ligthgbm_models/fraud-artem-new-val-no-day.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "_cell_guid": "b1076dfc-b9ad-4769-8c92-a6c4dae69d19",
    "_kg_hide-input": true,
    "_uuid": "8f2839f25d086af736a60e9eeb907d3b93b6e0e5"
   },
   "outputs": [],
   "source": [
    "# safe downcast\n",
    "def sd(col, max_loss_limit=0.001, avg_loss_limit=0.001, na_loss_limit=0, n_uniq_loss_limit=0, fillna=0):\n",
    "    \"\"\"\n",
    "    max_loss_limit - don't allow any float to lose precision more than this value. Any values are ok for GBT algorithms as long as you don't unique values.\n",
    "                     See https://en.wikipedia.org/wiki/Half-precision_floating-point_format#Precision_limitations_on_decimal_values_in_[0,_1]\n",
    "    avg_loss_limit - same but calculates avg throughout the series.\n",
    "    na_loss_limit - not really useful.\n",
    "    n_uniq_loss_limit - very important parameter. If you have a float field with very high cardinality you can set this value to something like n_records * 0.01 in order to allow some field relaxing.\n",
    "    \"\"\"\n",
    "    is_float = str(col.dtypes)[:5] == 'float'\n",
    "    na_count = col.isna().sum()\n",
    "    n_uniq = col.nunique(dropna=False)\n",
    "    try_types = ['float16', 'float32']\n",
    "\n",
    "    if na_count <= na_loss_limit:\n",
    "        try_types = ['int8', 'int16', 'float16', 'int32', 'float32']\n",
    "\n",
    "    for type in try_types:\n",
    "        col_tmp = col\n",
    "\n",
    "        # float to int conversion => try to round to minimize casting error\n",
    "        if is_float and (str(type)[:3] == 'int'):\n",
    "            col_tmp = col_tmp.copy().fillna(fillna).round()\n",
    "\n",
    "        col_tmp = col_tmp.astype(type)\n",
    "        max_loss = (col_tmp - col).abs().max()\n",
    "        avg_loss = (col_tmp - col).abs().mean()\n",
    "        na_loss = np.abs(na_count - col_tmp.isna().sum())\n",
    "        n_uniq_loss = np.abs(n_uniq - col_tmp.nunique(dropna=False))\n",
    "\n",
    "        if max_loss <= max_loss_limit and avg_loss <= avg_loss_limit and na_loss <= na_loss_limit and n_uniq_loss <= n_uniq_loss_limit:\n",
    "            return col_tmp\n",
    "\n",
    "    # field can't be converted\n",
    "    return col\n",
    "\n",
    "\n",
    "def reduce_mem_usage(df, deep=True, verbose=False, obj_to_cat=False):\n",
    "    numerics = ['int16', 'uint16', 'int32', 'uint32', 'int64', 'uint64', 'float16', 'float32', 'float64']\n",
    "    start_mem = df.memory_usage(deep=deep).sum() / 1024 ** 2\n",
    "    for col in df.columns:\n",
    "        col_type = df[col].dtypes\n",
    "\n",
    "        # collect stats\n",
    "        na_count = df[col].isna().sum()\n",
    "        n_uniq = df[col].nunique(dropna=False)\n",
    "        \n",
    "        # numerics\n",
    "        if col_type in numerics:\n",
    "            df[col] = sd(df[col])\n",
    "\n",
    "        # strings\n",
    "        if (col_type == 'object') and obj_to_cat:\n",
    "            df[col] = df[col].astype('category')\n",
    "        \n",
    "        if verbose:\n",
    "            print(f'Column {col}: {col_type} -> {df[col].dtypes}, na_count={na_count}, n_uniq={n_uniq}')\n",
    "        new_na_count = df[col].isna().sum()\n",
    "        if (na_count != new_na_count):\n",
    "            print(f'Warning: column {col}, {col_type} -> {df[col].dtypes} lost na values. Before: {na_count}, after: {new_na_count}')\n",
    "        new_n_uniq = df[col].nunique(dropna=False)\n",
    "        if (n_uniq != new_n_uniq):\n",
    "            print(f'Warning: column {col}, {col_type} -> {df[col].dtypes} lost unique values. Before: {n_uniq}, after: {new_n_uniq}')\n",
    "\n",
    "    end_mem = df.memory_usage(deep=deep).sum() / 1024 ** 2\n",
    "    percent = 100 * (start_mem - end_mem) / start_mem\n",
    "    if verbose:\n",
    "        print('Mem. usage decreased from {:5.2f} Mb to {:5.2f} Mb ({:.1f}% reduction)'.format(start_mem, end_mem, percent))\n",
    "    return df"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## <span id=\"1\"></span> ** 1. Importing Libraries and Reading the Dataset **"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "_cell_guid": "79c7e3d0-c299-4dcb-8224-4455121ee9b0",
    "_uuid": "d629ff2d2480ee46fbb7e2d37f6b5fab8052498a"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "/kaggle/input/ieee-fraud-detection/train_identity.csv\n",
      "/kaggle/input/ieee-fraud-detection/test_identity.csv\n",
      "/kaggle/input/ieee-fraud-detection/test_transaction.csv\n",
      "/kaggle/input/ieee-fraud-detection/sample_submission.csv\n",
      "/kaggle/input/ieee-fraud-detection/train_transaction.csv\n"
     ]
    }
   ],
   "source": [
    "import os\n",
    "for dirname, _, filenames in os.walk('/kaggle/input'):\n",
    "    for filename in filenames:\n",
    "        print(os.path.join(dirname, filename))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import pandas as pd\n",
    "from sklearn import preprocessing\n",
    "import gc, datetime, random\n",
    "import lightgbm as lgb\n",
    "import xgboost as xgb\n",
    "from sklearn.model_selection import KFold\n",
    "from sklearn.preprocessing import LabelEncoder\n",
    "import matplotlib.pyplot as plt\n",
    "import seaborn as sns\n",
    "from sklearn.metrics import roc_auc_score\n",
    "\n",
    "import pickle\n",
    "\n",
    "def save_to_disk(obj, filename):\n",
    "    with open(filename, 'wb') as handle:\n",
    "        pickle.dump(obj, handle, protocol=pickle.HIGHEST_PROTOCOL)\n",
    "\n",
    "pd.options.display.max_rows = 4000\n",
    "\n",
    "def seed_everything(seed=0):\n",
    "    random.seed(seed)\n",
    "    os.environ['PYTHONHASHSEED'] = str(seed)\n",
    "    np.random.seed(seed)\n",
    "\n",
    "SEED = 42\n",
    "seed_everything(SEED)\n",
    "START_DATE = datetime.datetime.strptime('2017-11-30', '%Y-%m-%d')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train shape:  (590540, 433)\n",
      "Test shape:  (506691, 432)\n",
      "CPU times: user 46.4 s, sys: 17.7 s, total: 1min 4s\n",
      "Wall time: 58.6 s\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "train_transaction = pd.read_csv('/kaggle/input/ieee-fraud-detection/train_transaction.csv', index_col='TransactionID')\n",
    "test_transaction = pd.read_csv('/kaggle/input/ieee-fraud-detection/test_transaction.csv', index_col='TransactionID')\n",
    "\n",
    "train_identity = pd.read_csv('/kaggle/input/ieee-fraud-detection/train_identity.csv', index_col='TransactionID')\n",
    "test_identity = pd.read_csv('/kaggle/input/ieee-fraud-detection/test_identity.csv', index_col='TransactionID')\n",
    "\n",
    "sample_submission = pd.read_csv('/kaggle/input/ieee-fraud-detection/sample_submission.csv', index_col='TransactionID')\n",
    "\n",
    "train = train_transaction.merge(train_identity, how='left', left_index=True, right_index=True)\n",
    "test = test_transaction.merge(test_identity, how='left', left_index=True, right_index=True)\n",
    "\n",
    "print(\"Train shape: \", train.shape)\n",
    "print(\"Test shape: \", test.shape)\n",
    "\n",
    "y = train['isFraud'].copy()\n",
    "del train_transaction, train_identity, test_transaction, test_identity\n",
    "\n",
    "# Drop target, fill in NaNs\n",
    "train = train.drop('isFraud', axis=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "train = reduce_mem_usage(train)\n",
    "test = reduce_mem_usage(test)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## <span id=\"2\"></span> ** 2. Feature Engineering **"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### <span id=\"3\"></span> ** Add New Features **"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "def addNewFeatures(data): \n",
    "    data['uid'] = data['card1'].astype(str)+'_'+data['card2'].astype(str)\n",
    "\n",
    "    data['uid2'] = data['uid'].astype(str)+'_'+data['card3'].astype(str)+'_'+data['card5'].astype(str)\n",
    "\n",
    "    data['uid3'] = data['uid2'].astype(str)+'_'+data['addr1'].astype(str)+'_'+data['addr2'].astype(str)\n",
    "\n",
    "    data['D9'] = np.where(data['D9'].isna(),0,1)\n",
    "    \n",
    "    return data\n",
    "\n",
    "train = addNewFeatures(train)\n",
