpyojZiH+nLHYiL0UdUmV0URlQ/JKfp4/ES+A3vsExNUq5Ipd+x1cmNZNRPTA3xYe1NMBjq5WDZ0P7JytOG/yRcSaeX4Ah81/T6aej/HtOGN8CKxcNNTVoYW4mIMcOe6GX78rh5s7TPxSbvK22SclqwJST5gWjVf7rhZlXy8TCo3v+2pFJWbBEdfX//DhT5AW1u+TVokEkEqlb73GpFIBEGQ7yP+7/4wDRs2RFRUFObNm4dXr16hT58+6N27d5nGqaurC7FYLLeVB/dCjFDDIVvuWHX7bCQ+rdxf1PfCqsLWRr4ZsbpNGhKT3jQpvE5uqlunYVqAD9Iz2Dm7OAx1c7Fi2H7k5Wtg8sZOyM2vDF9UAkZNvw+vdkmYPrwRnj0txmfl/39wa+u8//NPneXnaeDhLQM0aJkuOyYSCajfMgP3QipPx+vXNTiKbhVRuUlwnJycoK+vj+PHjxc65+bmhps3byIz880v4PPnz0NDQwMuLi4KPW7VqlURHx8v23/48CGysuQ7oInFYvTt2xfr1q3Dtm3bsHPnTiQnJ5dpnOXR7t8s4dogE33HxMHaLhttPnuBLl8lYe+myjE64V127XODq3MSvux1GzZWaWjbKgpdOjzEP4cKRglpakoxa8ppONd+gf8sawkNDQFmpq9gZvoKWloSFUdfevR18uBk/RxO1s8BADbmaXCyfg5L04IvILF+Npysn8O+2ksAgF3VFDhZP4e5UcHfY0Fysw96OnmYv6MNDHXzYG6UBXOjLGiI1PeLfPT34WjbJQELptXFq0xNmFnkwMwiBzq6Be8Vq+pZ6DMkCo5uaahqlQ03zxR8v+gWcnM0cfVc5W0qBoBdv1ZB56+S4fNFMmwdszHupyfQM5DiyFZzVYdWZgRBpJStIio3P3/09PTg7++PqVOnQkdHBy1atEBSUhLu3r2L/v37Y86cOfDz80NAQACSkpIwbtw4DBgwQNav5WO1a9cOq1atgpeXFyQSCfz9/eVqWJYsWQJra2s0aNAAGhoa+Ouvv2BlZVXkBIGlGWd59OCWEeaOcMRg/yfoPz4OCU90sTawJk7usVB1aCr1IKIK5v6nDQZ/fQP9v7iFhEQjrP29CU6ecQAAVDHPglfTJwCANUv2y107ZVYH3LqrnlMPuNVIxJpv9sr2J3a/CADYd80Z8/5qh1bu0Zjd55Ts/Pz+xwAA6442wm/HmsClehLq1ixoyt3l/6fcvXv89BXiX5aPmk1l69a34L2y4PcQueNLZrnj2D82yM3VRJ2GKfjs61gYifOQ8kIHd0LM8N3AxkhNrty1qaf/MYOJhQQDpyTArGo+Ht3Vx4z+9kh5ztGKlUG5SXAAYNasWdDS0sLs2bMRFxcHa2trjBw5EgYGBjh8+DC+/fZbNGnSBAYGBujVqxeWLFmi8GMuXrwYgwcPRqtWrWBjY4Ply5cjJOTNB4mxsTEWLFiAhw8fQlNTE02aNMGBAwegoVG48qs04yyvrpwwxZUTpqoOo9y5HFIDl0NqFHnuWZIRfHsOKOOIVO/6o+po5j/ynef3h7hif4jrR1+vrrp4+rz3fHKSLuaMbfDeMpXZPxuq4J8NlbcmSwqRwhP9KXq9qoiEtzugULmXlpYGExMTtNX+Aloi/hJ5H6HRu78w6Y0UJ/WcXFDZLA5FqDqECkGSVHk7NxdHvpCHU/gbqamppdan8vX3RLM946FlqFgfv/zMHFzusaJU4y0N5aYPDhEREVVsQUFBaNKkCYyNjVGtWjX06NFDbhQxUDDxrkgkkttGjpSvnY2JiUHXrl1hYGCAatWqYcqUKcjPlx8R9yHlqomKiIiIlEcZnYRLcv3p06cxZswYNGnSBPn5+fj+++/RsWNH3Lt3T24KleHDh2Pu3LmyfQODNyPbJBIJunbtCisrK1y4cAHx8fEYOHAgtLW18eOPPxY7FiY4REREaqqsJ/o7dOiQ3H5wcDCqVauGkJAQeHt7y44bGBi8c9LdI0eO4N69ezh27BgsLS1Rv359zJs3D/7+/ggICCj2SgRsoiIiIlJTyhwmnpaWJrf9ewmhd0lNTQUAmJvLD83fvHkzqlSpgrp162L69Oly07NcvHgRHh4ecqOPfX19kZaWhrt37xb7ubMGh4iIiD7I1tZWbn/OnDkICAh4Z3mpVIoJEyagRYsWqFu3ruz4V199BTs7O9jY2ODWrVvw9/dHeHg4du3aBQBISEgoNLXK6/2EhIRix8sEh4iISE0JSmiiel2DExsbKzeKSlf3/aOzxowZgzt37uDcuXNyx0eMGCH7t4eHB6ytrdG+fXtERkaidu3CCxJ/LDZRERERqSkBgCAouP3/Xm8vGfS+BGfs2LHYt28fTp48iRo1ip4T7LVmzQoWGo6IKJiGwcrKCs+ePZMr83q/JItlM8EhIiIipRAEAWPHjsXu3btx4sQJ2Nvbf/Ca0NBQAAULTwOAl5cXbt++LbcI9dGjRyEWi+Hu7l7sWNhERUREpKakEEFUhjMZjxkzBlu2bMHff/8NY2NjWZ8ZExMT6OvrIzIyElu2bEGXLl1gYWGBW7duYeLEifD29ka9evUAAB07doS7uzsGDBiABQsWICEhATNnzsSYMWM+2Cz2b0xwiIiI1FRZz4OzZs0aAAWT+f3bhg0bMGjQIOjo6ODYsWNYtmwZMjMzYWtri169emHmzJmyspqamti3bx9GjRoFLy8vGBoaws/PT27enOJggkNERERK8aHVn2xtbXH69OkP3sfOzg4HDhxQKBYmOERERGpKKoggKsOJ/soTJjhERERq6vVIKEXvURFxFBURERGpHdbgEBERqamy7mRcnjDBISIiUlNMcIiIiEjtVOZOxuyDQ0RERGqHNThERERqqjKPomKCQ0REpKYKEhxF++AoKZgyxiYqIiIiUjuswSEiIlJTHEVFREREakf4/6boPSoiNlERERGR2mENDhERkZpiExURERGpn0rcRsUEh4iISF0poQYHFbQGh31wiIiISO2wBoeIiEhNcSZjIiIiUjvsZEwVkiCRQBCxlfF9tJ68UHUIFcLlXZtUHUKF0OVQB1WHQETFxASHiIhIXQkixTsJswaHiIiIypPK3AeH7RtERESkdliDQ0REpK440R8RERGpG46i+oB//vmn2Df89NNPPzoYIiIiImUoVoLTo0ePYt1MJBJBIpEoEg8REREpUwVtYlJUsRIcqVRa2nEQERGRklXmJiqFRlFlZ2crKw4iIiJSNkFJWwVU4gRHIpFg3rx5qF69OoyMjPDo0SMAwKxZs7B+/XqlB0hERERUUiVOcObPn4/g4GAsWLAAOjo6suN169bFb7/9ptTgiIiISBEiJW0VT4kTnE2bNuHXX39F//79oampKTvu6emJ+/fvKzU4IiIiUgCbqIrv6dOncHR0LHRcKpUiLy9PKUERERERKaLECY67uzvOnj1b6PiOHTvQoEEDpQRFRERESlCJa3BKPJPx7Nmz4efnh6dPn0IqlWLXrl0IDw/Hpk2bsG/fvtKIkYiIiD5GJV5NvMQ1OJ999hn27t2LY8eOwdDQELNnz0ZYWBj27t2LDh06lEaMRERERCXyUWtRtWrVCkePHlV2LERERKREglCwKXqPiuijF9u8du0awsLCABT0y2nUqJHSgiIiIiIl4GrixffkyRP069cP58+fh6mpKQAgJSUFn3zyCbZu3YoaNWooO0YiIiKiEilxH5xhw4YhLy8PYWFhSE5ORnJyMsLCwiCVSjFs2LDSiJGIiIg+xutOxopuxRQUFIQmTZrA2NgY1apVQ48ePRAeHi5XJjs7G2PGjIGFhQWMjIzQq1cvPHv2TK5MTEwMunbtCgMDA1SrVg1TpkxBfn5+iZ56iROc06dPY82aNXBxcZEdc3FxwcqVK3HmzJmS3o6IiIhKiUhQzlZcp0+fxpgxY3Dp0iUcPXoUeXl56NixIzIzM2VlJk6ciL179+Kvv/7C6dOnERcXh549e8rOSyQSdO3aFbm5ubhw4QI2btyI4OBgzJ49u0TPvcRNVLa2tkVO6CeRSGBjY1PS2xEREVFpKeM+OIcOHZLbDw4ORrVq1RASEgJvb2+kpqZi/fr12LJlC9q1awcA2LBhA9zc3HDp0iU0b94cR44cwb1793Ds2DFYWlqifv36mDdvHvz9/REQECC3TNT7lLgGZ+HChRg3bhyuXbsmO3bt2jV8++23WLRoUUlvR0RERBVAWlqa3JaTk/PBa1JTUwEA5ubmAICQkBDk5eXBx8dHVsbV1RU1a9bExYsXAQAXL16Eh4cHLC0tZWV8fX2RlpaGu3fvFjveYtXgmJmZQSR60waXmZmJZs2aQUur4PL8/HxoaWlhyJAh6NGjR7EfnIiIiEqREif6s7W1lTs8Z84cBAQEvPMyqVSKCRMmoEWLFqhbty4AICEhATo6OrJBSq9ZWloiISFBVubfyc3r86/PFVexEpxly5YV+4ZERERUTiixiSo2NhZisVh2WFdX972XjRkzBnfu3MG5c+cUDODjFCvB8fPzK+04iIiIqBwTi8VyCc77jB07Fvv27cOZM2fkpo+xsrJCbm4uUlJS5Gpxnj17BisrK1mZK1euyN3v9Sir12WKo8R9cP4tOzu7UJscERERlRNlvNimIAgYO3Ysdu/ejRMnTsDe3l7ufKNGjaCtrY3jx4/LjoWHhyMmJgZeXl4AAC8vL9y+fRuJiYmyMkePHoVYLIa7u3uxYynxKKrMzEz4+/tj+/btePHiRaHzEomkpLckIiKi0lDGo6jGjBmDLVu24O+//4axsbGsz4yJiQn09fVhYmKCoUOHYtKkSTA3N4dYLMa4cePg5eWF5s2bAwA6duwId3d3DBgwAAsWLEBCQgJmzpyJMWPGfLBZ7N9KXIMzdepUnDhxAmvWrIGuri5+++03BAYGwsbGBps2bSrp7YiIiEhNrFmzBqmpqWjTpg2sra1l27Zt22Rlli5dim7duqFXr17w9vaGlZUVdu3aJTuvqamJffv2QVNTE15eXvj6668xcOBAzJ07t0SxlLgGZ+/evdi0aRPatGmDwYMHo1WrVnB0dISdnR02b96M/v37l/SWREREVBqUOIqqWEWLsTKnnp4eVq9ejdWrV7+zjJ2dHQ4cOFDsxy1KiWtwkpOT4eDgAKCgw1FycjIAoGXLlpzJmIiIqBwp65mMy5MS1+A4ODggKioKNWvWhKurK7Zv346mTZti7969hca1qxNBEPDNN99gx44dePnyJW7cuIH69eurOqxypc+YBAydHofdv1XF2gDbD1+gprr0eowuPWNgaf0KAPA4ygh//uaIkIvVAADaOhIM+zYM3h3joa0txfVLVfDzgrpISS5+23JFtHejBfZvqoJnsQWzkNq5ZKP/xAQ0aZeOhFgd+DUruvPgjF+i4N09FWnJmvhprB2iwvSR/lITJhb58PJNxeDp8TA0lpblUylTfYZE4ZP2Sahhn4ncHA2EhZri92WOePrYUFZm7KwwNGiWDPOqOcjO0sS9mybYsMwJT6IN33Nn9Ve3WQa+GJ0EJ48sWFjlI2BILVw8ZKLqsKiMlDjBGTx4MG7evInWrVtj2rRp6N69O1atWoW8vDwsWbKkNGIsFw4dOoTg4GCcOnUKDg4OqFKliqpDKlecPTPRtf9zPLqnr+pQVO75Mz0Er3ZBXKwhIBLg0/UpZi0KwfgBLRHzyBjDJ4ahSYtEBE1vgKwMbYycchcz/nMdU4Z7qTr0UlXVOg9Dvo9DdfscCIIIR/8yQ8Bge6w+8gC2jtn4M/SOXPkD/7XAjjXV0KRdOgBApAF4+aZikH88TCzyEReli1Xf10B6iham//xYFU+pTNRtnIJ922rgwV0xNDUF+I2LwPy1N/BNTy/kvNIEAETcM8ap/VZITNCDsTgP/Uc9wg9rr2NIl5aQShVsnqjA9AykeHRXD4f/NMec36NVHY5qlHEn4/KkxAnOxIkTZf/28fHB/fv3ERISAkdHR9SrV0+pwZUnkZGRsLa2xieffFJqj5Gbm1vsNTbKEz0DCfxXRmPZ1Jro923xZ5lUV1fOyc/AuWmNC7r0jIFr3RQ8f6aHjp/GYuGs+rh1rSBJXja3Hn756wxc6r5E+B0zVYRcJpp3lJ9GYvC0BOzbVAX3QwxQyyUb5tXkVwq+cNAE3t1ToG9YUDtjbCpBd783Izcta+Shu99z/LWmWukHr0KzRzeQ218yuw62njoDJ7c03Lle8H45tPPNPCOJcfrYtKo2ft5xGdVsXiHhiUGZxlueXDspxrWTxZu3hdSPQvPgAAUdgXr27KnWyc2gQYMwbtw4xMTEQCQSoVatWpBKpQgKCoK9vT309fXh6emJHTt2yK6RSCQYOnSo7LyLiwuWL19e6L49evTA/PnzYWNjI7dCe0Uydn4srhw3wY1z/CB5m4aGAO8OcdDTlyDstikc3VKhrS0g9MqbGsAnj42QGK8HN48U1QVaxiQS4NQeU+RkacCtcWah8w9v6SPyrgF8+xWeiuK1FwlaOH/QFPW8Mkoz1HLH0KggEUxP0y7yvK6+BB0+i0P8E308T9Ary9CoHBJBCX1wVP0kPlKxanBWrFhR7BuOHz/+o4Mpr5YvX47atWvj119/xdWrV6GpqYmgoCD897//xdq1a+Hk5IQzZ87g66+/RtWqVdG6dWtIpVLUqFEDf/31FywsLHDhwgWMGDEC1tbW6NOnj+zex48fh1gsxtGjR9/5+Dk5OXKLmpWnCRVbf5oMR48sjOvqqupQyhW72mlYvP4idHSkePVKEz9MbYjYKGM4OKchL1cDmRnyX04vk3VhZvHhhesquqgwPUzo7oTcHA3oG0oxe30U7JwLP+9Df1qgplM26jTJKnQuaJQdLh42QU62Bpp3SMXERbFlEXq5IBIJ+GbqA9y9YYLHEUZy57r2icWQiRHQN5AgNsoAM75pgPx8hX/DElVYxUpwli5dWqybiUQitUxwTExMYGxsDE1NTVhZWSEnJwc//vgjjh07Jpt50cHBAefOncMvv/yC1q1bQ1tbG4GBgbJ72Nvb4+LFi9i+fbtcgmNoaIjffvvtvU1TQUFBcvcqL6pa52JU4BNM/8oReTn8IP23p4+NMO7rljA0ykeLdvGYNOcW/Ec2U3VYKlejdg5+PhqOrHRNnN1nikXf2mHhrodySU7OKxFO7jbDVxOKbu78JvAp+k9KwNNHuvg9yBq/BFbHuKAnZfUUVGr09/dhVzsDkwc1LnTu5AFr3LhkAfMqOejp9xjTF97GZL/GyMvVVEGkVG6U8TDx8qRYCU5UVFRpx1GhREREICsrCx06dJA7npubiwYN3rSXr169Gr///jtiYmLw6tUr5ObmFhp55eHh8cF+N9OnT8ekSZNk+2lpaYVWdVUFx3pZMKuaj9UH78uOaWoBHs0y8OmgJHRzaFBpOzjm52sg/knBCJaI+yZwdk/FZ32jceaYNbR1pDA0ypOrxTEzz8HLF+o9igoAtHUEVLfPBQA41XuF8FAD7PmtKr5d8CZBObvfFDmvRPD5IrnIe5hXy4d5tXzUdMqBsakE333uhK8mJMDCMr/I8upi1PT7aOr9HFOHNMaLxMJNT1kZWsjK0EJcjAHu3zLB9nOn8Em7JJw+VPy1e0gNsZMxlURGRkGb//79+1G9enW5c6+nkd66dSsmT56MxYsXw8vLC8bGxli4cCEuX74sV97Q8MPDOHV1dUs0PXVZCT1njBHt3eSOfbf4MWIj9bD9Z8tKm9wURaQBaOtIERFmgrw8ETybPMeFk9YAgOo1M1DNOhtht01VG6QKCAKQlytf+3f4Tws075gGU4sPL/vyek6xt++hXgSMmh4Or3ZJmDa0EZ49LcZIxf//6WnrqO/weaIPYYLzEdzd3aGrq4uYmBi0bt26yDLnz5/HJ598gtGjR8uORUZGllWIZeJVpiYeh8t/2Ga/0kD6y8LHKxO/0fdx7WI1JCXoQd8gH2184+DR8AVmjW+CrExtHPnHFsMnhCEjTQdZmVoYOfkuwm6ZqvUIKgD4/UdrNGmXhqrV8/AqQwMnd5vh1gUjzN/y5u/iaZQObl8yxLz/Pip0/ZXjxniZpA2X+lnQM5TicbgefptngzpNMmBlm1uWT6VMjf4+HG06J2DuBE+8ytSU9dXKzNBCbo4mrKpnwdv3Ga5ftEDqSx1UsczGF0OikZujiavnKvd0FnoGEtjYv3lvWNnmwqHOK6SnaCLpacUbsfpRWINDJWFsbIzJkydj4sSJkEqlaNmyJVJTU3H+/HmIxWL4+fnByckJmzZtwuHDh2Fvb48//vgDV69eLbSyKqkfU/NcfDfnJsyr5CAzQwvREcaYNb4JQq9UBQCsW+oGQQp8/9N1aOu8mehP3aU818LC8XZITtSCgbEE9m7ZmL8lEo1avxkFdXirBapY56FR6/RC1+voCTi42QK/BFRHXq4IVW1y0aJzKvqOTSxUVp1061vQfLfg9xC540tmuePYPzbIzdVEnYYp+OzrWBiJ85DyQgd3Qszw3cDGSE2uJF/i7+Ds+QoLd75JoEcGxgEAjmwzw+KJNVUVVplSxkzEFXUmY5FQnIUjCMuWLcOyZcsQHR0NoGBm4xUrVmDNmjV49OgRTE1N0bBhQ3z//ffw9vZGTk4ORo4cid27d0MkEqFfv34wMTHBwYMHERoaCqBgmHhKSgr27NlToljS0tJgYmKCNho9oSUqeqgoFdCyYf+D4th/Zb+qQ6gQunh2+HAhgiQpSdUhlGv5Qh5O4W+kpqZCLC6d6TVef0/Umj8fGnqKTRcgzc5G9IwZpRpvaWCCUwExwSk+JjjFwwSneJjgFA8TnPcr0wTnByUlODMrXoLzUT3zzp49i6+//hpeXl54+vQpAOCPP/7AuXPnlBocERERKUBQ0lYBlTjB2blzJ3x9faGvr48bN27IJqBLTU3Fjz/+qPQAiYiIiEqqxAnODz/8gLVr12LdunXQ1n7TPNKiRQtcv35dqcERERHRx1N4mQYldFJWlRKPogoPD4e3t3eh4yYmJkhJSVFGTERERKQMlXgm4xLX4FhZWSEiIqLQ8XPnzsHBwUEpQREREZESsA9O8Q0fPhzffvstLl++DJFIhLi4OGzevBmTJ0/GqFGjSiNGIiIiohIpcRPVtGnTIJVK0b59e2RlZcHb2xu6urqYPHkyxo0bVxoxEhER0UeozBP9lTjBEYlEmDFjBqZMmYKIiAhkZGTA3d0dRkZGpREfERERfSwu1VByOjo6cHd3V2YsREREREpR4gSnbdu2EIne3aP6xIkTCgVERERESqKMYd6VpQanfv36cvt5eXkIDQ3FnTt34Ofnp6y4iIiISFFsoiq+pUuXFnk8ICAAGRkZRZ4jIiIiKksftRZVUb7++mv8/vvvyrodERERKaoSz4Pz0Z2M33bx4kXoKbhiKRERESkPh4mXQM+ePeX2BUFAfHw8rl27hlmzZiktMCIiIqKPVeIEx8TERG5fQ0MDLi4umDt3Ljp27Ki0wIiIiIg+VokSHIlEgsGDB8PDwwNmZmalFRMREREpQyUeRVWiTsaampro2LEjVw0nIiKqAF73wVF0q4hKPIqqbt26ePToUWnEQkRERKQUJU5wfvjhB0yePBn79u1DfHw80tLS5DYiIiIqRyrhEHGgBH1w5s6di++++w5dunQBAHz66adySzYIggCRSASJRKL8KImIiKjkKnEfnGInOIGBgRg5ciROnjxZmvEQERERKazYCY4gFKRwrVu3LrVgiIiISHk40V8xvW8VcSIiIipn2ERVPM7Ozh9McpKTkxUKiIiIiEhRJUpwAgMDC81kTEREROWTKpqozpw5g4ULFyIkJATx8fHYvXs3evToITs/aNAgbNy4Ue4aX19fHDp0SLafnJyMcePGYe/evdDQ0ECvXr2wfPlyGBkZFTuOEiU4X375JapVq1aSS4iIiEhVVNBElZmZCU9PTwwZMqTQ+pWvderUCRs2bJDt6+rqyp3v378/4uPjcfToUeTl5WHw4MEYMWIEtmzZUuw4ip3gsP8NERERfUjnzp3RuXPn95bR1dWFlZVVkefCwsJw6NAhXL16FY0bNwYArFy5El26dMGiRYtgY2NTrDiKPdHf61FUREREVEEoOsnfv2qA3p7YNycn56PDOnXqFKpVqwYXFxeMGjUKL168kJ27ePEiTE1NZckNAPj4+EBDQwOXL18u9mMUO8GRSqVsniIiIqpAlLkWla2tLUxMTGRbUFDQR8XUqVMnbNq0CcePH8d//vMfnD59Gp07d5ZNFJyQkFAo39DS0oK5uTkSEhKK/Tgl6oND5YxUAohKvNpGpSLk5Ko6hAqh7ZDhqg6hQhB5SlUdQoWgfSxJ1SHQa0rsgxMbGwuxWCw7/Ha/meL68ssvZf/28PBAvXr1ULt2bZw6dQrt27dXKNR/47cjERERfZBYLJbbPjbBeZuDgwOqVKmCiIgIAICVlRUSExPlyuTn5yM5Ofmd/XaKwgSHiIhIXSmxD05pefLkCV68eAFra2sAgJeXF1JSUhASEiIrc+LECUilUjRr1qzY92UTFRERkZpSxTw4GRkZstoYAIiKikJoaCjMzc1hbm6OwMBA9OrVC1ZWVoiMjMTUqVPh6OgIX19fAICbmxs6deqE4cOHY+3atcjLy8PYsWPx5ZdfFnsEFcAaHCIiIlKia9euoUGDBmjQoAEAYNKkSWjQoAFmz54NTU1N3Lp1C59++imcnZ0xdOhQNGrUCGfPnpVr8tq8eTNcXV3Rvn17dOnSBS1btsSvv/5aojhYg0NERKSuVDDRX5s2bd47tczhw4c/eA9zc/MSTepXFCY4REREaqoyrybOJioiIiJSO6zBISIiUlcqaKIqL5jgEBERqatKnOCwiYqIiIjUDmtwiIiI1JTo/5ui96iImOAQERGpq0rcRMUEh4iISE1xmDgRERGRGmENDhERkbpiExURERGppQqaoCiKTVRERESkdliDQ0REpKYqcydjJjhERETqqhL3wWETFREREakd1uAQERGpKTZRERERkfphExURERGR+mANDhERkZpiExURERGpn0rcRMUEh4iISF1V4gSHfXCIiIhI7bAGh4iISE2xDw4RERGpHzZREREREakP1uAQERGpKZEgQCQoVgWj6PWqotIEp02bNqhfvz6WLVumyjAwaNAgpKSkYM+ePSqNoyKq2ywDX4xOgpNHFiys8hEwpBYuHjJRdVgq1WdIFD5pn4Qa9pnIzdFAWKgpfl/miKePDWVlxs4KQ4NmyTCvmoPsLE3cu2mCDcuc8CTa8D13Vi8aIin8elxHh+YRMDd5hecpBjh83hl/7K0PQARNTSmGfn4NzerFwrpqOjJf6eD6PRv8uqMJXqRUntdpy+LtsKqaUej4nmOuWLHpE5iZZGHkl1fRqE4c9PXz8CTeBP/9xxNnr9Uq+2DLoe6DnqP3qESYV83Ho3v6+HlmdYSHGqg6rLJTiZuoWIMDYPny5RAqaIaqanoGUjy6q4fDf5pjzu/Rqg6nXKjbOAX7ttXAg7tiaGoK8BsXgflrb+Cbnl7IeaUJAIi4Z4xT+62QmKAHY3Ee+o96hB/WXseQLi0hlYpU/AzKRr8ut/BZmzD8tL41op6awaXWc/gPPYPMV9rYdawu9HTy4WT3HH/sbYDIWHMYGeRi3FcXMX/8UYyc20PV4ZeZUQHdoaHx5vPJvsZLLPI/jNNX7AEA00ecgZFBLmYu80Fqui7aez3C7LEnMWrOp4h4bKGqsMuF1p++xIg5cVg5rQbuXzfA58OTMH/LIwxt5YLUF9qqDo9KGRMcACYmlbvGQRHXTopx7aRY1WGUK7NHN5DbXzK7DraeOgMntzTcuW4GADi0s4bsfGKcPjatqo2fd1xGNZtXSHhSOX5d1nF8hvOhdrh0qyYA4NkLY7RvFglX+yQAQOYrHUxZ3EXumuX//QRrZ/+NauYZSEw2KvOYVSE1XV9u/6tut/D0mTFu3rcCANRxSsSy4E9w/1FVAMB//6mPXp3uwrnW80qf4PQc8RyHtpjjyDZzAMAK/xpo2j4Nvv2SsX2VpYqjKxuVeRSVyjsZS6VSTJ06Febm5rCyskJAQIDs3JIlS+Dh4QFDQ0PY2tpi9OjRyMh4U1UbHBwMU1NT7NmzB05OTtDT04Ovry9iY2NlZQICAlC/fn388ssvsLW1hYGBAfr06YPU1FRZmUGDBqFHjx6y/TZt2mD8+PHvjAsAUlJSMGzYMFStWhVisRjt2rXDzZs3Zedv3ryJtm3bwtjYGGKxGI0aNcK1a9cAAI8fP0b37t1hZmYGQ0ND1KlTBwcOHFDSK0rljaFRPgAgPa3oX4y6+hJ0+CwO8U/08TxBryxDU6m7EZZo6BaHGpYFf4u1bV+grlMCrty2fec1hga5kEqBjCydsgqzXNHSlMDnk0gcPOMMoKCm7+7DamjTPArGhjkQiQS0bfYIOtoShIZZqzZYFdPSlsKpXhaunzWWHRMEEW6cNYZ7oywVRlbGBCVtFZDKa3A2btyISZMm4fLly7h48SIGDRqEFi1aoEOHDtDQ0MCKFStgb2+PR48eYfTo0Zg6dSp+/vln2fVZWVmYP38+Nm3aBB0dHYwePRpffvklzp8/LysTERGB7du3Y+/evUhLS8PQoUMxevRobN68+aPiAoAvvvgC+vr6OHjwIExMTPDLL7+gffv2ePDgAczNzdG/f380aNAAa9asgaamJkJDQ6GtXfAFN2bMGOTm5uLMmTMwNDTEvXv3YGT07l+jOTk5yMnJke2npaV99OtNZUskEvDN1Ae4e8MEjyPk/4+79onFkIkR0DeQIDbKADO+aYD8fJX/5igzWw54wkA/Fxvn/wWpVAQNDQHrdzXGsUuORZbX1srHN72v4MTl2sjKrpwJTotGj2FkkIvDZ51kxwJXt8XsMafw95rNyM8XITtXC3OWt0dcYuWuWRWbS6CpBaQkyX/NvXyuBVvHnHdcRepE5QlOvXr1MGfOHACAk5MTVq1ahePHj6NDhw6YMGGCrFytWrXwww8/YOTIkXIJTl5eHlatWoVmzZoBKEhM3NzccOXKFTRt2hQAkJ2djU2bNqF69eoAgJUrV6Jr165YvHgxrKysShzXuXPncOXKFSQmJkJXVxcAsGjRIuzZswc7duzAiBEjEBMTgylTpsDV1VV2j9diYmLQq1cveHh4AAAcHBze+xoFBQUhMDCweC8olSujv78Pu9oZmDyocaFzJw9Y48YlC5hXyUFPv8eYvvA2Jvs1Rl6upgoiLXttmjyCT/NI/PBrW0Q/NYNjzRcY0+8SXqQY4PAFZ7mymppSzBl1AhABS/9ooaKIVa9L64e4cqsGXqS8acYc0us6jAxy8N1PnZCarouWjWIwe8xJfDu/C6KemKswWioP2ESlQvXq1ZPbt7a2RmJiIgDg2LFjaN++PapXrw5jY2MMGDAAL168QFbWm+pFLS0tNGnSRLbv6uoKU1NThIWFyY7VrFlTltwAgJeXF6RSKcLDwz8qrps3byIjIwMWFhYwMjKSbVFRUYiMjAQATJo0CcOGDYOPjw9++ukn2XEAGD9+PH744Qe0aNECc+bMwa1bt977Gk2fPh2pqamy7d9NcFR+jZp+H029n2Pa8EZ4kVi46SkrQwtxMQa4c90MP35XD7b2mfikXZIKIlWNkX2u4M8Dnjh5pTainprj6EUn7DhSF191vSlXriC5OQ6rKhmYsqhzpa29sbTIQMM6cdh/+k3yZ1MtDZ93CMPC31rhxj0bPIq1wKY9DRAebYHPfMLeczf1l5asCUk+YFo1X+64WZV8vExS+W/7slOJm6hUnuC8brZ5TSQSQSqVIjo6Gt26dUO9evWwc+dOhISEYPXq1QCA3NxclcUFABkZGbC2tkZoaKjcFh4ejilTpgAo6Ptz9+5ddO3aFSdOnIC7uzt2794NABg2bBgePXqEAQMG4Pbt22jcuDFWrlz5zlh0dXUhFovlNirPBIyafh9e7ZIwfXgjPHuq/+FL/j9wSltHWrqhlSO6OvmQvvV0pVIRRP/6ufg6ualRLQ3fLeqMtMzK00fpbZ28HyAlTQ+XQt/0UdLVKfjylgryI++kUg1oVNSf3UqSn6eBh7cM0KBluuyYSCSgfssM3AupHB35gTc1OIpuFZHKE5x3CQkJgVQqxeLFi9G8eXM4OzsjLi6uULn8/HxZ510ACA8PR0pKCtzc3GTHYmJi5K69dOkSNDQ04OLi8lGxNWzYEAkJCdDS0oKjo6PcVqVKFVk5Z2dnTJw4EUeOHEHPnj2xYcMG2TlbW1uMHDkSu3btwnfffYd169Z9VCyqpmcggUOdV3Co8woAYGWbC4c6r1C1euknoeXV6O/D0bZLAhZMq4tXmZows8iBmUUOdHQlAACr6lnoMyQKjm5pqGqVDTfPFHy/6BZyczRx9VyVD9xdfVwMrYmvu4Wieb0YWFqko2XDaHzhewfnrtcCUJDcBI4+BpdazzF/XRtoiASYibNgJs6ClqZEtcGXMZFIQKdWD3HknCOk0jcf2zHxpniSIMakQefh6pAEm2pp+KLTbTSq8xTnQ+xUGHH5sOvXKuj8VTJ8vkiGrWM2xv30BHoGUhzZyqa7yqDc1tM5OjoiLy8PK1euRPfu3XH+/HmsXbu2UDltbW2MGzcOK1asgJaWFsaOHYvmzZvL+t8AgJ6eHvz8/LBo0SKkpaVh/Pjx6NOnzzv733yIj48PvLy80KNHDyxYsECWfO3fvx+ff/456tSpgylTpqB3796wt7fHkydPcPXqVfTq1QsAMGHCBHTu3BnOzs54+fIlTp48KZeQVSTOnq+wcOeb5reRgQWJ5JFtZlg8saaqwlKpbn2fAAAW/B4id3zJLHcc+8cGubmaqNMwBZ99HQsjcR5SXujgTogZvhvYGKnJlaf5ZcUWLwz5PATffn0BZuKCif72nnLFpn8KhtlXMc1EiwYxAIDfAnfLXTvhP11wM9ymzGNWlUZ14mBZJfP/o6fekEg0MH1xBwzvcw0/TDwKfb18xD0zxn9+9cblW+8ejVZZnP7HDCYWEgyckgCzqvl4dFcfM/rbI+V5JZoDhxP9lT+enp5YsmQJ/vOf/2D69Onw9vZGUFAQBg4cKFfOwMAA/v7++Oqrr/D06VO0atUK69evlyvj6OiInj17okuXLkhOTka3bt3kOiqXlEgkwoEDBzBjxgwMHjwYSUlJsLKygre3NywtLaGpqYkXL15g4MCBePbsGapUqYKePXvKOgpLJBKMGTMGT548gVgsRqdOnbB06dKPjkeVbl00gq+Np6rDKFe6ePq893xyki7mjG3w3jKVwatsHaz+0wur//Qq8vyzF8ZoO2RYGUdVPl27Ux3tBg4p8tzTZyYIWNm+jCOqOP7ZUAX/bKg8NaNFqahNTIoSCRV4Ct/g4GBMmDABKSkp7ywTEBCAPXv2IDQ0tMziKm1paWkwMTFBG3wGLVEl+iXyETSrVlV1CBXCq0a1VB1ChSDKrzx9pBShfSzkw4UqsXwhD6fwN1JTU0utT+Xr74lGfeZDS1uxvmv5edkI2T6jVOMtDeW2BoeIiIgUJAgFm6L3qIDKbSdjIiIiUowqRlGdOXMG3bt3h42NDUQiUaGFrAVBwOzZs2FtbQ19fX34+Pjg4cOHcmWSk5PRv39/iMVimJqaYujQoXIrGRRHhU5wXq8C/j4BAQFq1TxFRERUnmVmZsLT01M2tcvbFixYgBUrVmDt2rW4fPkyDA0N4evri+zsbFmZ/v374+7duzh69Cj27duHM2fOYMSIESWKg01URERE6koFo6g6d+6Mzp07F30rQcCyZcswc+ZMfPbZZwCATZs2wdLSEnv27MGXX36JsLAwHDp0CFevXkXjxgWzwK9cuRJdunTBokWLYGNTvBGUFboGh4iIiN5NJFXOBhR0XP739u81EosrKioKCQkJ8PF5M9rUxMQEzZo1w8WLFwEAFy9ehKmpqSy5AQqmZ9HQ0MDly5eL/VhMcIiIiOiDbG1tYWJiItuCgoJKfI+EhAQAgKWlpdxxS0tL2bmEhARUq1ZN7ryWlhbMzc1lZYqDTVRERETqSolNVLGxsXLDxF8vNl1esQaHiIhITSlzFNXbayJ+TILzegWBZ8+eyR1/9uyZ7JyVlZVscevX8vPzkZycXKIVCJjgEBERqavX8+AouimJvb09rKyscPz4cdmxtLQ0XL58GV5eBbOae3l5ISUlBSEhbyaMPHHiBKRSKZo1a1bsx2ITFRERESlNRkYGIiIiZPtRUVEIDQ2Fubk5atasiQkTJuCHH36Ak5MT7O3tMWvWLNjY2KBHjx4AADc3N3Tq1AnDhw/H2rVrkZeXh7Fjx+LLL78s9ggqgAkOERGR2vqYifqKukdJXLt2DW3btpXtT5o0CQDg5+eH4OBgTJ06FZmZmRgxYgRSUlLQsmVLHDp0CHp6b5aU2Lx5M8aOHYv27dtDQ0MDvXr1wooVK0oUBxMcIiIidaWCeXDatGmD9y1zKRKJMHfuXMydO/edZczNzbFly5aSPfBb2AeHiIiI1A5rcIiIiNSUKpqoygsmOEREROqKq4kTERERqQ/W4BAREakpNlERERGR+lHBKKrygk1UREREpHZYg0NERKSm2ERFRERE6kcqFGyK3qMCYoJDRESkrtgHh4iIiEh9sAaHiIhITYmghD44Somk7DHBISIiUlecyZiIiIhIfbAGh4iISE1xmDgRERGpH46iIiIiIlIfrMEhIiJSUyJBgEjBTsKKXq8qTHAqMg1NQKSp6ijKN0Gq6ggqBI08vk7FoRsapeoQKgSpFr9a3kckCEB+GT2Y9P+boveogNhERURERGqHaTYREZGaYhMVERERqZ9KPIqKCQ4REZG64kzGREREROqDNThERERqijMZExERkfphExURERGR+mANDhERkZoSSQs2Re9RETHBISIiUldsoiIiIiJSH6zBISIiUlec6I+IiIjUTWVeqoFNVERERKR2WINDRESkripxJ2MmOEREROpKAKDoMO+Kmd8wwSEiIlJX7INDREREpEZYg0NERKSuBCihD45SIilzTHCIiIjUVSXuZMwmKiIiIlKKgIAAiEQiuc3V1VV2Pjs7G2PGjIGFhQWMjIzQq1cvPHv2rFRiYYJDRESkrqRK2kqgTp06iI+Pl23nzp2TnZs4cSL27t2Lv/76C6dPn0ZcXBx69uyp2HN8BzZRERERqSlljqJKS0uTO66rqwtdXd1C5bW0tGBlZVXoeGpqKtavX48tW7agXbt2AIANGzbAzc0Nly5dQvPmzRWK822swSEiIqIPsrW1hYmJiWwLCgoqstzDhw9hY2MDBwcH9O/fHzExMQCAkJAQ5OXlwcfHR1bW1dUVNWvWxMWLF5UeL2twiIiI1JUSOxnHxsZCLBbLDhdVe9OsWTMEBwfDxcUF8fHxCAwMRKtWrXDnzh0kJCRAR0cHpqamctdYWloiISFBsRiLwASHiIhIXSkxwRGLxXIJTlE6d+4s+3e9evXQrFkz2NnZYfv27dDX11csjhJiExURERGVClNTUzg7OyMiIgJWVlbIzc1FSkqKXJlnz54V2WdHUUxwiIiI1NXrGhxFt4+UkZGByMhIWFtbo1GjRtDW1sbx48dl58PDwxETEwMvLy9lPFs5bKIiIiJSV1IAIiXco5gmT56M7t27w87ODnFxcZgzZw40NTXRr18/mJiYYOjQoZg0aRLMzc0hFosxbtw4eHl5KX0EFcAEh4iISG2V9WKbT548Qb9+/fDixQtUrVoVLVu2xKVLl1C1alUAwNKlS6GhoYFevXohJycHvr6++PnnnxWK712Y4BAREZFSbN269b3n9fT0sHr1aqxevbrUY2GC838ikQi7d+9Gjx49VB1KhdFtQBK6DkyCZY1cAMDjB/rYvMwK106aqDgy1eozNBqftE9CDfss5OZoICzUBL8vq42n0YayMp16PUWbLs/g6JYOAyMJvmjRCpnp2iqMuuxtXrIdVlUzCh3/+5grtu33wJalfxV5XeDKtjhzxb60wys3+gx9jE98/v9+ytZA2E0T/L60Np5GG8jKaOtIMHxKJLw7PYO2joDr582xer4zUl7oqDBy1fp6Yhy+nhgvdyw2QhfD29VVUUQqUonXomKCQx8tKV4bvwdVx9MoXYgAdPjiBQLWP8KYTq54/KBshwOWJ3Ubp2Df1hp4cNcYmpoC/MY/wvy1ofjm8+bIeaUJANDVlyLkvDlCzptj8IRHKo5YNUbP6Q4NjTcfnPY1XmLhtMM4fdkeSS8M0Xvsl3Llu7UNR58ut3HlZo2yDlWlCt5P1fHgjrjg/fRtJOb/EopvejSTvZ9GTI1AE+8XCPquLjIztDDq+weYufQ2Jg9spOLoVSs6XA/Tv3KW7UvyFe2MUgFJBUCkYIIiZYJDlczlY6Zy+8ELqqPbwOdwbZhZqROc2aPqy+0vmeWGrafPwck9DXdCzAAAf//XFgDg0fhlWYdXbqSmy79H+nW7hafPjHHzvhUAEV6mGsidb9HoMU5fsUd2TuWq6Zo9ylNuf8lMN2w9cx5O7um4E2IKA6N8dOwZjwX+7rh5peD9tXSWK3795wpc6qUi/FblrVGV5IvwMqlyvV/ojQo7THzHjh3w8PCAvr4+LCws4OPjg8zMTFy9ehUdOnRAlSpVYGJigtatW+P69ety1z58+BDe3t7Q09ODu7s7jh49Knc+OjoaIpEIu3btQtu2bWFgYABPT89CU0mfO3cOrVq1gr6+PmxtbTF+/HhkZmbKzv/8889wcnKCnp4eLC0t0bt37w/GX1FpaAho/WkydPWlCAsx/PAFlYihUT4AID2VH7TvoqUpgU+LSBw67Yyihnw41XoOp1rJOHDaufDFlcyb91PB71Mn93RoawsIvWQmK/MkyhCJcbpw80wr8h6VRXX7HGy+egsbzt3G1OVRqGqTq+qQyp6Kh4mrUoVMcOLj49GvXz8MGTIEYWFhOHXqFHr27AlBEJCeng4/Pz+cO3cOly5dgpOTE7p06YL09HQAgFQqRc+ePaGjo4PLly9j7dq18Pf3L/JxZsyYgcmTJyM0NBTOzs7o168f8vMLPlwiIyPRqVMn9OrVC7du3cK2bdtw7tw5jB07FgBw7do1jB8/HnPnzkV4eDgOHToEb2/vD8ZflJycHKSlpclt5UUt11fYEx6KfY9uYHxQLOYOd0DMw8pbe/M2kUjAN1Mf4u51EzyOMFJ1OOVWi0aPYWSQi8NnnYo837n1Azx+aop7Dy3LOLLyRSQS8I1/hNz7yaxKLvJyRYX6cL18oQOzKpXwC/3/7t8wxOLvamHmAEes+r4mrGxzsGhHOPQNJaoOrYwpI7mpmAlOhWyiio+PR35+Pnr27Ak7OzsAgIeHBwDIVih97ddff4WpqSlOnz6Nbt264dixY7h//z4OHz4MGxsbAMCPP/4oN730a5MnT0bXrl0BAIGBgahTpw4iIiLg6uqKoKAg9O/fHxMmTAAAODk5YcWKFWjdujXWrFmDmJgYGBoaolu3bjA2NoadnR0aNGjwwfiLEhQUhMDAQAVesdLzJFIXo31dYWAsRauuLzF56WNM6e3EJOf/Rs94ADvHTEwe1FDVoZRrnVs/xJVbNfAixaDQOR3tfLT3eoT//u1ZxJWVi+z95NdA1aGUe9dOvWmai7oP3A81xKYLt+Hd7SUOb6uiwsiorFTIGhxPT0+0b98eHh4e+OKLL7Bu3Tq8fFnQl+HZs2cYPnw4nJycYGJiArFYjIyMDNlqpmFhYbC1tZUlNwDeOYNivXr1ZP+2trYGACQmJgIAbt68ieDgYBgZGck2X19fSKVSREVFoUOHDrCzs4ODgwMGDBiAzZs3Iysr64PxF2X69OlITU2VbbGxsQq8esqVn6eBuGg9RNw2wIafqiPqnj56DE1SdVjlwqjp4Wjq/RzThjXAi2d6qg6n3KpmkYGGdeNw4FTRzU/eTaOhq5uPI+ccyziy8mXU9w/QtPULTBtaX+799PK5DrR1BBga58mVN7PIxcvnlXcU1dsy07TwNEoPNrVyVB1K2WITVcWiqamJo0eP4uDBg3B3d8fKlSvh4uKCqKgo+Pn5ITQ0FMuXL8eFCxcQGhoKCwsL5OaWvKpWW/tNla9IVNAvQCotmNIxIyMD33zzDUJDQ2XbzZs38fDhQ9SuXRvGxsa4fv06/vzzT1hbW2P27Nnw9PRESkrKe+Mviq6urmyRs+IsdqZKIg0B2jolmPZSLQkYNT0cXu2SMH1YAzx7ytqs9+nk/QApaXq4FGpb5PnOrR/g4vWahTolVx4CRn3/oOD9NLR+offTw3vGyMsToX6zNz+SqtfKQjWbHITdLL+fFWVNz0ACa7scJCdWsr5wUkE5WwVUIRMcoCDhaNGiBQIDA3Hjxg3o6Ohg9+7dOH/+PMaPH48uXbqgTp060NXVxfPnz2XXubm5ITY2FvHxb+ZHuHTpUokfv2HDhrh37x4cHR0LbTo6Bb+atLS04OPjgwULFuDWrVuIjo7GiRMn3ht/RTJ42lPUbZYOyxo5qOX6CoOnPUU9rwyc3G2u6tBUavSMB2jb9RkWTKuDV5maMLPIgZlFDnR037T9m1nkwMElHTY1XwEAajllwsElHUbivHfdVi2JRAI6eT/EkbOOkEoLfxzZVEtDPZeEd9buVAZv3k/uRb6fsjK0cGSXNYZPiUC9Ji/h6J6OifPCcC9UXKlHUA2b8QQe//98cmuUgdnrIiGRiHDqb7MPX0xqoUL2wbl8+TKOHz+Ojh07olq1arh8+TKSkpLg5uYGJycn/PHHH2jcuDHS0tIwZcoUuSXafXx84OzsDD8/PyxcuBBpaWmYMWNGiWPw9/dH8+bNMXbsWAwbNgyGhoa4d+8ejh49ilWrVmHfvn149OgRvL29YWZmhgMHDkAqlcLFxeW98VckplXyMWXZY5hXy0NWuiaiwvQxo78jrp+t3L8au/V9CgBYsOGG3PElM91w7J+Cps4ufZ6i/6ho2bmFwdcLlakMGtaJg2WVTBw6U3QC07n1AyQlG+LaneplHFn50e3LOADAgg2hcseXzHTFsb8L3iu/LnCEIIgwY+kdaGtLEXLBHD//UHmTQgCoYp2LaauiYGyaj9RkLdy9aoSJPVyRmlzJanAEacGm6D0qoAqZ4IjFYpw5cwbLli1DWloa7OzssHjxYnTu3BlWVlYYMWIEGjZsCFtbW/z444+YPHmy7FoNDQ3s3r0bQ4cORdOmTVGrVi2sWLECnTp1KlEM9erVw+nTpzFjxgy0atUKgiCgdu3a6Nu3L4CCJeJ37dqFgIAAZGdnw8nJCX/++Sfq1KmDsLCwd8ZfkSydbKfqEMqlLvXafbDM5jUO2LzGoQyiKd9C7lRH+wFD3nl+/V+Nsf6vxmUYUfnTxaPtB8vk5Wri5/nO+Hl+5U5q/u2nsfz7AlCpZzIWCe8am0zlVlpaGkxMTNBGoye0RJXs10gJaZqbqjqECiHHs5aqQ6gQdEOL7idH8qSp5Wcqi/IoX8jDyfydSE1NLbU+la+/J3yqj4SWhq5C98qX5uDY07WlGm9pqLB9cIiIiIjepUI2UREREVExVOImKiY4RERE6kqAEhIcpURS5thERURERGqHNThERETqik1UREREpHakUgAKzmMjrZjz4LCJioiIiNQOa3CIiIjUFZuoiIiISO1U4gSHTVRERESkdliDQ0REpK6kAhSeyEZaMWtwmOAQERGpKUGQQlBwNXBFr1cVJjhERETqShAUr4FhHxwiIiKi8oE1OEREROpKUEIfnApag8MEh4iISF1JpYBIwT40FbQPDpuoiIiISO2wBoeIiEhdsYmKiIiI1I0glUJQsImqog4TZxMVERERqR3W4BAREakrNlERERGR2pEKgKhyJjhsoiIiIiK1wxocIiIidSUIABSdB6di1uAwwSEiIlJTglSAoGATlcAEh4iIiMoVQQrFa3A4TJyIiIgIq1evRq1ataCnp4dmzZrhypUrZR4DExwiIiI1JUgFpWwlsW3bNkyaNAlz5szB9evX4enpCV9fXyQmJpbSsywaExwiIiJ1JUiVs5XAkiVLMHz4cAwePBju7u5Yu3YtDAwM8Pvvv5fSkywa++BUQK87fOULeSqOpPwTpLmqDqFCyM/PVnUIFYIm30/FIuVn03u9/uwui867+chTeJ6/fBTEm5aWJndcV1cXurq6csdyc3MREhKC6dOny45paGjAx8cHFy9eVCyQEmKCUwGlp6cDAM4JexV+46q9F6oOoII4peoAiCqf9PR0mJiYlMq9dXR0YGVlhXMJB5RyPyMjI9ja2sodmzNnDgICAuSOPX/+HBKJBJaWlnLHLS0tcf/+faXEUlxMcCogGxsbxMbGwtjYGCKRSNXhACjI7G1tbREbGwuxWKzqcMotvk7Fw9fpw/gaFU95fJ0EQUB6ejpsbGxK7TH09PQQFRWF3Fzl1DoKglDo++bt2pvyhglOBaShoYEaNWqoOowiicXicvMhUp7xdSoevk4fxteoeMrb61RaNTf/pqenBz09vVJ/nH+rUqUKNDU18ezZM7njz549g5WVVZnGwk7GREREpBQ6Ojpo1KgRjh8/LjsmlUpx/PhxeHl5lWksrMEhIiIipZk0aRL8/PzQuHFjNG3aFMuWLUNmZiYGDx5cpnEwwSGl0NXVxZw5c8p9m6yq8XUqHr5OH8bXqHj4OpW9vn37IikpCbNnz0ZCQgLq16+PQ4cOFep4XNpEQkVdZIKIiIjoHdgHh4iIiNQOExwiIiJSO0xwiIiISO0wwSFSsqysLPTq1QtisRgikQgpKSkK31MkEmHPnj0K30dVBEHAiBEjYG5uDpFIhNDQUFWHpJA2bdpgwoQJqg4DgwYNQo8ePVQdRrlR0f9OSLk4iopIyTZu3IizZ8/iwoULqFKlilIm9IqPj4eZmZkSolONQ4cOITg4GKdOnYKDgwOqVKmi6pDUwvLly8tkPSOiiogJDpVLeXl50NbWVnUYHyUyMhJubm6oW7eu0u5Z1jOAKltkZCSsra3xySeflNpj5ObmQkdHp9TuXx6VxWy4RBUVm6gquUOHDqFly5YwNTWFhYUFunXrhsjISABAdHQ0RCIRdu3ahbZt28LAwACenp6FVoRdt24dbG1tYWBggM8//xxLliyBqampXJm///4bDRs2hJ6eHhwcHBAYGIj8/HzZeZFIhDVr1uDTTz+FoaEh5s+fX2rPWSqVYsGCBXB0dISuri5q1qwpe7zbt2+jXbt20NfXh4WFBUaMGIGMjAzZta+bBBYtWgRra2tYWFhgzJgxyMsrWG23TZs2WLx4Mc6cOQORSIQ2bdrInt/bVeempqYIDg4GUPDlPHbsWFhbW0NPTw92dnYICgqSe33+fb2icZalQYMGYdy4cYiJiYFIJEKtWrUglUoRFBQEe3t76Ovrw9PTEzt27JBdI5FIMHToUNl5FxcXLF++vNB9e/Togfnz58PGxgYuLi5l+rykUimmTp0Kc3NzWFlZyS06uGTJEnh4eMDQ0BC2trYYPXq03P9PcHAwTE1NsWfPHjg5OUFPTw++vr6IjY2VlQkICED9+vXxyy+/yP6++vTpg9TUVFmZt5uo2rRpg/Hjx78zLgBISUnBsGHDULVqVYjFYrRr1w43b96Unb958ybatm0LY2NjiMViNGrUCNeuXQMAPH78GN27d4eZmRkMDQ1Rp04dHDjw8Ys57tixAx4eHrL3sY+PDzIzM3H16lV06NBBVgPaunVrXL9+Xe7ahw8fwtvbG3p6enB3d8fRo0flzhf38+vcuXNo1aoV9PX1YWtri/HjxyMzM1N2/ueff5b9H1laWqJ3794fjJ/KCYEqtR07dgg7d+4UHj58KNy4cUPo3r274OHhIUgkEiEqKkoAILi6ugr79u0TwsPDhd69ewt2dnZCXl6eIAiCcO7cOUFDQ0NYuHChEB4eLqxevVowNzcXTExMZI9x5swZQSwWC8HBwUJkZKRw5MgRoVatWkJAQICsDAChWrVqwu+//y5ERkYKjx8/LrXnPHXqVMHMzEwIDg4WIiIihLNnzwrr1q0TMjIyBGtra6Fnz57C7du3hePHjwv29vaCn5+f7Fo/Pz9BLBYLI0eOFMLCwoS9e/cKBgYGwq+//ioIgiC8ePFCGD58uODl5SXEx8cLL168kD2/3bt3y8VhYmIibNiwQRAEQVi4cKFga2srnDlzRoiOjhbOnj0rbNmyRe71eX29MuIsSykpKcLcuXOFGjVqCPHx8UJiYqLwww8/CK6ursKhQ4eEyMhIYcOGDYKurq5w6tQpQRAEITc3V5g9e7Zw9epV4dGjR8J///tfwcDAQNi2bZvcczQyMhIGDBgg3LlzR7hz506ZPafWrVsLYrFYCAgIEB48eCBs3LhREIlEwpEjRwRBEISlS5cKJ06cEKKiooTjx48LLi4uwqhRo2TXb9iwQdDW1hYaN24sXLhwQbh27ZrQtGlT4ZNPPpGVmTNnjmBoaCi0a9dOuHHjhnD69GnB0dFR+Oqrr+Reg88++6zYcQmCIPj4+Ajdu3cXrl69Kjx48ED47rvvBAsLC9l7tU6dOsLXX38thIWFCQ8ePBC2b98uhIaGCoIgCF27dhU6dOgg3Lp1S4iMjBT27t0rnD59+qNew7i4OEFLS0tYsmSJEBUVJdy6dUtYvXq1