    "test = addNewFeatures(test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "i_cols = ['card1','card2','card3','card5','uid','uid2','uid3']\n",
    "\n",
    "for col in i_cols:\n",
    "    for agg_type in ['mean','std']:\n",
    "        new_col_name = col+'_TransactionAmt_'+agg_type\n",
    "        temp_df = pd.concat([train[[col, 'TransactionAmt']], test[[col,'TransactionAmt']]])\n",
    "        #temp_df['TransactionAmt'] = temp_df['TransactionAmt'].astype(int)\n",
    "        temp_df = temp_df.groupby([col])['TransactionAmt'].agg([agg_type]).reset_index().rename(\n",
    "                                                columns={agg_type: new_col_name})\n",
    "\n",
    "        temp_df.index = list(temp_df[col])\n",
    "        temp_df = temp_df[new_col_name].to_dict()   \n",
    "\n",
    "        train[new_col_name] = train[col].map(temp_df)\n",
    "        test[new_col_name]  = test[col].map(temp_df)\n",
    "\n",
    "train = train.replace(np.inf,999)\n",
    "test = test.replace(np.inf,999)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "train['TransactionAmt'] = np.log1p(train['TransactionAmt'])\n",
    "test['TransactionAmt'] = np.log1p(test['TransactionAmt'])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### <span id=\"4\"></span> ** Handle Email Domains **"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "emails = {'gmail': 'google', 'att.net': 'att', 'twc.com': 'spectrum', 'scranton.edu': 'other', 'optonline.net': 'other',\n",
    "          'hotmail.co.uk': 'microsoft', 'comcast.net': 'other', 'yahoo.com.mx': 'yahoo', 'yahoo.fr': 'yahoo',\n",
    "          'yahoo.es': 'yahoo', 'charter.net': 'spectrum', 'live.com': 'microsoft', 'aim.com': 'aol', 'hotmail.de': 'microsoft',\n",
    "          'centurylink.net': 'centurylink', 'gmail.com': 'google', 'me.com': 'apple', 'earthlink.net': 'other', \n",
    "          'gmx.de': 'other', 'web.de': 'other', 'cfl.rr.com': 'other', 'hotmail.com': 'microsoft', 'protonmail.com': 'other',\n",
    "          'hotmail.fr': 'microsoft', 'windstream.net': 'other', 'outlook.es': 'microsoft', 'yahoo.co.jp': 'yahoo',\n",
    "          'yahoo.de': 'yahoo', 'servicios-ta.com': 'other', 'netzero.net': 'other', 'suddenlink.net': 'other',\n",
    "          'roadrunner.com': 'other', 'sc.rr.com': 'other', 'live.fr': 'microsoft', 'verizon.net': 'yahoo',\n",
    "          'msn.com': 'microsoft', 'q.com': 'centurylink', 'prodigy.net.mx': 'att', 'frontier.com': 'yahoo',\n",
    "          'anonymous.com': 'other', 'rocketmail.com': 'yahoo', 'sbcglobal.net': 'att', 'frontiernet.net': 'yahoo',\n",
    "          'ymail.com': 'yahoo', 'outlook.com': 'microsoft', 'mail.com': 'other', 'bellsouth.net': 'other',\n",
    "          'embarqmail.com': 'centurylink', 'cableone.net': 'other', 'hotmail.es': 'microsoft', 'mac.com': 'apple',\n",
    "          'yahoo.co.uk': 'yahoo', 'netzero.com': 'other', 'yahoo.com': 'yahoo', 'live.com.mx': 'microsoft', 'ptd.net': 'other',\n",
    "          'cox.net': 'other', 'aol.com': 'aol', 'juno.com': 'other', 'icloud.com': 'apple'}\n",
    "us_emails = ['gmail', 'net', 'edu']\n",
    "\n",
    "for c in ['P_emaildomain', 'R_emaildomain']:\n",
    "    train[c + '_bin'] = train[c].map(emails)\n",
    "    test[c + '_bin'] = test[c].map(emails)\n",
    "    \n",
    "    train[c + '_suffix'] = train[c].map(lambda x: str(x).split('.')[-1])\n",
    "    test[c + '_suffix'] = test[c].map(lambda x: str(x).split('.')[-1])\n",
    "    \n",
    "    train[c + '_suffix'] = train[c + '_suffix'].map(lambda x: x if str(x) not in us_emails else 'us')\n",
    "    test[c + '_suffix'] = test[c + '_suffix'].map(lambda x: x if str(x) not in us_emails else 'us')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### <span id=\"5\"></span> ** Handle P Email Domain and R Email Domain **"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "p = 'P_emaildomain'\n",
    "r = 'R_emaildomain'\n",
    "uknown = 'email_not_provided'\n",
    "\n",
    "def setDomain(df):\n",
    "    df[p] = df[p].fillna(uknown)\n",
    "    df[r] = df[r].fillna(uknown)\n",
    "    \n",
    "    # Check if P_emaildomain matches R_emaildomain\n",
    "    df['email_check'] = np.where((df[p]==df[r])&(df[p]!=uknown),1,0)\n",
    "\n",
    "    df[p+'_prefix'] = df[p].apply(lambda x: x.split('.')[0])\n",
    "    df[r+'_prefix'] = df[r].apply(lambda x: x.split('.')[0])\n",
    "    \n",
    "    return df\n",
    "    \n",
    "train=setDomain(train)\n",
    "test=setDomain(test)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### <span id=\"6\"></span> ** Set Time **"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "def setTime(df):\n",
    "    df['TransactionDT'] = df['TransactionDT'].fillna(df['TransactionDT'].median())\n",
    "    # Temporary\n",
    "    df['DT'] = df['TransactionDT'].apply(lambda x: (START_DATE + datetime.timedelta(seconds = x)))\n",
    "#     df['DT_M'] = (df['DT'].dt.year-2017)*12 + df['DT'].dt.month\n",
    "#     df['DT_W'] = (df['DT'].dt.year-2017)*52 + df['DT'].dt.weekofyear\n",
    "#     df['DT_D'] = (df['DT'].dt.year-2017)*365 + df['DT'].dt.dayofyear\n",
    "    \n",
    "    df['DT_hour'] = df['DT'].dt.hour\n",
    "    df['DT_day_week'] = df['DT'].dt.dayofweek\n",
    "#     df['DT_day'] = df['DT'].dt.day\n",
    "    \n",
    "    return df\n",
    "    \n",
    "train=setTime(train)\n",
    "test=setTime(test)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Normalize D* features by trend"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "trends = {\n",
    "    'D1': [(453.557, 613.017), (335.379, 487.827)],\n",
    "    'D2': [(453.557, 613.017), (335.379, 487.827)],\n",
    "    'D3': [(515.803, 630.), (335.163, 457.923)],\n",
    "    'D4': [(515.298, 689.073), (335.163, 488.54)],\n",
    "    'D5': [(514.,670.), (335.595, 465.903)],\n",
    "    'D6': [(515.514, 602.702), (335.595, 389.906)],\n",
    "    'D7': [(515.947, 559.162), (335.379, 377.205)],\n",
    "    'D8': [(514.865, 1298.05), (335.163, 1113.72)],\n",
    "    'D10': [(515.37, 675.59), (335.451, 492.4)],\n",
    "    'D11': [(515.298, 674.197), (335.379,484.004)],\n",
    "    'D12': [(515.947,651.695), (335.812,470.355)],\n",
    "    'D13': [(499.512,810.655), (337.758,702.105)],\n",
    "    'D14': [(515.298,871.374), (336.028,697.996)],\n",
    "    'D15': [(515.298, 678.383), (335.163, 483.718)]\n",
    "}\n",
    "\n",
    "def normalize_by_trend(df, col, period):\n",
    "    if col in trends:\n",
    "        end_x, end_y = trends[col][0]\n",
    "        start_x, start_y = trends[col][1]\n",
    "        trend = period.apply(lambda x: start_y + (x - start_x)*(end_y-start_y)/(end_x-start_x))\n",
    "        return df[col]/trend\n",
    "    return df[col]\n",
    "\n",
    "for df in [train, test]:\n",
    "    dt = df['TransactionDT'].apply(lambda x: (START_DATE + datetime.timedelta(seconds = x)))\n",
    "    dt_d = ((dt.dt.year-2017)*365 + dt.dt.dayofyear).astype(np.int16)\n",
    "    for col in trends:\n",
    "        df[col] = normalize_by_trend(df, col, dt_d)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### <span id=\"7\"></span> ** Handle Browser Version **"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "train[\"lastest_browser\"] = np.zeros(train.shape[0])\n",
    "test[\"lastest_browser\"] = np.zeros(test.shape[0])\n",
    "\n",
    "def setBrowser(df):\n",
    "    df.loc[df[\"id_31\"]==\"samsung browser 7.0\",'lastest_browser']=1\n",
    "    df.loc[df[\"id_31\"]==\"opera 53.0\",'lastest_browser']=1\n",
    "    df.loc[df[\"id_31\"]==\"mobile safari 10.0\",'lastest_browser']=1\n",
    "    df.loc[df[\"id_31\"]==\"google search application 49.0\",'lastest_browser']=1\n",
    "    df.loc[df[\"id_31\"]==\"firefox 60.0\",'lastest_browser']=1\n",
    "    df.loc[df[\"id_31\"]==\"edge 17.0\",'lastest_browser']=1\n",
    "    df.loc[df[\"id_31\"]==\"chrome 69.0\",'lastest_browser']=1\n",
    "    df.loc[df[\"id_31\"]==\"chrome 67.0 for android\",'lastest_browser']=1\n",
    "    df.loc[df[\"id_31\"]==\"chrome 63.0 for android\",'lastest_browser']=1\n",
    "    df.loc[df[\"id_31\"]==\"chrome 63.0 for ios\",'lastest_browser']=1\n",
    "    df.loc[df[\"id_31\"]==\"chrome 64.0\",'lastest_browser']=1\n",
    "    df.loc[df[\"id_31\"]==\"chrome 64.0 for android\",'lastest_browser']=1\n",
    "    df.loc[df[\"id_31\"]==\"chrome 64.0 for ios\",'lastest_browser']=1\n",
    "    df.loc[df[\"id_31\"]==\"chrome 65.0\",'lastest_browser']=1\n",
    "    df.loc[df[\"id_31\"]==\"chrome 65.0 for android\",'lastest_browser']=1\n",
    "    df.loc[df[\"id_31\"]==\"chrome 65.0 for ios\",'lastest_browser']=1\n",
    "    df.loc[df[\"id_31\"]==\"chrome 66.0\",'lastest_browser']=1\n",
    "    df.loc[df[\"id_31\"]==\"chrome 66.0 for android\",'lastest_browser']=1\n",
    "    df.loc[df[\"id_31\"]==\"chrome 66.0 for ios\",'lastest_browser']=1\n",
    "    return df\n",
    "\n",
    "train=setBrowser(train)\n",
    "test=setBrowser(test)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### <span id=\"8\"></span> ** Handle Device Type **"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "def setDevice(df):\n",
    "    df['DeviceInfo'] = df['DeviceInfo'].fillna('unknown_device').str.lower()\n",
    "    \n",
    "    df['device_name'] = df['DeviceInfo'].str.split('/', expand=True)[0]\n",
    "\n",
    "    df.loc[df['device_name'].str.contains('SM', na=False), 'device_name'] = 'Samsung'\n",
    "    df.loc[df['device_name'].str.contains('SAMSUNG', na=False), 'device_name'] = 'Samsung'\n",
    "    df.loc[df['device_name'].str.contains('GT-', na=False), 'device_name'] = 'Samsung'\n",
    "    df.loc[df['device_name'].str.contains('Moto G', na=False), 'device_name'] = 'Motorola'\n",
    "    df.loc[df['device_name'].str.contains('Moto', na=False), 'device_name'] = 'Motorola'\n",
    "    df.loc[df['device_name'].str.contains('moto', na=False), 'device_name'] = 'Motorola'\n",
    "    df.loc[df['device_name'].str.contains('LG-', na=False), 'device_name'] = 'LG'\n",
    "    df.loc[df['device_name'].str.contains('rv:', na=False), 'device_name'] = 'RV'\n",
    "    df.loc[df['device_name'].str.contains('HUAWEI', na=False), 'device_name'] = 'Huawei'\n",
    "    df.loc[df['device_name'].str.contains('ALE-', na=False), 'device_name'] = 'Huawei'\n",
    "    df.loc[df['device_name'].str.contains('-L', na=False), 'device_name'] = 'Huawei'\n",
    "    df.loc[df['device_name'].str.contains('Blade', na=False), 'device_name'] = 'ZTE'\n",
    "    df.loc[df['device_name'].str.contains('BLADE', na=False), 'device_name'] = 'ZTE'\n",
    "    df.loc[df['device_name'].str.contains('Linux', na=False), 'device_name'] = 'Linux'\n",
    "    df.loc[df['device_name'].str.contains('XT', na=False), 'device_name'] = 'Sony'\n",
    "    df.loc[df['device_name'].str.contains('HTC', na=False), 'device_name'] = 'HTC'\n",
    "    df.loc[df['device_name'].str.contains('ASUS', na=False), 'device_name'] = 'Asus'\n",
    "\n",
    "    df.loc[df.device_name.isin(df.device_name.value_counts()[df.device_name.value_counts() < 200].index), 'device_name'] = \"Others\"\n",
    "    df['had_id'] = 1\n",
    "    gc.collect()\n",
    "    \n",
    "    return df\n",
    "\n",
    "train=setDevice(train)\n",
    "test=setDevice(test)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### <span id=\"9\"></span> ** Set Frequency **"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "i_cols = ['card1','card2','card3','card5',\n",
    "          'C1','C2','C3','C4','C5','C6','C7','C8','C9','C10','C11','C12','C13','C14',\n",
    "#           'D1','D2','D3','D4','D5','D6','D7','D8',\n",
    "          'addr1','addr2',\n",
    "          'dist1','dist2',\n",
    "          'P_emaildomain', 'R_emaildomain',\n",
    "          'DeviceInfo','device_name',\n",
    "          'id_30','id_33',\n",
    "          'uid','uid2','uid3',\n",
    "         ]\n",
    "\n",
    "for col in i_cols:\n",
    "    temp_df = pd.concat([train[[col]], test[[col]]])\n",
    "    fq_encode = temp_df[col].value_counts(dropna=False).to_dict()   \n",
    "    train[col+'_fq_enc'] = train[col].map(fq_encode)\n",
    "    test[col+'_fq_enc']  = test[col].map(fq_encode)\n",
    "\n",
    "\n",
    "# for col in ['DT_M','DT_W','DT_D']:\n",
    "#     temp_df = pd.concat([train[[col]], test[[col]]])\n",
    "#     fq_encode = temp_df[col].value_counts().to_dict()\n",
    "            \n",
    "#     train[col+'_total'] = train[col].map(fq_encode)\n",
    "#     test[col+'_total']  = test[col].map(fq_encode)\n",
    "        \n",
    "\n",
    "# periods = ['DT_M','DT_W','DT_D']\n",
    "# i_cols = ['uid']\n",
    "# for period in periods:\n",
    "#     for col in i_cols:\n",
    "#         new_column = col + '_' + period\n",
    "            \n",
    "#         temp_df = pd.concat([train[[col,period]], test[[col,period]]])\n",
    "#         temp_df[new_column] = temp_df[col].astype(str) + '_' + (temp_df[period]).astype(str)\n",
    "#         fq_encode = temp_df[new_column].value_counts().to_dict()\n",
    "            \n",
    "#         train[new_column] = (train[col].astype(str) + '_' + train[period].astype(str)).map(fq_encode)\n",
    "#         test[new_column]  = (test[col].astype(str) + '_' + test[period].astype(str)).map(fq_encode)\n",
    "        \n",
    "#         train[new_column] /= train[period+'_total']\n",
    "#         test[new_column]  /= test[period+'_total']"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## <span id=\"10\"></span> ** 3. Data Preprocessing **"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {
    "_kg_hide-input": false
   },
   "outputs": [],
   "source": [
    "def get_too_many_null_attr(data):\n",
    "    many_null_cols = [col for col in data.columns if data[col].isnull().sum() / data.shape[0] > 0.9]\n",
    "    return many_null_cols\n",
    "\n",
    "def get_too_many_repeated_val(data):\n",
    "    big_top_value_cols = [col for col in train.columns if train[col].value_counts(dropna=False, normalize=True).values[0] > 0.9]\n",
    "    return big_top_value_cols\n",
    "\n",
    "def get_useless_columns(data):\n",
    "    too_many_null = get_too_many_null_attr(data)\n",
    "    print(\"More than 90% null: \" + str(len(too_many_null)))\n",
    "    too_many_repeated = get_too_many_repeated_val(data)\n",
    "    print(\"More than 90% repeated value: \" + str(len(too_many_repeated)))\n",
    "    cols_to_drop = list(set(too_many_null + too_many_repeated))\n",
    "    #cols_to_drop.remove('isFraud')\n",
    "    return cols_to_drop"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "More than 90% null: 12\n",
      "More than 90% repeated value: 70\n"
     ]
    }
   ],
   "source": [
    "cols_to_drop = get_useless_columns(train)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "train = train.drop(cols_to_drop, axis=1)\n",
    "test = test.drop(cols_to_drop, axis=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(590540, 423)\n",
      "(506691, 423)\n",
      "(590540,)\n"
     ]
    }
   ],
   "source": [
    "print(train.shape)\n",
    "print(test.shape)\n",
    "print(y.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [],
   "source": [
    "numerical_cols = train.select_dtypes(exclude = 'object').columns\n",
    "categorical_cols = train.select_dtypes(include = 'object').columns"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Index(['ProductCD', 'card4', 'card6', 'P_emaildomain', 'R_emaildomain'], dtype='object')"
      ]
     },
     "execution_count": 21,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "categorical_cols[:5]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Label Encoding\n",
    "for f in train.columns:\n",
    "    if train[f].dtype.name =='object' or test[f].dtype.name =='object': \n",
    "        lbl = preprocessing.LabelEncoder()\n",
    "        lbl.fit(list(train[f].values) + list(test[f].values))\n",
    "        train[f] = lbl.transform(list(train[f].values))\n",
    "        test[f] = lbl.transform(list(test[f].values))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [],
   "source": [
    "train = train.fillna(-999)\n",
    "test = test.fillna(-999)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0\n",
      "0\n"
     ]
    }
   ],
   "source": [
    "print(train.isnull().sum().max())\n",
    "print(test.isnull().sum().max())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "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>TransactionDT</th>\n",
       "      <th>TransactionAmt</th>\n",
       "      <th>ProductCD</th>\n",
       "      <th>card1</th>\n",
       "      <th>card2</th>\n",
       "      <th>card3</th>\n",
       "      <th>card4</th>\n",