kJ6eLhw/flz4448/hLCwMOHevXvC0KFDBUtLSyEtLU0QBEGQSCRC3bp1hfbt2wuhoaHC6dOnhQYNGsj9nRTn8ysiIkIwNDQUli5dKjx48EA4f/680KBBA2HQoEGCIAjC1atXBU1NTWHLli1CdHS0cP36dWH58uUfjJ/KByY4JCcpKUkAINy+fVv2AfHbb7/Jzt+9e1cAIISFhQmCIAh9+/YVunbtKneP/v37yyU47du3F3788Ue5Mn/88YdgbW0t2wcgTJgwoRSekby0tDRBV1dXWLduXaFzv/76q2BmZiZkZGTIju3fv1/Q0NAQEhISBEEo+EKxs7MT8vPzZWW++OILoW/fvrL9b7/9VmjdurXcvT+U4IwbN05o166dIJVKi4z739crK86ytHTpUsHOzk4QBEHIzs4WDAwMhAsXLsiVGTp0qNCvX7933mPMmDFCr169ZPt+fn6CpaWlkJOTUyoxv0/r1q2Fli1byh1r0qSJ4O/vX2T5v/76S7CwsJDtb9iwQQAgXLp0SXYsLCxMACBcvnxZEISCBEdTU1N48uSJrMzBgwcFDQ0NIT4+XhCEohOc98V19uxZQSwWC9nZ2XJlateuLfzyyy+CIAiCsbGxEBwcXOTz8PDwkPthooiQkBABgBAdHf3BshKJRDA2Nhb27t0rCIIgHD58WNDS0hKePn0qK3Pw4MEiE5z3fX4NHTpUGDFihNxjnT17VtDQ0BBevXol7Ny5UxCLxbLE6mPjJ9VgE1Ul9/DhQ/Tr1w8ODg4Qi8WoVasWACAmJkZWpl69erJ/W1tbAwASExMBAOHh4WjatKncPd/ev3nzJubOnQsjIyPZNnz4cMTHxyMrK0tWrnHjxkp9bkUJCwtDTk4O2rdvX+Q5T09PGBoayo61aNECUqkU4eHhsmN16tSBpqambN/a2lr2enysQYMGITQ0FC4uLhg/fjyOHDny3uegqjiVISIiAllZWejQoYPce2LTpk2y5lEAWL16NRo1aoSqVavCyMgIv/76q9z7EgA8PDxU1u/m338XgPzre+zYMbRv3x7Vq1eHsbExBgwYgBcvXsi937W0tNCkSRPZvqurK0xNTREWFiY7VrNmTVSvXl227+XlVej/uSRx3bx5ExkZGbCwsJB77aOiomSv/aRJkzBs2DD4+Pjgp59+kvs/GT9+PH744Qe0aNECc+bMwa1bt4r9er3N09MT7du3h4eHB7744gusW7cOL1++BFCw8vTw4cPh5OQEExMTiMViZGRkyP7/w8LCYGtrCxsbG7nX5kOvx9ufXzdv3kRwcLDca+Hr6wupVIqoqCh06NABdnZ2cHBwwIABA7B582bZ/+H74qfygQlOJde9e3ckJydj3bp1uHz5Mi5fvgygoE/Ia//u7CsSiQAU9D8oroyMDAQGBiI0NFS23b59Gw8fPoSenp6s3L+/sEuLvr6+wvd4u/OzSCT64OshEokKjXb5d3+Yhg0bIioqCvPmzcOrV6/Qp08fubb+soqzLLzui7J//36598S9e/dk/XC2bt2KyZMnY+jQoThy5AhCQ0MxePBgufclUDbvmXd51+sbHR2Nbt26oV69eti5cydCQkKwevVqACgUf1nGBRS89tbW1nKve2hoKMLDwzFlyhQABX1/7t69i65du+LEiRNwd3fH7t27AQDDhg3Do0ePMGDAANy+fRuNGzfGypUrPypOTU1NHD16FAcPHoS7uztWrlwJFxcXREVFwc/PD6GhoVi+fDkuXLiA0NBQWFhYfNTr977Pr4yMDHzzzTdyr8XNmzfx8OFD1K5dG8bGxrh+/Tr+/PNPWFtbY/bs2fD09ERKSsp746fygQlOJfbixQuEh4dj5syZaN++Pdzc3Er8C8TFxQVXr16VO/b2fsOGDREeHg5HR8dCm4ZG2b4FnZycoK+vj+PHjxc65+bmhps3b8p1Ejx//jw0NDQU7sBatWpVxMfHy/YfPnwo92seAMRiMfr27Yt169Zh27Zt2LlzJ5KTk8s0zrLg7u4OXV1dxMTEFHo/2NraAih4Pp988glGjx6NBg0awNHRUa4moTwLCQmBVCrF4sWL0bx5czg7OyMuLq5Qufz8fFnnXaCgNjQlJQVubm6yYzExMXLXXrp0SaH/54YNGyIhIQFaWlqFXvt/D913dnbGxIkTceTIEfTs2RMbNmyQnbO1tcXIkSOxa9cufPfdd1i3bt1HxQIUJBwtWrRAYGAgbty4AR0dHezevRvnz5/H+PHj0aVLF9SpUwe6urp4/vy57Do3NzfExsbK/U1dunSpxI/fsGFD3Lt3r8jPptc1g1paWvDx8cGCBQtw69YtREdH48SJE++Nn8oHDhOvxMzMzGBhYYFff/0V1tbWiImJwbRp00p0j3HjxsHb2xtLlixB9+7dceLECRw8eFD2SwkAZs+ejW7duqFmzZro3bs3NDQ0cPPmTdy5cwc//PCDsp/We+np6cHf3x9Tp06Fjo4OWrRogaSkJNy9exf9+/fHnDlz4Ofnh4CAACQlJWHcuHEYMGCAwqvgtmvXDqtWrYKXlxckEgn8/f3lflkuWbIE1tbWaNCgATQ0NPDXX3/Bysqq0Gg0AKUaZ1kwNjbG5MmTMXHiREilUrRs2RKpqak4f/48xGIx/Pz84OTkhE2bNuHw4cOwt7fHH3/8gatXr8Le3l7V4X+Qo6Mj8vLysHLlSnTv3h3nz5/H2rVrC5XT1tbGuHHjsGLFCmhpaWHs2LFo3ry5XBOvnp4e/Pz8sGjRIqSlpWH8+PHo06fPR08b4OPjAy8vL/To0QMLFiyQJV/79+/H559/jjp16mDKlCno3bs37O3t8eTJE1y9ehW9evUCAEyYMAGdO3eGs7MzXr58iZMnT8olZCVx+fJlHD9+HB07dkS1atVw+fJlJCUlwc3NDU5OTvjjjz/QuHFjpKWlYcqUKXK1rz4+PnB2doafnx8WLlyItLQ0zJgxo8Qx+Pv7o3nz5hg7diyGDRsGQ0ND3Lt3D0ePHsWqVauwb98+PHr0CN7e3jAzM8OBAwcglUrh4uLy3vipfGANTiWmoaGBrVu3IiQkBHXr1sXEiROxcOHCEt2jRYsWWLt2LZYsWQJPT08cOnQIEydOlGt68vX1xb59+3DkyBE0adIEzZs3x9KlS2FnZ6fsp1Qss2bNwnfffYfZs2fDzc0Nffv2RWJiIgwMDHD48GEkJyejSZMm6N27N9q3b49Vq1Yp/JiLFy+Gra0tWrVqha+++gqTJ0+GgYGB7LyxsTEWLFiAxo0bo0mTJoiOjsaBAweKrOEqzTjLyrx58zBr1iwEBQXBzc0NnTp1wv79+2UJzDfffIOePXuib9++aNasGV68eIHRo0erOOri8fT0xJIlS/Cf//wHdevWxebNm+WG/L9mYGAAf39/fPXVV2jRogWMjIywbds2uTKOjo7o2bMnunTpgo4dO6JevXr4+eefPzo2kUiEAwcOwNvbG4MHD4azszO+/PJLPH78GJaWltDU1MSLFy8wcOBAODs7o0+fPujcuTMCAwMBFAzfHzNmjOz/zNnZ+aPjEYvFOHPmDLp06QJnZ2fMnDkTixcvRufOnbF+/Xq8fPkSDRs2xIABAzB+/HhUq1ZNdq2GhgZ2796NV69eoWnTphg2bNhHTS1Rr149nD59Gg8ePECrVq3QoEEDzJ49W9a3x9TUFLt27UK7du3g5uaGtWvX4s8//0SdOnXeGz+VDyLh7Y4BRAoaPnw47t+/j7Nnz6o6FKJyKTg4GBMmTHjvMh4BAQHYs2dPhV/WgkhV2ERFClu0aBE6dOgAQ0NDHDx4EBs3blToVyYREZGimOCQwq5cuYIFCxYgPT0dDg4OWLFiBYYNG6bqsIiIqBJjExURERGpHXYyJiIiIrXDBIeIiIjUDhMcIiIiUjtMcIiIiEjtMMEhIiIitcMEh4g+yqBBg9CjRw/Zfps2bTBhwoQyj+PUqVMQiUTvnTRPJBJhz549xb5nQEAA6tevr1Bc0dHREIlEnKiPSEWY4BCpkUGDBkEkEkEkEkFHRweOjo6YO3cu8vPzS/2xd+3ahXnz5hWrbHGSEiIiRXCiPyI106lTJ2zYsAE5OTk4cOAAxowZA21tbUyfPr1Q2dzcXNmqyYoyNzdXyn2IiJSBNThEakZXVxdWVlaws7PDqFGj4OPjg3/++QfAm2al+fPnw8bGBi4uLgCA2NhY9OnTB6ampjA3N8dnn32G6Oho2T0lEgkmTZoEU1NTWFhYYOrUqXh7jtC3m6hycnLg7+8PW1tb6OrqwtHREevXr0d0dDTatm0LoGBFe5FIhEGDBgEApFIpgoKCYG9vD319fXh6emLHjh1yj3PgwAE4OztDX18fbdu2lYuzuPz9/eHs7AwDAwM4ODhg1qxZyMvLK1Tul19+ga2tLQwMDNCnTx+kpqbKnf/tt9/g5uYGPT09uLq6cokSonKECQ6RmtPX10dubq5s//jx4wgPD8fRo0exb98+5OXlwdfXF8bGxjh79izOnz8PIyMjdOrUSXbd4sWLERwcjN9//x3nzp1DcnIydu/e/d7HHThwIP7880+sWLECYWFh+OWXX2BkZARbW1vs3LkTABAeHo74+HgsX74cABAUFIRNmzZh7dq1uHv3LiZOnIivv/4ap0+fBlCQiPXs2RPdu3dHaGgohg0bhmnTppX4NTE2NkZwcDDu3buH5cuXY926dVi6dKlcmYiICGzfvh179+7FoUOHcOPGDbkVzTdv3ozZs2dj/vz5CAsLw48//ohZs2Zh48aNJY6HiEqBQERqw8/PT/jss88EQRAEqVQqHD16VNDV1RUmT54sO29paSnk5OTIrvnjjz8EFxcXQSqVyo7l5OQI+vr6wuHDhwVBEARra2thwYIFsvN5eXlCjRo1ZI8lCILQunVr4dtvvxUEQRDCw8MFAMLRo0eLjPPkyZMCAOHly5eyY9nZ2YKBgYFw4cIFubJDhw4V+vXrJwiCIEyfPl1wd3eXO+/v71/oXm8DIOzevfud5xcuXCg0atRItj9nzhxBU1NTePLkiezYwYMHBQ0NDSE+Pl4QBEGoXbu2sGXLFrn7zJs3T/Dy8hIEQRCioqIEAMKNGzfe+bhEVHrYB4dIzezbtw9GRkbIy8uDVCrFV199hYCAANl5Dw8PuX43N2/eREREBIyNjeXuk52djcjISKSmpiI+Ph7NmjWTndPS0kLjxo0LNVO9FhoaCk1NTbRu3brYcUdERCArKwsdOnSQO56bm4sGDRoAAMLCwuTiAAAvL69iP8Zr27Ztw4oVKxAZGYmMjAzk5+dDLBbLlalZsyaqV68u9zhSqRTh4eEwNjZGZGQkhg4diuHDh8vK5Ofnw8TEpMTxEJHyMcEhUjNt27bFmjVroKOjAxsbG2hpyf+ZGxoayu1nZGSgUaNG2Lx5c6F7Va1a9aNi0NfXL/E1GRkZAID9+/fLJRZAQb8iZbl48SL69++PwMBA+Pr6wsTEBFu3bsXixYtLHOu6desKJVyamppKi5WIPh4THCI1Y2hoCEdHx2KXb9iwIbZt24Zq1aoVqsV4zdraGpcvX4a3tzeAgpqKkJAQNGzYsMjyHh4ekEqlOH36NHx8fAqdf12DJJFIZMfc3d2hq6uLmJiYd9b8uLm5yTpMv3bp0qUPP8l/uXDhAuzs7DBjxgzZscePHxcqFxMTg7i4ONjY2MgeR0NDAy4uLrC0tISNjQ0ePXqE/v37l+jxiahssJMxUSXXv39/VKlSBZ999hnOnj2LqKgonDp1CuPHj8eTJ08AAN9++y1++ukn7NmzB/fv38fo0aPfO4dNrVq14OfnhyFDhmDPnj2ye27fvh0AYGdnB5FIhH379iEpKQkZGRkwNjbG5MmTMXHiRGzcuBGRkZG4fv06Vq5cKeu4O3LkSDx8+BBTpkxBeHg4tmzZguDg4BI9XycnJ8TExGDr1q2IjIzEihUriuwwraenBz8/P9y8eRNnz57F+PHj0adPH1hZWQEAAgMDERQUhBUrVuDBgwe4ffs2NmzYgCVLlpQoHiIqHUxwiCo5AwMDnDlzBjVr1kTPnj3h5uaGoUOHIjs7W1aj891332HAgAHw8/ODl5cXjI2N8fnnn7/3vmvWrEHv3r0xevRouLq6Yvjw4cjMzAQAVK9eHYGBgZg2bRosLS0xduxYAMC8efMwa9YsBAUFwc3NDZ06dcL+/fthb28PoKBfzM6dO7Fnzx54enpi7dq1+PHHH0v0fD/99FNMnDgRY8eORf369XHhwgXMmjWrUDlHR0f07NkTXbp0QceOHVGvXj25YeDDhg3Db7/9hg0bNsDDwwOtW7dGcHCwLFYiUi2R8K5egkREREQVFGtwiIiISO0wwSEiIiK1wwSHiIiI1A4THCIiIlI7THCIiIhI7TDBISIiIrXDBIeIiIjUDhMcIiIiUjtMcIiIiEjtMMEhIiIitcMEh4iIiNTO/wCxpNzP/xWaowAAAABJRU5ErkJggg==",
      "text/plain": [
       "<Figure size 640x480 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "cm = confusion_matrix(y_test, y_pred, labels=model.classes_)\n",
    "disp = ConfusionMatrixDisplay(confusion_matrix=cm,\n",
    "                               display_labels=model.classes_)\n",
    "disp.plot()\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "cc8fa3d8-ffa2-4da3-bccb-ccf2f1927910",
   "metadata": {},
   "source": [
    "## Cross validation\n",
    "https://scikit-learn.org/stable/modules/cross_validation.html"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "id": "bb94bd7a-705b-43b1-93a6-ab680ab91d43",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Mean f1_macro: 0.40828471010718187\n",
      "Standard deviation f1_macro: 0.01451110152051428\n"
     ]
    }
   ],
   "source": [
    "# Perform cross-validation and print the mean accuracy\n",
    "scoring = 'f1_macro'\n",
    "scores = cross_val_score(model, X, y, cv=5, scoring=scoring, n_jobs = -1)\n",
    "print(f\"Mean {scoring}: {scores.mean()}\")\n",
    "print(f\"Standard deviation {scoring}: {scores.std()}\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "8d358895-ae2b-4952-8e2d-631110754f9c",
   "metadata": {},
   "source": [
    "# Naive Bayes from scratch"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 102,
   "id": "ea5881b2-7baf-4616-9b7d-8f48c98caf63",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "1040     happiness\n",
       "25797         fear\n",
       "16279         fear\n",
       "1840         anger\n",
       "26616         fear\n",
       "Name: emotion, dtype: object"
      ]
     },
     "execution_count": 102,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "y_train.head()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "0ebe7bf7-f47b-41b9-945f-34c5b0728d2b",
   "metadata": {},
   "source": [
    "### Calculate priors and vocabulary"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 103,
   "id": "bd9c7da0-2b53-41e7-84b0-1b877a7ad046",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{'happiness': 0.1820796460176991, 'fear': 0.3911504424778761, 'anger': 0.14115044247787611, 'confusion': 0.18871681415929203, 'sadness': 0.09690265486725663}\n"
     ]
    }
   ],
   "source": [
    "from collections import Counter\n",
    "\n",
    "count_c = Counter(y_train)\n",
    "\n",
    "priors = {k:v/len(y_train) for k, v in count_c.items()}\n",
    "print(priors)\n",
    "\n",
    "train_strings = \" \".join(X_train)\n",
    "train_tokens = word_tokenize(train_strings)\n",
    "vocabulary = set(train_tokens)\n",
    "V = len(vocabulary)\n",
    "classes = set(y_train)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "3d55c7f6-cc07-4307-8619-4647a0cccdab",
   "metadata": {},
   "source": [
    "### Calculate likelihoods "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 104,
   "id": "f3239b70-a813-47da-989d-01733bb628d2",
   "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>document</th>\n",
       "      <th>emotion</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>a man gets out of prison. he's walking around ...</td>\n",
       "      <td>happiness</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>another the agency dream.  at first i was outs...</td>\n",
       "      <td>fear</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>i was at memorial stadium, supposed to do a dj...</td>\n",