       "      <th>card5</th>\n",
       "      <th>card6</th>\n",
       "      <th>addr1</th>\n",
       "      <th>...</th>\n",
       "      <th>dist1_fq_enc</th>\n",
       "      <th>P_emaildomain_fq_enc</th>\n",
       "      <th>R_emaildomain_fq_enc</th>\n",
       "      <th>DeviceInfo_fq_enc</th>\n",
       "      <th>device_name_fq_enc</th>\n",
       "      <th>id_30_fq_enc</th>\n",
       "      <th>id_33_fq_enc</th>\n",
       "      <th>uid_fq_enc</th>\n",
       "      <th>uid2_fq_enc</th>\n",
       "      <th>uid3_fq_enc</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TransactionID</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>2987000</td>\n",
       "      <td>86400</td>\n",
       "      <td>4.241327</td>\n",
       "      <td>4</td>\n",
       "      <td>13926</td>\n",
       "      <td>-999.0</td>\n",
       "      <td>150.0</td>\n",
       "      <td>1</td>\n",
       "      <td>142.0</td>\n",
       "      <td>1</td>\n",
       "      <td>315.0</td>\n",
       "      <td>...</td>\n",
       "      <td>4707</td>\n",
       "      <td>163648</td>\n",
       "      <td>824070</td>\n",
       "      <td>863508</td>\n",
       "      <td>863508</td>\n",
       "      <td>949007</td>\n",
       "      <td>953271</td>\n",
       "      <td>6</td>\n",
       "      <td>6</td>\n",
       "      <td>2</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>2987001</td>\n",
       "      <td>86401</td>\n",
       "      <td>3.401197</td>\n",
       "      <td>4</td>\n",
       "      <td>2755</td>\n",
       "      <td>404.0</td>\n",
       "      <td>150.0</td>\n",
       "      <td>2</td>\n",
       "      <td>102.0</td>\n",
       "      <td>1</td>\n",
       "      <td>325.0</td>\n",
       "      <td>...</td>\n",
       "      <td>643488</td>\n",
       "      <td>435803</td>\n",
       "      <td>824070</td>\n",
       "      <td>863508</td>\n",
       "      <td>863508</td>\n",
       "      <td>949007</td>\n",
       "      <td>953271</td>\n",
       "      <td>1328</td>\n",
       "      <td>1328</td>\n",
       "      <td>105</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>2987002</td>\n",
       "      <td>86469</td>\n",
       "      <td>4.094345</td>\n",
       "      <td>4</td>\n",
       "      <td>4663</td>\n",
       "      <td>490.0</td>\n",
       "      <td>150.0</td>\n",
       "      <td>4</td>\n",
       "      <td>166.0</td>\n",
       "      <td>2</td>\n",
       "      <td>330.0</td>\n",
       "      <td>...</td>\n",
       "      <td>38</td>\n",
       "      <td>9934</td>\n",
       "      <td>824070</td>\n",
       "      <td>863508</td>\n",
       "      <td>863508</td>\n",
       "      <td>949007</td>\n",
       "      <td>953271</td>\n",
       "      <td>1787</td>\n",
       "      <td>1787</td>\n",
       "      <td>46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>2987003</td>\n",
       "      <td>86499</td>\n",
       "      <td>3.931826</td>\n",
       "      <td>4</td>\n",
       "      <td>18132</td>\n",
       "      <td>567.0</td>\n",
       "      <td>150.0</td>\n",
       "      <td>2</td>\n",
       "      <td>117.0</td>\n",
       "      <td>2</td>\n",
       "      <td>476.0</td>\n",
       "      <td>...</td>\n",
       "      <td>643488</td>\n",
       "      <td>182784</td>\n",
       "      <td>824070</td>\n",
       "      <td>863508</td>\n",
       "      <td>863508</td>\n",
       "      <td>949007</td>\n",
       "      <td>953271</td>\n",
       "      <td>7602</td>\n",
       "      <td>7602</td>\n",
       "      <td>362</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>2987004</td>\n",
       "      <td>86506</td>\n",
       "      <td>3.931826</td>\n",
       "      <td>1</td>\n",
       "      <td>4497</td>\n",
       "      <td>514.0</td>\n",
       "      <td>150.0</td>\n",
       "      <td>2</td>\n",
       "      <td>102.0</td>\n",
       "      <td>1</td>\n",
       "      <td>420.0</td>\n",
       "      <td>...</td>\n",
       "      <td>643488</td>\n",
       "      <td>435803</td>\n",
       "      <td>824070</td>\n",
       "      <td>9</td>\n",
       "      <td>42979</td>\n",
       "      <td>3573</td>\n",
       "      <td>1430</td>\n",
       "      <td>30</td>\n",
       "      <td>30</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>5 rows × 423 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "               TransactionDT  TransactionAmt  ProductCD  card1  card2  card3  \\\n",
       "TransactionID                                                                  \n",
       "2987000                86400        4.241327          4  13926 -999.0  150.0   \n",
       "2987001                86401        3.401197          4   2755  404.0  150.0   \n",
       "2987002                86469        4.094345          4   4663  490.0  150.0   \n",
       "2987003                86499        3.931826          4  18132  567.0  150.0   \n",
       "2987004                86506        3.931826          1   4497  514.0  150.0   \n",
       "\n",
       "               card4  card5  card6  addr1  ...  dist1_fq_enc  \\\n",
       "TransactionID                              ...                 \n",
       "2987000            1  142.0      1  315.0  ...          4707   \n",
       "2987001            2  102.0      1  325.0  ...        643488   \n",
       "2987002            4  166.0      2  330.0  ...            38   \n",
       "2987003            2  117.0      2  476.0  ...        643488   \n",
       "2987004            2  102.0      1  420.0  ...        643488   \n",
       "\n",
       "               P_emaildomain_fq_enc  R_emaildomain_fq_enc  DeviceInfo_fq_enc  \\\n",
       "TransactionID                                                                  \n",
       "2987000                      163648                824070             863508   \n",
       "2987001                      435803                824070             863508   \n",
       "2987002                        9934                824070             863508   \n",
       "2987003                      182784                824070             863508   \n",
       "2987004                      435803                824070                  9   \n",
       "\n",
       "               device_name_fq_enc  id_30_fq_enc  id_33_fq_enc  uid_fq_enc  \\\n",
       "TransactionID                                                               \n",
       "2987000                    863508        949007        953271           6   \n",
       "2987001                    863508        949007        953271        1328   \n",
       "2987002                    863508        949007        953271        1787   \n",
       "2987003                    863508        949007        953271        7602   \n",
       "2987004                     42979          3573          1430          30   \n",
       "\n",
       "               uid2_fq_enc  uid3_fq_enc  \n",
       "TransactionID                            \n",
       "2987000                  6            2  \n",
       "2987001               1328          105  \n",
       "2987002               1787           46  \n",
       "2987003               7602          362  \n",
       "2987004                 30            1  \n",
       "\n",
       "[5 rows x 423 columns]"
      ]
     },
     "execution_count": 25,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [],
   "source": [
    "TransactionDT1 = train['TransactionDT'].apply(lambda x: (START_DATE + datetime.timedelta(seconds = x)))\n",
    "dt_m = TransactionDT1.dt.date.astype('str').str[:7]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [],
   "source": [
    "X = train.drop(['TransactionDT', 'DT'], axis=1)\n",
    "X_test = test.drop(['TransactionDT', 'DT'], axis=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "X: (590540, 421)\n",
      "y_train: (590540,)\n",
      "X_test: (506691, 421)\n"
     ]
    }
   ],
   "source": [
    "print(\"X:\", X.shape)\n",
    "print(\"y_train:\", y.shape)\n",
    "print(\"X_test:\", X_test.shape)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## <span id=\"11\"></span> ** 4. Model **"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### <span id=\"12\"></span> ** LightGBM **"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [],
   "source": [
    "params = {'num_leaves': 546,\n",
    "          'min_child_weight': 0.03454472573214212,\n",
    "          'feature_fraction': 0.1797454081646243,\n",
    "          'bagging_fraction': 0.2181193142567742,\n",
    "          'min_data_in_leaf': 106,\n",
    "          'objective': 'binary',\n",
    "          'max_depth': -1,\n",
    "          'learning_rate': 0.005883242363721497,\n",
    "          \"boosting_type\": \"gbdt\",\n",
    "          \"bagging_seed\": 11,\n",
    "          \"metric\": 'auc',\n",
    "          \"verbosity\": -1,\n",
    "          'reg_alpha': 0.3299927210061127,\n",
    "          'reg_lambda': 0.3885237330340494,\n",
    "          'random_state': 42,\n",
    "}"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Fit predict"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(453219, 421) (453219,) (137321, 421) (137321,)\n",
      "Training until validation scores don't improve for 500 rounds.\n",
      "[200]\ttraining's auc: 0.957642\tvalid_1's auc: 0.887381\n",
      "[400]\ttraining's auc: 0.974969\tvalid_1's auc: 0.899758\n",
      "[600]\ttraining's auc: 0.985691\tvalid_1's auc: 0.906795\n",
      "[800]\ttraining's auc: 0.991855\tvalid_1's auc: 0.910679\n",
      "[1000]\ttraining's auc: 0.995418\tvalid_1's auc: 0.913451\n",
      "[1200]\ttraining's auc: 0.997365\tvalid_1's auc: 0.915105\n",
      "[1400]\ttraining's auc: 0.998433\tvalid_1's auc: 0.915929\n",
      "[1600]\ttraining's auc: 0.99908\tvalid_1's auc: 0.916515\n",
      "[1800]\ttraining's auc: 0.999465\tvalid_1's auc: 0.916592\n",
      "[2000]\ttraining's auc: 0.999692\tvalid_1's auc: 0.916839\n",
      "[2200]\ttraining's auc: 0.999818\tvalid_1's auc: 0.917177\n",
      "[2400]\ttraining's auc: 0.999896\tvalid_1's auc: 0.91711\n",
      "[2600]\ttraining's auc: 0.999941\tvalid_1's auc: 0.917104\n",
      "Early stopping, best iteration is:\n",
      "[2220]\ttraining's auc: 0.999829\tvalid_1's auc: 0.917205\n",
      "Fold 1 | AUC: 0.9172053775734671\n",
      "(497955, 421) (497955,) (92585, 421) (92585,)\n",
      "Training until validation scores don't improve for 500 rounds.\n",
      "[200]\ttraining's auc: 0.954431\tvalid_1's auc: 0.915118\n",
      "[400]\ttraining's auc: 0.971986\tvalid_1's auc: 0.926424\n",
      "[600]\ttraining's auc: 0.983982\tvalid_1's auc: 0.934676\n",
      "[800]\ttraining's auc: 0.991051\tvalid_1's auc: 0.939987\n",
      "[1000]\ttraining's auc: 0.995073\tvalid_1's auc: 0.943028\n",
      "[1200]\ttraining's auc: 0.997233\tvalid_1's auc: 0.944819\n",
      "[1400]\ttraining's auc: 0.998396\tvalid_1's auc: 0.945769\n",
      "[1600]\ttraining's auc: 0.999058\tvalid_1's auc: 0.946599\n",
      "[1800]\ttraining's auc: 0.999447\tvalid_1's auc: 0.946683\n",
      "[2000]\ttraining's auc: 0.999677\tvalid_1's auc: 0.946788\n",
      "[2200]\ttraining's auc: 0.999805\tvalid_1's auc: 0.946762\n",
      "[2400]\ttraining's auc: 0.999886\tvalid_1's auc: 0.946753\n",
      "Early stopping, best iteration is:\n",
      "[1994]\ttraining's auc: 0.999671\tvalid_1's auc: 0.946809\n",
      "Fold 2 | AUC: 0.9468091596609053\n",
      "(504519, 421) (504519,) (86021, 421) (86021,)\n",
      "Training until validation scores don't improve for 500 rounds.\n",
      "[200]\ttraining's auc: 0.954233\tvalid_1's auc: 0.92883\n",
      "[400]\ttraining's auc: 0.972292\tvalid_1's auc: 0.938429\n",
      "[600]\ttraining's auc: 0.983858\tvalid_1's auc: 0.944802\n",
      "[800]\ttraining's auc: 0.990767\tvalid_1's auc: 0.948581\n",
      "[1000]\ttraining's auc: 0.994759\tvalid_1's auc: 0.950272\n",
      "[1200]\ttraining's auc: 0.996981\tvalid_1's auc: 0.950988\n",
      "[1400]\ttraining's auc: 0.998215\tvalid_1's auc: 0.951399\n",
      "[1600]\ttraining's auc: 0.998929\tvalid_1's auc: 0.951416\n",
      "[1800]\ttraining's auc: 0.99937\tvalid_1's auc: 0.951164\n",
      "Early stopping, best iteration is:\n",
      "[1464]\ttraining's auc: 0.998486\tvalid_1's auc: 0.951496\n",
      "Fold 3 | AUC: 0.9514955417366173\n",
      "(488908, 421) (488908,) (101632, 421) (101632,)\n",
      "Training until validation scores don't improve for 500 rounds.\n",
      "[200]\ttraining's auc: 0.956058\tvalid_1's auc: 0.916149\n",
      "[400]\ttraining's auc: 0.973599\tvalid_1's auc: 0.928181\n",
      "[600]\ttraining's auc: 0.985112\tvalid_1's auc: 0.935951\n",
      "[800]\ttraining's auc: 0.991882\tvalid_1's auc: 0.940611\n",
      "[1000]\ttraining's auc: 0.995585\tvalid_1's auc: 0.942921\n",
      "[1200]\ttraining's auc: 0.997558\tvalid_1's auc: 0.94427\n",
      "[1400]\ttraining's auc: 0.998611\tvalid_1's auc: 0.944906\n",
      "[1600]\ttraining's auc: 0.999207\tvalid_1's auc: 0.945114\n",
      "[1800]\ttraining's auc: 0.999544\tvalid_1's auc: 0.945089\n",
      "[2000]\ttraining's auc: 0.999739\tvalid_1's auc: 0.945013\n",
      "[2200]\ttraining's auc: 0.999848\tvalid_1's auc: 0.94481\n",
      "Early stopping, best iteration is:\n",
      "[1703]\ttraining's auc: 0.999404\tvalid_1's auc: 0.945182\n",
      "Fold 4 | AUC: 0.9451823945261882\n",
      "(506885, 421) (506885,) (83655, 421) (83655,)\n",
      "Training until validation scores don't improve for 500 rounds.\n",
      "[200]\ttraining's auc: 0.953259\tvalid_1's auc: 0.920849\n",
      "[400]\ttraining's auc: 0.971359\tvalid_1's auc: 0.936306\n",
      "[600]\ttraining's auc: 0.9833\tvalid_1's auc: 0.945499\n",
      "[800]\ttraining's auc: 0.990393\tvalid_1's auc: 0.951445\n",
      "[1000]\ttraining's auc: 0.994479\tvalid_1's auc: 0.954519\n",
      "[1200]\ttraining's auc: 0.996782\tvalid_1's auc: 0.956188\n",
      "[1400]\ttraining's auc: 0.99807\tvalid_1's auc: 0.957259\n",
      "[1600]\ttraining's auc: 0.998833\tvalid_1's auc: 0.958046\n",
      "[1800]\ttraining's auc: 0.999298\tvalid_1's auc: 0.95842\n",
      "[2000]\ttraining's auc: 0.999583\tvalid_1's auc: 0.958536\n",
      "[2200]\ttraining's auc: 0.999742\tvalid_1's auc: 0.958717\n",
      "[2400]\ttraining's auc: 0.999845\tvalid_1's auc: 0.958643\n",
      "[2600]\ttraining's auc: 0.999906\tvalid_1's auc: 0.958604\n",
      "Early stopping, best iteration is:\n",
      "[2201]\ttraining's auc: 0.999743\tvalid_1's auc: 0.958726\n",
      "Fold 5 | AUC: 0.958725831896644\n",
      "(501214, 421) (501214,) (89326, 421) (89326,)\n",
      "Training until validation scores don't improve for 500 rounds.\n",
      "[200]\ttraining's auc: 0.954408\tvalid_1's auc: 0.910157\n",
      "[400]\ttraining's auc: 0.972302\tvalid_1's auc: 0.923066\n",
      "[600]\ttraining's auc: 0.983945\tvalid_1's auc: 0.931467\n",
      "[800]\ttraining's auc: 0.990807\tvalid_1's auc: 0.935993\n",
      "[1000]\ttraining's auc: 0.994819\tvalid_1's auc: 0.938118\n",
      "[1200]\ttraining's auc: 0.997017\tvalid_1's auc: 0.938858\n",
      "[1400]\ttraining's auc: 0.998237\tvalid_1's auc: 0.939215\n",
      "[1600]\ttraining's auc: 0.998953\tvalid_1's auc: 0.939299\n",
      "[1800]\ttraining's auc: 0.999388\tvalid_1's auc: 0.93928\n",
      "[2000]\ttraining's auc: 0.999644\tvalid_1's auc: 0.939049\n",
      "Early stopping, best iteration is:\n",
      "[1541]\ttraining's auc: 0.998785\tvalid_1's auc: 0.939443\n",
      "Fold 6 | AUC: 0.9394431624856543\n",
      "\n",
      "Mean AUC = 0.9431435779799127\n",
      "CPU times: user 7h 49min 9s, sys: 8min 17s, total: 7h 57min 26s\n",
      "Wall time: 2h 2min 47s\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "\n",
    "columns = X.columns\n",
    "y_preds = np.zeros(X_test.shape[0])\n",
    "score = 0\n",
    "\n",
    "feature_importances = pd.DataFrame()\n",
    "feature_importances['feature'] = columns\n",
    "\n",
    "NFOLDS = len(np.unique(dt_m)) # 6\n",
    "\n",
    "for fold_n, month in enumerate(np.unique(dt_m)):\n",
    "    X_train, y_train = X[~dt_m.isin([month])][columns], y[~dt_m.isin([month])]\n",
    "    X_valid, y_valid = X[dt_m.isin([month])][columns], y[dt_m.isin([month])]\n",
    "    print(X_train.shape, y_train.shape, X_valid.shape, y_valid.shape)\n",
    "    \n",
    "    dtrain = lgb.Dataset(X_train, label=y_train)\n",
    "    dvalid = lgb.Dataset(X_valid, label=y_valid)\n",
    "\n",
    "    clf = lgb.train(params, dtrain, 10000, valid_sets = [dtrain, dvalid], verbose_eval=200, early_stopping_rounds=500)\n",
    "    \n",
    "    feature_importances[f'fold_{fold_n + 1}'] = clf.feature_importance()\n",