       "      <td>fear</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>i am driving my van, but can't find the keys t...</td>\n",
       "      <td>anger</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>there are four of us and we were trying to ste...</td>\n",
       "      <td>fear</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                            document    emotion\n",
       "0  a man gets out of prison. he's walking around ...  happiness\n",
       "1  another the agency dream.  at first i was outs...       fear\n",
       "2  i was at memorial stadium, supposed to do a dj...       fear\n",
       "3  i am driving my van, but can't find the keys t...      anger\n",
       "4  there are four of us and we were trying to ste...       fear"
      ]
     },
     "execution_count": 104,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_df = pd.DataFrame(zip(X_train, y_train), columns = [\"document\", \"emotion\"])\n",
    "train_df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 113,
   "id": "14f4a5a5-fe90-42ec-b313-4242d0da56d1",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "7036\n",
      "0.04408346854268335\n"
     ]
    }
   ],
   "source": [
    "mega_dict = dict()\n",
    "prob_dict = dict()\n",
    "k = 1\n",
    "\n",
    "for c in classes:\n",
    "    class_df = train_df.query(f\"emotion == '{c}'\")\n",
    "    mega_document = class_df[\"document\"]\n",
    "    mega_strings = \" \".join(mega_document)\n",
    "    mega_tokens = word_tokenize(mega_strings)\n",
    "    mega_count = Counter(mega_tokens)\n",
    "    mega_dict[c] = mega_count\n",
    "    prob_dict[c] = dict()\n",
    "    \n",
    "\n",
    "    for token in vocabulary:\n",
    "        prob_dict[c][token] = (mega_dict[c][token] + k ) / (len(mega_tokens) + V)\n",
    "\n",
    "print(mega_dict[\"anger\"][\"i\"])\n",
    "print(prob_dict[\"anger\"][\"i\"])"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "a6102d5f-01cd-448d-b17e-c6bc34f43aef",
   "metadata": {},
   "source": [
    "### Prediction"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 114,
   "id": "e592166b-db31-423f-9da1-55d713ddc13e",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{'anger': 3.8438023650338485e-12, 'happiness': 5.738926582807367e-12, 'confusion': 7.1664344130907345e-12, 'sadness': 2.4003655911873422e-12, 'fear': 2.254136682259967e-11}\n",
      "fear\n"
     ]
    }
   ],
   "source": [
    "import math\n",
    "\n",
    "test_tokens = word_tokenize(\"i don't like it\")\n",
    "\n",
    "pred_dict = dict()\n",
    "\n",
    "for c in classes:\n",
    "    sum = math.log2(priors[c])\n",
    "    for token in test_tokens:\n",
    "        #print(prob_dict[c][token])\n",
    "        sum += math.log2(prob_dict[c][token])\n",
    "    sum = math.pow(2, sum)\n",
    "    pred_dict[c] = sum\n",
    "print(pred_dict)\n",
    "print(max(pred_dict, key=pred_dict.get))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 115,
   "id": "2b7afc94-4bb6-47f3-9880-b66cd63b7475",
   "metadata": {},
   "outputs": [],
   "source": [
    "import math\n",
    "\n",
    "test_tokens = [word_tokenize(text) for text in X_test]\n",
    "\n",
    "pred_dict = dict()\n",
    "preds = list()\n",
    "\n",
    "for report_tokens in test_tokens:\n",
    "    for c in classes:\n",
    "        sum = math.log2(priors[c])\n",
    "        for token in report_tokens:\n",
    "            if token in prob_dict[c]: # remove unknown tokens\n",
    "                sum += math.log2(prob_dict[c][token])\n",
    "        sum = math.pow(2, sum)\n",
    "        pred_dict[c] = sum\n",
    "    preds.append(max(pred_dict, key=pred_dict.get))"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "584e7728-e8ce-4530-9a91-e82c8cc415f2",
   "metadata": {},
   "source": [
    "### Evaluation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 116,
   "id": "d410d612-9800-4a3a-901b-98862e531dc0",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "              precision    recall  f1-score   support\n",
      "\n",
      "       anger       0.75      0.18      0.30       754\n",
      "   confusion       0.10      0.53      0.17        38\n",
      "        fear       0.33      0.54      0.41       262\n",
      "   happiness       0.21      0.53      0.30        72\n",
      "     sadness       0.03      1.00      0.06         4\n",
      "\n",
      "    accuracy                           0.30      1130\n",
      "   macro avg       0.28      0.56      0.25      1130\n",
      "weighted avg       0.59      0.30      0.32      1130\n",
      "\n"
     ]
    }
   ],
   "source": [
    "from sklearn.metrics import classification_report\n",
    "\n",
    "print(classification_report(preds, list(y_test)))"
   ]
  }
 ],
 "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": "3.9.18"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}