    "    \n",
    "    y_pred_valid = clf.predict(X_valid)\n",
    "    save_to_disk(y_pred_valid, 'y_pred_valid_fold{}.pkl'.format(fold_n))\n",
    "    print(f\"Fold {fold_n + 1} | AUC: {roc_auc_score(y_valid, y_pred_valid)}\")\n",
    "    \n",
    "    score += roc_auc_score(y_valid, y_pred_valid) / NFOLDS\n",
    "    \n",
    "    y_pred_test = clf.predict(X_test)\n",
    "    save_to_disk(y_pred_test, 'y_pred_test_fold{}.pkl'.format(fold_n))\n",
    "    y_preds += y_pred_test / NFOLDS\n",
    "    \n",
    "    del X_train, X_valid, y_train, y_valid\n",
    "    gc.collect()\n",
    "    \n",
    "print(f\"\\nMean AUC = {score}\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## <span id=\"13\"></span> ** 5. Submission **"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [],
   "source": [
    "submission = pd.read_csv('/kaggle/input/ieee-fraud-detection/sample_submission.csv')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [],
   "source": [
    "submission['isFraud'] = y_preds"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "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>TransactionID</th>\n",
       "      <th>isFraud</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>0</td>\n",
       "      <td>3663549</td>\n",
       "      <td>0.000496</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>3663550</td>\n",
       "      <td>0.001555</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>2</td>\n",
       "      <td>3663551</td>\n",
       "      <td>0.001414</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>3</td>\n",
       "      <td>3663552</td>\n",
       "      <td>0.001150</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>4</td>\n",
       "      <td>3663553</td>\n",
       "      <td>0.000959</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   TransactionID   isFraud\n",
       "0        3663549  0.000496\n",
       "1        3663550  0.001555\n",
       "2        3663551  0.001414\n",
       "3        3663552  0.001150\n",
       "4        3663553  0.000959"
      ]
     },
     "execution_count": 33,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "submission.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [],
   "source": [
    "submission.to_csv('submission.csv', index=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAABDUAAAOjCAYAAACr6D8xAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADl0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uIDMuMC4zLCBodHRwOi8vbWF0cGxvdGxpYi5vcmcvnQurowAAIABJREFUeJzs3XmYXVWV9/HvTxEIAsEwKQhEQUGZIkZAQUTEgW5sB1CgUZzoiK823S2KtNqICIrazhONM+KAoCiDyqAogoCEIQlgFBWVSSRMEgnzev84u+RS1piQVG7y/TxPPXXuOXtY59yKcletvStVhSRJkiRJUr95xEQHIEmSJEmStDBMakiSJEmSpL5kUkOSJEmSJPUlkxqSJEmSJKkvmdSQJEmSJEl9yaSGJEmSJEnqSyY1JEnSiJK8KcmNSeYnWXOi4xmPJO9M8oWJjkOLLsm6Sc5JckeSj4zSduck145w/StJjnj4o5QkLWkmNSRJWgKS/DTJXS0xMD/Jrwdd/9ckf0zytyTfSzJliDE27Ok/P0m19gOvn93aPSvJT9qHv9uTnJLkqT3j7JzkgdbnjiS/TvK6YeJ+FPBR4AVVtWpV3bwIz2Bqi3mFhR1jvKrq/VW1/5KabyRJDkty3ETHsTRJskqSzyaZ135Wzxmh+QxgHrB6VR20hEKUJC3lTGpIkrTkvKUlBlatqk0HTibZHPg/4NXAusCdwGcHd66qP/X0X7Wd3rrn3M+TPBM4A/g+sB7wBGAWcF6SJ/YMd30bY3XgHcDnexMfPdYFVgauWMR7X2Tp9OV/uyzJRM7SaIT7PwaYAjylff+vEYbZCLiyquphDm+psrz/rEjSePXlfxhIkrSM2Rc4parOqar5wP8AL0+y2kKM9SHg2Kr6RFXdUVW3VNW7gQuAwwY3rs73gFuBhyQ1kjwZGKgouS3JT9r5zZKcmeSWVuXxyp4+/5zk0iR/TXJNkt45z+kZa36SZw6uXhhczdEqXI5Mch5dsueJSSYn+WKSG5Jcl+SIJI8c6mH0jt8z9utabLcmOSDJM5LMTnJbkk/39H1tkvOSfKpVEcxN8rye6+slObk9h98m+bdB856Y5LgkfwUOAN4J7NXufVZr97okv2oVM79P8saeMXZOcm2Sg5L8pd3v63quT0rykVbhc3uSc5NMate2T/KLdk+zkuw81PNpbZ/SnvNtSa5I8i89Y/y599kmeVmS2e34EUkOSfK7JDcn+XZahVHPs35Dkj8BPxli3k2BfwFmVNVNVXV/VV08TIxfAV4DHNye365JVkry8STXt6+PJ1lpmP5PS3JJe87H0yXqBq6tleTUdv+3JPn5cMmzJJ9oPzt/TXJxHqyOWi/JgvRUWLU556WrdiLJ69t7fWuS05Ns1NO2krw5yVXAVSPN1a5NSvLVNtavkhycnuU2LZ7vJLkpydVJDhzqfiRpWWBSQ5KkJecD7UPOeYM+ZG5OV00BQFX9DrgHePJ4Bk+yCvAs4IQhLn8beP4QfR6R5GXAGsCc3mtV9ZsWG8AaVbVLkkcDZwLfANYB9gE+m67aBOBvwH5tvH8G3pTkpe3aTj1jrVpV54/x1l5Nt/RgNeCPwFeB+4BNgKcBLwDGs8RkO+BJwF7Ax4F3Abu2e31lkucMavt7YC3gPcB3ez64fhO4lq4iZk/g/b1JD+AlwIl0z+KLwPuB49u9b93a/AXYna5i5nXAx5Js0zPGY4HJwPrAG4DPJHlMu/a/wNPp3vMpwMHAA0nWB04Djmjn3wZ8J8nagx9E+8B9Cl11zzrAvwNfT7JpVV1A937u0tPlX+nee4ADgZcCz2nP4FbgM4OmeA5dFcYLB89N92z/CLy3/buYk2SPIdpRVa8Fvg58qD2/s+jet+2BacDWwLbAu4e4xxWB7wFfa8/jBKB3noPo3se16SqT3gkMVw1yUZtvSnsOJyRZuaquB84fNO6/AidW1b3t38A7gZe3eX5O9/PT66XtmQwkF4ecq117DzAVeCLdv+tX9dzvI+je01l0PzfPA/4zyVDvgST1PZMakiQtGe+g+wCyPl3J/SlJNm7XVgVuH9T+droP8eMxhe7/228Y4toNdB/MB6yX5Da6PQreA7y6qn49RL/Bdgf+UFVfrqr7quoS4Dt0H+qpqp9W1ZyqeqCqZtN9cHvOCOONxVeq6oqquo/uHncD/rOq/lZVfwE+Buw9jvHeV1V3VdUZdB/av1lVf6mq6+g+bD6tp+1fgI9X1b1VdTxd5co/J9kA2BF4RxvrMuALdAmYAedX1ffas1gwVCBVdVpV/a5VzPyMLrnw7J4m9wKHt/l/AMwHNm0fXF8P/EdVXdeqHH5RVXfTfcD9QVX9oM19JjAT+KchQtie7ufvqKq6p6p+ApxKl6yC7v3bByBd5dA/8eCH8TcC76qqa9u8hwF75qHLJw5r79NQ9/94YAu6n/X1gLcAX03ylKGe1RD2bc/mL1V1E/BeHvr8e+/xUTz4Pp5IlzAYcC/wOGCjdv3nwy1xqarjqurm9rP/EWAlYGAp2Td48FmF7mdyIAH0RuADVfWr9nP8fmBab7VGu37LwLMaZa5XAu+vqlur6lrgkz3jPANYu6oOb+/p74HPM75/I5LUN0xqSJK0BFTVhW05yN1V9VXgPB78kDmf7jf1vVYH7hjnNLcCD9B9QBvscXQJjAHXV9UaVTWlqqZV1bfGOMdGwHatVP+2lhjZl66igCTbJTm7lb3fTrfsYq0RxhuLawbN/yjghp75/4+uymCsbuw5XjDE61V7Xl836APuH+k+gK8H3FJVdwy6tv4wcQ8pyW5JLmjLHm6j+5nofV43tw/BA+5s8a1Ft4Tid0MMuxHwikHv0Y4M/XOxHnBNVT0wzH18g24p1Ep0VQaXVNUfe+Y5qWeOXwH301U7jOUZLKBLKBzRPnz/DDibrvJmLNZrsfbGvd4w7YZ6Hwd8GPgtcEa6JUCHDDdhuqVAv0q33Oc2uiqagffrROCZSdajq0oquiQZdM/qEz3P6hYgjPDzMspc6w1qP/jfyHqD3v938tD3RZKWGSY1JEmaGEX3oQa6TTgHliOQbkPPlYDfjGvAqr/RlcC/YojLrwR+vFCRPtQ1wM9aQmTga9WqelO7/g3gZGCDqpoMHM2D9znUb7//BqzS8/qxQ7Tp7XcNcDewVs/8q1fV5kP0ezis337rPmBD4Pr2NSUP3fdkQ+C6YeL+h9ctUfAdumUk61bVGsAPePB5jWQecBew8RDXrgG+Nug9enRVHTVE2+uBDQbtIfH3+6iqK+kSALvx0KUnA/PsNmielVvFy5D3PMjs0W5yFNfTfYDvjfv6IdrdwNDvIwAt2XhQVT0ReDHw1kHLiABoe1q8g+7f0mPa+3U77f2qqtvoKm1eSfesvtmTSLkGeOOgZzWpqn7RM0WNda52T4/v6btBz/E1wNWD5lqtqoaq1JGkvmdSQ5KkxSzJGklemGTlJCsk2ZfuN7mntyZfB16c5Nltz4rDge8OqgIYq0OA1yQ5MMlqSR6T5AjgmXTl+YvqVODJSV6d5FHt6xk9SwZWo6tguCvJtnQf7gbcRFdJ0vtXWC4Ddkr352onA/890uRVdQPdB8ePJFm97Qmy8aB9MB5O6wAHtvt8Bd3+ED+oqmuAX9Dtk7Jykq3o9rz4+ghj3QhM7UkgrEiXvLoJuC/JboyxSqFVVnwJ+GjbFPKR6TZeXQk4ju7n6YXt/MrpNh19/BBDXUiXWDq43ePOdB/seyt3vkG3f8ZOPHS/lqOBIweWUCRZO8lLxhJ/cw7wJ+C/27+LHYCdefDfxWi+Cby7zbsWcCjdvQ92Pt0eLAe2eV5Ot/8GLe7dk2zSkh5/pas2uX+IcVZr49wErJDkUP6xwuobdHvK7MFDE0BHt/vcvM05uf08DWe0ub7dxntMuj1U3tJz7ZfAX5O8I92Goo9MskWSZ4wwnyT1LZMakiQtfo+i27TxJrrfsP878NJqe1hU1RV0yzS+TreHw2rA/1uYiarqXLpNGV9O99vcP9LtEbFjVV21aLfR/Vab7oP33nS/Ff8z8EG6D+fQxX14kjvoPmR+u6fvncCRdH9e9rYk27f9Ho6n+639xXRJk9HsR5cQuJJuyc2JDL204uFwId2movPoYt+zqm5u1/ah26zxeuAk4D3tfoYzkBC4Ockl7VkeSPeMbqVLAJ08jtjeRre560V0yxk+CDyiJVxeQrfk4Ca639y/nSH+u6+q7qH7CyS7tXv8LLBfVc3tafZNumTDT6qqdwnTJ1q8Z7T3+wK6jS7HpKrubXH+E10VwueHmHskR9DtFTKb7jlc0s4Nnuceun8Pr6V7znsB3+1p8iTgLLplYOcDn62qnw4x3+nAD+kqqP5IVykzeHnNyW28G6uqd/Pfk+jen2+l+2s4l9M98+GMNtfhdJubXt1iP5Gugomqup8uMTWtXZ9Ht9/L5BHmk6S+lWH2QZIkSVquJXktsH9V7TjRsUgjSfImYO+qWlwVS5K01LJSQ5IkSeojSR6XZIe2/GpTuj9Le9JExyVJE2GF0ZtIkiRJWoqsSPdXf54A3Ea3B8pnJzQiSZogLj+RJEmSJEl9yeUnkiRJkiSpL5nUkCRJkiRJfck9NdSX1lprrZo6depEhyFJkiRJWgwuvvjieVW19mjtTGqoL02dOpWZM2dOdBiSJEmSpMUgyR/H0s6khvrSfTfdwk2fO26iw5AkSZKkvrH2m1410SE87NxTQ5IkSZIk9SWTGpIkSZIkqS+Z1JAkSZIkSX2pr5MaSQ5PsusQ53dOcmo7fkmS2UkuSzIzyY7DjLVla3NZkluSXN2Oz1rc9zFWSV6f5LE9r7+cZNNFHPPyJF9biH5TkhywEP0ekeSQEa4fkeQ/xzuuJEmSJGn509dJjao6tKpGSzr8GNi6qqYBrwe+MMxYc6pqWmt3MvD29vohSZMkE7m56uuBvyc1qup1VfXrhR0syVbAfcAuSSaNs/sUYNxJDbqfuWGTGpIkSZIkjVVfJDWSTE1yec/rtyU5LMlXkuzZzr0oydwk5wIvH2hbVfOrqtrLRwPFOCXZNclZSb4FXNrOnZLk4iRXJNm/nVshyW1JjkoyK8n5SdZp1/ZuVRGzkpzdzm2c5OdJLm1jbdcz5zuTzGntj0yyFzANOL5VkKyY5Nwk01r7V7X2lyd5/2jxNPsAxwI/AXbvmfvcJB9tsV2ZZHqSk5JcleSw1uwoYNMWy1HDPLf121iXtbie1fqt1s4d29odmuTXSc4EnjTe90eSJEmStHxaJv6ka5KVgc8DuwC/BY4fdP1lwAeAdYB/XshptgeeWlV/aq9fU1W3JFkFmJnkO8AdwGTgZ1V1SJKP0lVXHAW8B9i5qm5MskYb4wbg+VV1V5LNgK8C2yV5MbAbsG1VLUgypc3178Bbquqydl8D9/d44AhgOnA7cFaS3YEfjRAPwCuBnYC5wP7ACT33u6Cqnp3kIOB7wNPb2L9P8nG6aotNWmXLcF4FnFJVH0zySGAS8Etg/4F+SbYF9qBL2KwIXAacP9RgSWYAMwAeP2XNEaaVJEmSJC0P+qJSYww2A66uqqtaVcZxvRer6qSq2gx4KfC+hZzj/J6EBsB/JZlF9wH88cDG7fyCqvphO74YmNqOzwOObVUdA899JeCLrQrlW8BT2/ldgS9V1YIW/y2jxLYd8JOqmldV9wLfoEtWDBtPkmcC11bVdcCZdMmUyT1jnty+zwHmVNWNVXUX8Id2v2NxEbB/kvcAW1TV/CHa7AR8p6oWVNXtwCnDDVZVx1TV9Kqavuaqq48xBEmSJEnSsqpfkhr38dBYVx6izajLSqrqHGDjJGstRAx/Gzhom5PuBGxfVVsDs3tiuqenz/08WA3zb3TVGlOBWUkeAxwEXANsCWxLl+QAyFjup0dGuDZcPPsAWyT5A3AVsDrwsp62d7fvD/QcD7weU4VPVf0E2JmuIuXrSfYdrulYxpMkSZIkqVe/JDVuBNZJsmaSlejZ/6GZCzwhyUC1xD4DF5JskrZOI8k2dEscbl7EeCYDt7SlIZsDzxhDnydW1QXA/wC3Auu3cW5o1SWv4cHkxBnAGwY270wypZ2/A1htiLEvAJ7bns8KwN7Az4YLpC0F2YNuOc3UqppKtw/JPsP1GcJwsfTOsxHw56o6BvgK8LSquq9dG0iMnAO8PMnKSVbnH99bSZIkSZKG1Bd7alTVvUkOBy4ErqZLYvRev6vtt3BaknnAucAW7fIewH5J7gUWAHv1bBy6sE4DZrTlJ3NbXKP5WJIn0CUuzqiqy5N8GjgxyT7AWbSKiKo6NcnWdHt13Eu3JON/gC8DX0iygK6yY+D+r01yKPDTNv4pVXVahv9LLc+lW65zY8+5s4Hjkqw7lgfQ9gaZmWQOcFpVDfUXTZ4HvLXdw3y6PTYAvgjMTjKzqvZLchIwi25pyzljmV+SJEmSpCz653tpyZu20RPrzEMOn+gwJEmSJKlvrP2mV43eaCmR5OKqmj5au35ZfiJJkiRJkvQQfbH85OGWZEvga4NO311V201EPP0uyTS6PTN63VlVz1pcc66w9pS+yjJKkiRJkh5+y2VSo6rmANMmOo5lRVVdhs9TkiRJkrSEufxEkiRJkiT1peWyUkP9776bbuKmo4+Z6DAkSZIkLQfWPmDGRIegYVipIUmSJEmS+pJJDUmSJEmS1JdMakiSJEmSpL5kUmMZkeTwJLsOcX7nJKe245ckmZ3ksiQzk+w4ypgfTnJFkg8vrrglSZIkSVpYbhS6jKiqQ8fQ7MfAyVVVSbYCvg1sNkL7NwJrV9XdD0eMkiRJkiQ9nKzU6DNJpia5vOf125IcluQrSfZs516UZG6Sc4GXD7StqvlVVe3lo4FiGElObm0uTLJXkickOT/JRUnel2T+KHG+vbWdneS9PbH/KsnnWwXIGUkmtWubJDkryawklyTZeCEfkSRJkiRpOWFSYxmTZGXg88CLgWcDjx10/WVJ5gKnAa8fbpyq+hdgQVVNq6rjgU8An6uqZwB/HiWGFwBPArYFpgFPT7JTu/wk4DNVtTlwG7BHO//1dn5r4FnADUOMO6Mtm5l58/wRcyqSJEmSpOWASY1lz2bA1VV1VavKOK73YlWdVFWbAS8F3jeOcXcAvtmOvzZK2xe0r0uBS1pMT2rXrq6qy9rxxcDUJKsB61fVSS3Gu6rqzsGDVtUxVTW9qqavueqq4whdkiRJkrQsck+N/nMfD01GrTxEm2GXlfy9QdU5STZOslZVzRvj3KOO2wT4QFX930NOJlOB3v057gcmtfaSJEmSJI2LlRr950ZgnSRrJlkJ2H3Q9bnAE3r2pNhn4ELbtyLteBtgReDmMc57HrB3O953lLanA69Psmqba/0k6wzXuKr+Clyb5KWt/UpJVhljXJIkSZKk5ZRJjT5TVfcChwMXAqfSJTF6r98FzABOaxuF/rHn8h7A5UkuAz4D7NWzceho/gN4c5KLgMmjxHgG8A3g/CRzgBOB1UYZ/9XAgUlmA79g0F4gkiRJkiQNlrF/ppUelGR+VU3YxhbTNtqozvzvd03U9JIkSZKWI2sfMGOiQ1juJLm4qqaP1s5KDUmSJEmS1JfcKHQ5l2RL/vGvmdxdVduN1K+qVl3Yvg+HFdZe22ypJEmSJC3nTGos56pqDjBtSfeVJEmSJGlRufxEkiRJkiT1JZMakiRJkiSpL7n8RH3p3ptu5MbPfWSiw5AkSZI0QdZ900ETHYKWAlZqSJIkSZKkvmRSQ5IkSZIk9SWTGpIkSZIkqS8tVUmNJIcn2XWI8zsnObUdvyTJ7CSXJZmZZMdhxtqytbksyS1Jrm7HZy3u+xirJK9P8tie119Osukijnl5kq8tRL8pSQ5YlLklSZIkSVqSlqqNQqvq0DE0+zFwclVVkq2AbwObDTHWHGAaQJKvAKdW1YmD2yVZoaruW6TAF97rgUuAPwNU1esWZbD2PO4DdkkyqaoWjKP7FOAA4OhFiUGSJEmSpCVlQio1kkxNcnnP67clOSzJV5Ls2c69KMncJOcCLx9oW1Xzq6ray0cDxTgl2TXJWUm+BVzazp2S5OIkVyTZv51bIcltSY5KMivJ+UnWadf2blURs5Kc3c5tnOTnSS5tY23XM+c7k8xp7Y9Mshdd0uX4VkGyYpJzkwwkYl7V2l+e5P2jxdPsAxwL/ATYvWfuc5N8tMV2ZZLpSU5KclWSw1qzo4BNWyxHjfDczk5yYut7RJL9klzUqmemtnbrJvluq6T5ZZLt2/ntW8yXJjkvyZPa+f3bmKe3cT8w3vdUkiRJkrT8WaqWnwxIsjLweeDFwLOBxw66/rIkc4HT6KodFsb2wMFVtWV7/ZqqejrwDOCtSR7Tzk8GflZVWwPn98z3HuB57fzL2rkbgOdX1dOAfYFPtnhfDOwGbNvaf6SqjgcuA/aqqmlVdU/P/T0eOAJ4LvA0YIckA0mK4eIBeCVwPPBNugRHrwVV9Wzgi8D36KoytgRmJFkDOAT4dYvlkBGe29bAm1vf/YGpVfUM4KvAW1qbTwIfqqrpLaYvtPO/AnZsz+d97R57x90T2Ap4VZL1Bk+cZEZLlMy8Zf7fRghRkiRJkrQ8WCqTGnTLSa6uqqtaVcZxvRer6qSq2gx4Kd2H44VxflX9qef1fyWZRZcoeDywcTu/oKp+2I4vBqa24/OAY1tVx8BzXAn4YqtC+Rbw1HZ+V+BLA8tBquqWUWLbDvhJVc2rqnuBbwA7jRRPkmcC11bVdcCZwHZJJveMeXL7PgeYU1U3VtVdwB/a/Y7VhT19fw+c3jPu1J77PTrJZXQJlMckmQSsAXy3PZ//BTbvGfesqrqjPaO5wIaDJ66qY6pqelVNn7Lqo8cRsiRJkiRpWTRRSY37Bs298hBtRl1WUlXnABsnWWshYvj7r/rTbU66E7B9q4CY3RPTPT197ufBfUj+ja5aYyowq1V2HARcQ1fFsC1dkgMgY7mfHhnh2nDx7ANskeQPwFXA6jxYQQJwd/v+QM/xwOvx7K0yuG/vuAPjhK4qZVr7Wr8lK44ETq+qLegSUr3ve++4vfclSZIkSdKQJiqpcSOwTpI1k6xEz/4PzVzgCUkGqiX+vpQiySZJ0o63AVYEbl7EeCYDt1TVgiSb0y1BGc0Tq+oC4H+AW4H12zg3tOqS1/BgcuIM4A2tWoEkU9r5O4DVhhj7AuC57fmsAOwN/Gy4QJI8EtgDeGpVTa2qqXT7kAxegjKS4WJZGGfRLVEZiG9aO5wMXNeOX/swzSVJkiRJWk5NSFKjLak4HLgQOJUuidF7/S5gBnBa2yj0jz2X9wAub0sbPkO3J8W4Nwsd5DRglbb85NAW12g+lmQO3bKLs6rqcuDTwP5JLgA2olUfVNWpwI+AmS3u/2pjfBn4wsBGoQMDV9W1LY6f0u27cUFVnTZCLM+lW65zY8+5s4FpSdYdw73Q+s5sm5MOuVHoOLyZbh+Q2UmupKtqAfgg8OEk5y3i+JIkSZIkkUXPB0hL3tYbbVBnHPKfEx2GJEmSpAmy7psOmugQtBglubj98YkRuW+B+tKj1l7X/xGTJEmSpOXcMpHUSLIl8LVBp++uqu0mIp5+1/bA+Mqg03dW1bMmIBxJkiRJkoa0TCQ1qmoOMG3UhhqTqroMn6ckSZIkaSk3UX/9RJIkSZIkaZEsE5UaWv7ce9N13PDZd050GJIkSdJS6XH/7/0THYK0RFipIUmSJEmS+pJJDUmSJEmS1JdMamhCJPlpkunt+Mgk1ySZP9FxSZIkSZL6h0kNLXZJRtu75RRg2yURiyRJkiRp2eFGoRqXJPsBbwMKmA18G3g3sCJwM7BvVd2Y5DBgPWAqMC/JG4AvA08FfgVMGhizqi5oYy+x+5AkSZIk9T+TGhqzJJsD7wJ2qKp5SabQJTe2r6pKsj9wMHBQ6/J0YMeqWpDkrcCdVbVVkq2ASybiHiRJkiRJyw6TGhqPXYATq2oeQFXdkmRL4Pgkj6Or1ri6p/3JVbWgHe8EfLL1m51k9ngnTzIDmAGw/pTVF/4uJEmSJEnLBPfU0HiErjKj16eAT1fVlsAbgZV7rv1tUNvBfcelqo6pqulVNX3NVVdZlKEkSZIkScsAkxoajx8Dr0yyJkBbfjIZuK5df80Ifc8B9m39tgC2WoxxSpIkSZKWAyY1NGZVdQVwJPCzJLOAjwKHASck+Tkwb4TunwNWbctODgZ+OXAhyYeSXAuskuTatsmoJEmSJEkjck8NjUtVfRX46qDT3x+i3WGDXi8A9h5mzIPpEh2SJEmSJI2ZlRqSJEmSJKkvmdSQJEmSJEl9yeUn6kuPWnt9Hvf/3j/RYUiSJEmSJpCVGpIkSZIkqS+Z1JAkSZIkSX3J5SfqS/f85Wqu+dS+Ex2GJEmSNOE2+PevT3QI0oSxUkOSJEmSJPUlkxqSJEmSJKkvmdRYzJKsmeSy9vXnJNf1vF5xKYjv5Uk263l9ZJLnLuKYpyX5+UL0e0SSQxZlbkmSJEnS8sM9NRazqroZmAaQ5DBgflX9b2+bJAFSVQ8s+Qh5OfAAMBegqt61KIMlWRPYErgryYZV9adxdH8EcAhw1KLEIEmSJElaPlipMUGSbJLk8iRHA5cAj0tyTJKZSa5IcmhP22uTHJbk0iSzkzy5nd8lyaxW9XFJkkcnWT3JT9rr2Ul27xnnde3crCRfTvJs4J+Aj7UxpiY5LslLW/vnt/Nzknx+oLJkuHiaPYHvAccDe/XMfVySzyQ5O8nvkuyU5KtJ5ib5Ymt2FLBam/PYxfHcJUmSJEnLDpMaE+upwBer6mlVdR1wSFVNB7YGnp/kqT1tb6yqpwFfAN7azr0dmFFV04CdgLuABcBLqmobYFfgYwBJtgbeAexcVVsDB1XVz4EfAP9VVdOq6g8DkyVZBfgSsEdVbQmsAswYJR6AfYBvtq99Bt3v5Kp6LnAwcArwwfYMnp5kC7oqjTtaLPuN/TFKkiRJkpZHJjUm1u+q6qKe1/skuYSucuMpdB/4B3y3fb8YmNqOzwM+nuTfgdWr6n4gwAeTzAbOADZIshawC3B8Vd0CMPB9BE8Brqqq37XXx9IlToaNJ8n6wIbABVV1JfDI3v066BIZAHOA66vqyrbk5sqeexpWkhmtkmWj9BwtAAAgAElEQVTmLfPvGq25JEmSJGkZZ1JjYv1t4CDJk4D/AHapqq2AHwEr97S9u32/n7YXSlUdAbwRWBW4qI2xHzAZ2KZVcMxr4wSoccSWUa7/Qzx0y03WBK5O8ge6BMfeQ/R5oOd44PWo+7tU1TFVNb2qpk9ZdeXRmkuSJEmSlnEmNZYeqwN3AH9N8jjghaN1SLJxVc2uqg8AlwKb0iU0/lJV9yV5PrB+a34WsHeSKa3vlHb+DmC1IYa/EnhSkie2168CfjZKSPsAu1bV1KqaCmzLPy5BGVZV3ddicwNbSZIkSdKoTGosPS6hSyRcDnyebmnJaN7WNhudDdxGt9zka8CzkswEXgFcBVBVs4EPAeckuQz4cBvjm8A7BzYKHRi4qu4E3gB8N8kcusqKzw8XSJKNgccCM3vGuAq4O8nTx3AvA74IzHajUEmSJEnSaFI1nhUJ0tJhqw3XrNPe/qKJDkOSJEmacBv8+9cnOgTpYZfk4vaHNEZkpYYkSZIkSepLJjUkSZIkSVJfckNG9aUV13mCZXaSJEmStJyzUkOSJEmSJPUlkxqSJEmSJKkvmdSQJEmSJEl9yT011Jfu+stvmfuZl0x0GJIkSRIAm735+xMdgrRcslJDkiRJkiT1JZMakiRJkiSpL5nUkCRJkiRJfcmkxjIiyeFJdh3i/M5JTm3H+yaZ3b5+kWTrUcY8MMmvknx9ccUtSZIkSdLCcqPQZURVHTqGZlcDz6mqW5PsBhwDbDdC+/8H7FZVVz8cMUqSJEmS9HCyUqPPJJma5PKe129LcliSryTZs517UZK5Sc4FXj7Qtqp+UVW3tpcXAI8fYZ6jgScCJyf5ryRrJjkjyaVJ/i/JH5OsNUL/VyX5ZZLLWvtHtvPzkxyZZFaSC5Ks286vm+Skdn5WkmctwmOSJEmSJC0HTGosY5KsDHweeDHwbOCxwzR9A/DD4capqgOA64HnVtXHgPcA51bV04CTgQ1HiOEpwF7ADlU1Dbgf2LddfjRwQVVtDZwD/Fs7/0ngZ+38NsAVQ4w7I8nMJDNvnX/PcNNLkiRJkpYTLj9Z9mwGXF1VVwEkOQ6Y0dsgyXPpkho7jmPcnWhVH1V1WpJbR2j7PODpwEVJACYBf2nX7gFObccXA89vx7sA+7Xx7wduHzxoVR1Dt2SGLTZco8YRuyRJkiRpGWRSo//cx0MrbFYeos2wH/iTbAV8gW6vjJvHOfdYEwkBvlpV/z3EtXuramCc+/FnUJIkSZK0kFx+0n9uBNZpe1ysBOw+6Ppc4AlJNm6v9xm4kGRD4LvAq6vqN+Oc9xzaEpK2yehjRmj7Y2DPJOu09lOSbDTK+D8G3tTaPzLJ6uOMT5IkSZK0nDGp0Weq6l7gcOBCumUccwddv4tuuclpbaPQP/ZcPhRYE/hs28Bz5jimfi+wU5JLgBcAfxohxiuBdwNnJJkNnAk8bpTx/wN4bpI5dMtSNh9HbJIkSZKk5VAeXAkgjV2SPwDTq2reRMy/xYZr1InveM5ETC1JkiT9g83e/P2JDkFapiS5uKqmj9bOSg1JkiRJktSX3KRxOZdkTbr9LAZ73kgbiVbV1Lavx2Xj7ftwWHmdTcyGS5IkSdJyzqTGcq4lH6Yt6b6SJEmSJC0ql59IkiRJkqS+ZKWG+tKdN/2WS45+8USHIUmSpGXYNgecMtEhSBqFlRqSJEmSJKkvmdSQJEmSJEl9yaSGJkSSw5PsOsT5nZOcOhExSZIkSZL6i3tqaEJU1aETHYMkSZIkqb+Z1NBilWQqcGpVbdFevw1YFRg4f2KSFwEfB+YBl0xMpJIkSZKkfuPyE02oJCsDnwdeDDwbeOzERiRJkiRJ6hcmNTTRNgOurqqrqqqA44ZrmGRGkplJZt46/54lF6EkSZIkaalkUkOL23089Ods5SHa1FgGqqpjqmp6VU1/zKorPizBSZIkSZL6l0kNLW43AuskWTPJSsDug67PBZ6QZOP2ep8lGp0kSZIkqW+5UagWq6q6N8nhwIXA1XRJjN7rdyWZAZyWZB5wLrDFko9UkiRJktRvTGposauqTwKfHOH6j+j21pAkSZIkacxcfiJJkiRJkvqSSQ1JkiRJktSXXH6ivrTK2puwzQGnTHQYkiRJkqQJZKWGJEmSJEnqSyY1JEmSJElSXzKpIUmSJEmS+pJ7aqgvzb/pt5x3zO4THYYkSZImyA4zTp3oECQtBazUkCRJkiRJfcmkhiRJkiRJ6ksmNSRJkiRJUl8yqbGMS/LTJNPb8ZFJrkkyfwz91k5yYZJLkzx78UcqSZIkSdL4mNRYhiQZbePXU4Btxzjc84C5VfW0qvr5okUmSZIkSdLDz79+spRKsh/wNqCA2cC3gXcDKwI3A/tW1Y1JDgPWA6YC85K8Afgy8FTgV8CkgTGr6oI29mhzTwM+BExKchnwTGBv4L+BG4DfAHdX1VuG6b82cDSwYTv1n1V1Xot1Q+CJ7fvHq+qTQ91vVb16DI9JkiRJkrQcM6mxFEqyOfAuYIeqmpdkCt2H/e2rqpLsDxwMHNS6PB3YsaoWJHkrcGdVbZVkK+CS8c5fVZclORSYXlVvSfI44L1tntuBs4FLRxjiE8DHqurcJBsCpwNPadc2A54LrAb8OsnngCcPcb9DPZcZwAyAdadMGqqJJEmSJGk5YlJj6bQLcGJVzQOoqluSbAkc3xIMKwJX97Q/uaoWtOOdgE+2frOTzH4Y4tkO+GlV3QSQ5Hi6RMRwdgWe2lMRsnqS1drxaVV1N3B3kr8A6zLE/Q41aFUdAxwDsNlGa9Si3ZIkSZIkqd+Z1Fg6ha4yo9engI9W1clJdgYO67n2t0FtF8cH/vGM+QjgmT2JFuDvy17u7jl1P93P4FD3K0mSJEnSiNwodOn0Y+CVSdYEaMsxJgPXteuvGaHvOcC+rd8WwFYPQzwXAjsnWTPJo4BXjNL+DODv+220PTpGMtT9SpIkSZI0IpMaS6GqugI4EvhZklnAR+kqM05I8nNg3gjdPwes2padHAz8cuBCkg8luRZYJcm1bePOscRzQ5v/fOAsRt+n40BgepLZSa4EDhhl/KHuV5IkSZKkEaXKqn+NT5LX0jYRnagYNttojfriu3acqOklSZI0wXaYcepEhyBpMUpycVVNH62de2qoL6269ib+H5kkSZIkLedMaiznkryLf9wj44SqOnK4PlX1FeArC9NXkiRJkqSHi8tP1JemT59eM2fOnOgwJEmSJEmLwViXn7hRqCRJkiRJ6ksuP1Ff+uu8qzjrC/800WFIkiQtk3bd/wcTHYIkjYmVGpIkSZIkqS+Z1JAkSZIkSX3JpIYkSZIkSepLfZHUSPLTJNPb8ZFJrkkyf5Q+70pyWfu6v+f4wCUT9ciSTElyQM/rDZIcv4hjPiNJJXneQvTdJcn2izK/JEmSJElL0lKX1Egy2ualpwDbjjZOVR1ZVdOqahqwYOC4qj45zvkWlynA35MaVXVNVe21iGPuA5zbvo/XLoBJDUmSJElS31isSY0k+yWZnWRWkq8leXGSC5NcmuSsJOu2doclOSbJGcCxSSYl+VbrezwwaWDMqrqgqm5YxLiOS/KRJGcD70+yfZLzW1znJXlSa7d/khOTnJ7kqiQfaOdXaPczJ8nlA9UfSQ5IclG73xOSTGrnH5vk+z3PYjvgKGDTVj1yVJJNklzW2k9K8tU2/iVJdhopnnbtEcAewGuA3ZKs2M5v0mL8UpIrkhyb5IVJfpHkN0mmJ9kY2B94e4vnWSM8t88kOTvJ75Ls1OKcm+SLPe12a8/zkiTHJ3l0O//e9nwuT3J0krTz57Zn8Mskvx5ufkmSJEmSei22KoUkmwPvAnaoqnlJpgAFbF9VlWR/4GDgoNbl6cCOVbUgyVuBO6tqqyRbAZcshhA3Bp5XVQ8kmdzmvj/Ji4AjgIGqia2BbYD7gN8k+RSwAbBWVW3Z7nWN1vaEqjq6nTsKeC3wOeAzwJlV9elWGbIKcAiwSaskIckmPbEdCNxTVVu25/iDgUTLUPFU1fXATsDcqvp9kvOAFwEntz6bAq8E5tI9y7ur6llJ9gAOqao9k3wBmFdVHx/luU2uque2vqcAzxwYN8kWwF/avT2vqu5M8i7gP4D3A5+oqve0ZMY3Wow/bOOmqrZN8i/Aoe3aQySZAcwAWGfKyqOEKUmSJEla1i3OpRe7ACdW1TyAqrolyZbA8UkeB6wIXN3T/uSqWtCOdwI+2frNTjJ7McR3QlU90I7XoKsQ2XiIdmdV1R0ASeYCGwJX0VVZfAL4AXBGa7tVksPbeKsBp7bzOwN7t/u5D/hrknVGiG1H4MOt/RVJrgcGkh5DxXM93ZKTb7U232qvB5Iav62qK1ufK4Gz2vk5wH+PEMdQTunpe/2gcae2OJ8K/KIVYqxItyQG4HlJ3g6sDKwFXMyDSY3vtu8Xt3H+QVUdAxwD8OSpk2uccUuSJEmSljGLM6kRusqMXp8CPlpVJyfZGTis59rfBrVd3B9ae+c7Eji9qj7bKiZ+1HPt7p7j+4EVqurmVkGyG11VxR50FQTHArtV1eWtEqV3j4rx3E9GuPYP8SR5FPAy4J+SvIduWdEaA8s+BvV5oOf1A4z/Z6C37+BxV2ix/6iqXt3bKckqwKeBbarquiRH0CU3Bo97/0LEJEmSJElaDi3OPTV+DLwyyZrQ/bUPYDJwXbv+mhH6ngPs2/ptAWy1GONkUFyvHa1xkrXplkucALyHbjkIwKOBP7ckw7/2dDmbtilokkcmWR24g66aYyi99/8U4HHAb0cI6QXARVW1QVVNraoN6Soq/mW0e+kxUjzj8QvgOUmeCJDk0W3pzCS6xMe8JKvRJYIkSZIkSVpoiy2pUVVX0FVA/CzJLOCjdJUZJyT5OTBvhO6fA1Zty04OBn45cCHJh5JcC6yS5Nokhz0M4X4Q+HDbi2IsNgDOaRt7fh54Zzt/aIv1TODKnvZvAV6YZA4wE9isqm4EZrbNQI8aNP6ngEmt/deB/arqnhHi2Qc4adC57/DQxMpovk+XhLp0UTbqbPf1BrplRrPokhxPrqqbga8Cl7dYL1zYOSRJkiRJgq7aYKJjkMbtyVMn12ffvcNEhyFJkrRM2nX/H0x0CJKWc0kurqrpo7VbrH/SVZIkSZIkaXFZJjZkbH829BWDTp9QVUdORDz9LsmhwMsHnf5WVQ1eJjNhVl/rSf4GQZIkSZKWcy4/UV+aPn16zZw5c6LDkCRJkiQtBi4/kSRJkiRJy7RlYvmJlj+3z7uKU7+020SHIUmStFTb/fU/nOgQJGmxslJDkiRJkiT1JZMakiRJkiSpL5nU0MMmyWuTfHqYa/NH6PelJH9Jcvnii06SJEmStKwxqaEJk+SR7fArwIsmMBRJkiRJUh8yqaExS/K9JBcnuSLJjHbudUl+k+RnwA49bZ+Q5PwkFyV5X8/5nZOcneQbwByAqjoHuGUJ344kSZIkqc/51080Hq+vqluSTAIuSnIa8F7g6cDtwNnApa3tJ4DPVdWxSd48aJxtgS2q6uolFbgkSZIkadljpYbG48Aks4ALgA2AVwM/raqbquoe4PietjsA32zHXxs0zi8XJqGRZEaSmUlm3j7/noUIX5IkSZK0LDGpoTFJsjOwK/DMqtqariJjLlAjdBvu2t8WJoaqOqaqplfV9MmrrrgwQ0iSJEmSliEmNTRWk4Fbq+rOJJsB2wOTgJ2TrJnkUcAretqfB+zdjvddsqFKkiRJkpYHJjU0Vj8CVkgyG3gf3RKUG4DDgPOBs4BLetr/B/DmJBfRJUSGleSbbYxNk1yb5A0Pf/iSJEmSpGWNG4VqTKrqbmC3IS79FPjyEO2vBp7Zc+qodv6nrU9v230epjAlSZIkScsRKzUkSZIkSVJfMqkhSZIkSZL6kstP1Jcmr/Ukdn/9Dyc6DEmSJEnSBLJSQ5IkSZIk9SWTGpIkSZIkqS+Z1JAkSZIkSX3JPTXUl26ddxUnfvlFEx2GJEnShNvzdT+a6BAkacJYqSFJkiRJkvqSSQ1JkiRJktSXTGpIkiRJkqS+tEwmNZL8NMn0dnxkkmuSzB+lz7uSXNa+7u85PnDJRD2yJFOSHNDzeoMkxy/imM9IUkmetxB9d0my/UL02ybJsJthJLk2yRrjHVeSJEmStPzp+6RGktE2Oz0F2Ha0carqyKqaVlXTgAUDx1X1yXHOt7hMAf6e1Kiqa6pqr0Uccx/g3PZ9vHYBxp3UALYB3OFTkiRJkrTIlqqkRpL9ksxOMivJ15K8OMmFSS5NclaSdVu7w5Ick+QM4Ngkk5J8q/U9Hpg0MGZVXVBVNyxiXMcl+UiSs4H3J9k+yfktrvOSPKm12z/JiUlOT3JVkg+08yu0+5mT5PKB6o8kByS5qN3vCUkmtfOPTfL9nmexHXAUsGmrHjkqySZJLmvtJyX5ahv/kiQ7jRRPu/YIYA/gNcBuSVZs5zdpMX4pyRVJjk3ywiS/SPKbJNOTbAzsD7y9xfOsYZ7b3m2sWUnObvd3KLBv67dnkrWTnNni/hyQRXmvJEmSJEnLj6XmT7om2Rx4F7BDVc1LMgUoYPuqqiT7AwcDB7UuTwd2rKoFSd4K3FlVWyXZCrhkMYS4MfC8qnogyeQ29/1tKcURwEDVxNZ01Qj3Ab9J8ilgA2Ctqtqy3evA8ooTqurodu4o4LXA54DPAGdW1adbZcgqwCHAJq2ShCSb9MR2IHBPVW3ZnuMPBhItQ8VTVdcDOwFzq+r3Sc6jq544ufXZFHglMJfuWd5dVc9KsgdwSFXtmeQLwLyq+vgIz+w9wM5VdWOSNdp7dTiwRVX9Z7uPzwJnV9X7k7yEnmqUwZLMAGYArLXmyiNMK0mSJElaHixNlRq7ACdW1TyAqroFeDxwepI5wNuBzXvan1xVC9rxTsBxrd9sYPZiiO+EqnqgHa8BfDfJ5cD/DorrrKq6o8U2F9gQ+C1dlcUnkrwQuL213SrJz9v97d0zzs7A/7X7ua+q/jpKbDsCX2vtrwCuBwaSHkPFA92Sk2+142/x0CUov62qK9v9Xgmc1c7PAaaOEkuv8+gqafZn+J+13vfu+8Adww1WVcdU1fSqmr76qiuOIwxJkiRJ0rJoaUpqhK4yo9engE+3Coc3Ar2/nv/boLaD+z7ceuc7Eji9qrYAXjoorrt7ju8HVqiqm4Gt6PavOJCWsACOBd7U7u+IQeOM535GWrLxD/EkeRTwMuDwJH8APg78c5JHD9HngZ7XDzC+6p5/o6vWmArMSvKYYdot7vdOkiRJkrQMWpqSGj8GXplkTej+2gcwGbiuXX/NCH3PAfZt/bagSyAsTr1xvXa0xknWBlJVJ9B9yN+mXXo08OeWZPjXni5n05ZhJHlkktXpKhhWG2aK3vt/CvA4uuqQ4bwAuKiqNqiqqVW1Id2Gqv8y2r30GCmeAU+sqguA/wFuBdYfol9v7C8ew5iSpP/P3r1H21WV9/9/fwC5CQQJAlahgUBF5M6pUBFERAErBQEViqIWfqn9adUqWr6ljZFCBbWKiNUGEUVaQRAViQKCIBcFCSEJl0ajYr+KFhoBRbkZeL5/7Hl0czyXnCQnJ/uc92uMM85ac80517NWMkbGfvLMuSVJkgSsRkmNtmziVOBbSRYAHwZmARcluR5YMszwTwAbJFlIZ9+N7/ZfSPKBJD8F1k/n60JnrYRwTwc+2PaiWBZbAte1jT3PBv6htc9ssX6DzjKPfm8FDmzLUuYC21fVvcDcthnoaQPm/xiwXuv/H8CxVfX4MPEcDXxpQNsXeWpiZSRfoZOEum2ojUKBj7SYbqezDOYO4JvALm3ckXSSPAckmUdn2c09Q8wlSZIkSdJTpMrKf/We6dOm1Onv/bPxDkOSJGncHfmmy8c7BEla6ZLcWlV9I/VbbSo1JEmSJEmSRmO1+UrXVSnJScCrBzRfVFWnjkc8vS7JTODwAc0XVNXAZTIrzTM23c7/lZAkSZKkSc7lJ+pJfX19NXfu3PEOQ5IkSZI0Blx+IkmSJEmSJrRJufxEve/+Xyzm/M8cON5hSJIkjanXvfGK8Q5BklZrVmpIkiRJkqSeZFJDkiRJkiT1JJMakiRJkiSpJ61WSY0kJyc5YJD2/ZJc1o6PSbKw/Xw7yS5DzDU1yfz28z9J7uk6X3usn2VZJDk8yfZd56cmeckKzjknyfXLMW6NJCeuyL0lSZIkSVqVVquNQqtq5jJ0uxt4cVU9kORgYDaw5yBz/QLYFSDJLODXVfWhgf2ShM5X2z65IrEvp8OBJ4FFAFV10opMlmQqsBPwaJKtqur/jmL4GsCJwGkrEoMkSZIkSavKuFRqJJmW5I6u8xOSzErymSRHtraDkixKcgOdD/8AVNW3q+qBdnoT8JzluP+2Se5I8klgHvCsJLOTzE1yZ5KZXX1/2mK7rVWH/Elr3z/Jglb5MS/J05NslOSb7Xxhkld2zfOm1rYgyblJ9gFeAXykzTEtyflJDmv9X9bab09ydn91yVDxNEcCXwYuBF7bde/zk3w8yTVJfphk3ySfbe/3nNbtNGDDds/zRnhvn27v6bwkB7aKme8n6Wv9Nmh/lt9tcR7S2qcnub613Zpkz9Z+QJKrk1yS5HtD3V+SJEmSpG6r1fKTfknWBc4GDgH2AbYYoutxwNeX8zY7AOdU1W5VdQ9wYlX1AbsAL0uyQ1ffe6tqN+BTwDtb27uBGVW1K7Av8CjwCHBoVe0OHAB8pD3PLsDfA/tV1S7Au6rqeuBrwN9V1a5V9eOu518f+DRwRFXtBKwPzBghHoCjgc+3n6MHPO+UqnoJ8B7gq8Dp7R3skWRHOlUaD7VYjh3mvT0X+BCdipCdgSOr6oXA/2lzAMwELq+qFwD7A//a/kx/DrysxX4McGbXvLsDb2kxPS/JXgNvnGRGSzzN/dVDjw8ToiRJkiRpMlgtkxrA9sDdVbW4qgo4f2CHtvfEcXSSBcvjh1V1S9f50Unm0anceB6dD9f9Lmm/bwWmteMbgTOS/C2wUVU9AQQ4PclC4EpgyySb0vlgf2FV3Q/Q/3sYzwMWV9UP2/l5dBInQ8aT5NnAVsBNVXUXsGb3fh10EhkAtwM/q6q72pKbu7qeaVn8YMDYq7rm7Z/n5cBJSeYD1wDrttjWAc5pVToX8NR3fFNV/by9x/mDxVRVs6uqr6r6NtpwtdgWRZIkSZI0jsZrT42lPDWhsu4gfWqowUl2plOlcHDbO2N5/KZrvu2AtwMvqKoHk5w/IKbH2u8naO+sqk5Jcinw58AtSfYDXgxMAXavqqVJftrmyXDPM4iMcP0P4qGz3GQqcHdnmxCmAEcBswaMebLruP98NH8PBo7tnrd/ngCHdSVlOo3JKcBPgNcBTwN+PcS83c8lSZIkSdKgxqtS415gs3S+oWQd4JUDri8Ctk4yvZ3/bilFkq3oVCq8vqq+v5Li2Qh4CPhVkmcBB440IMn0qlpYVe8HbqOzLGMKcF9LaLwMeHbrfhVwVJJN2thNWvtDwIaDTH8XsF2Sbdr564BvjRDS0cABVTWtqqYBL+APl6AMqaqWtthWRjLhCuBt/SdJdmuHU4Cft+qbNzBy8kaSJEmSpCGNS1Kjqn4LnAzcDFxG+/aPruuP0tlDYk7bKPS/uy7PpFOR8G9tU8u5KyGkeXQSCXfQ2cvjxmUYc0LbNHMh8CCd5SafA17YYno1sLg9z0LgA8B1bUnGB9scnwf+oX+j0P6Jq+phOktrLklyO50qhrOHCqQlf7YAfvcuqmox8FiSPZbhWfqdAyxcCRt1vg9Yv21yeie/rxY5Czg+yU3AH/PU6gxJkiRJkkYlnf80l3rLNltPqZPf+wd7iUqSJE0or3vjFeMdgiSNiyS3ti/zGNbqulGoJEmSJEnSsCbEZoxJpgJXD3LppSuwkeiklWQzOstpBtqvqh5c1fEMZpOp2/k/F5IkSZI0yU2IpEZLXOw63nFMFFV1H75PSZIkSdJqzuUnkiRJkiSpJ5nUkCRJkiRJPWlCLD/R5LPkF4s557wDxzsMSZKkMXPcse4fJkkjsVJDkiRJkiT1JJMakiRJkiSpJ5nUkCRJkiRJPcmkxgSX5OQkBwzSvl+Sy4YZt06Sq5LMT/LasY1SkiRJkqTRc6PQCa6qZi7n0N2Ap1XVriszHkmSJEmSVhYrNSaIJNOS3NF1fkKSWUk+k+TI1nZQkkVJbgAOH2auzYDzgV1bpcb07rFJzhyhyuPpST6d5JYktyU5tLW/McklSS5PsjjJB7rGHJRkXpIFSa4eYt4ZSeYmmfvQQ4+P+h1JkiRJkiYWKzUmiSTrAmcD+wM/AC4cqm9V3ZfkeOCEqnplG3vtsoxtTgK+WVV/lWRj4LtJrmrXdqVTBfIY8L0kHwMebbHtW1V3J9lkiLhmA7MBpm09pZbhsSVJkiRJE5iVGpPH9sDdVbW4qopOJcZYjX05cGKS+XSSIesCW7VrV1fVL6vqUeAu4I+BvYDrqupugKq6fxSxSZIkSZImKSs1Jo6lPDVJte4gfVakumE0YwMcUVXfe0pjsiedCo1+T9D5O5gVjE2SJEmSNAlZqTFx3AtslmRqknWAVw64vgjYOsn0dn70KOYe7dgrgL9NEoAku43Q/zvAi5Ns3foPuvxEkiRJkqRuJjUmiKr6LXAycDNwGZ1ERPf1R4EZwJy2Ueh/j2Lu0Y79Z+BpwMK2eek/jzD//7b5L0mygJH37JAkSZIkiXS2SJCWXZL9aJuIjlcM07aeUv/0vr3G6/aSJElj7rhjrxjvECRp3CS5tar6RurnnhrqSZtO3c5/6CVJkiRpkjOpMckleRPw9gHNN1bVW4YaU1XXAtcuz1hJkiRJklYWkxqTXFWdC5y7qsdKkiRJkrSi3ChUkiRJkiT1JCs11JPuu38xHz//wPEOQ5Ikabm85XXuDSZJK4OVGpIkSZIkqSeZ1JAkSZIkST3JpIYkSZIkSepJJjV6UJI3JjlriGu/HjM1BgEAACAASURBVGbcp5Pcl+SOZbjH9knmJ7ktyfQViVeSJEmSpLFgUmMSSLJmO/wMcNAyDjsM+EpV7VZVPxyTwCRJkiRJWgEmNVZDSb6c5NYkdyaZ0drelOT7Sb4F7N3Vd+sk30lyS5J/7mrfL8k1Sf4TuB2gqq4D7l+G+78CeAdwfJJrWttJSb6X5Kokn09ywjDjpye5vD3D9Um2b+2fSXJmkm8n+VGSI7vGvCfJ7UkWJDltlK9MkiRJkjQJ+ZWuq6e/qqr7k6wH3JJkDvA+YA/gl8A1wG2t70eBT1TVeUneMmCeFwA7VtXdo7l5VX0tySeBX1fVh5LsARwF7Ebn78w84NZhppgNvLmqFifZE/g3YP927VnAi4DtgUuBi5McTKcyZM+qejjJJoNN2hI8MwCeMXXd0TySJEmSJGkCMqmxenpbkle14y2B1wPXVtX/AiS5EPiTdn1v4Ih2/Dng9K55vjvahMYQ9gG+VFUPt/tfOlTHJBsALwQuStLfvE5Xly9X1ZPAXUk2b20HAOf2z19Vg1aTVNVsOgkTttpmSi3/40iSJEmSJgKTGquZJPvR+ZD/Z61q4VpgEfC8YYYN9QH/NysxtGVNIqwBPFhVuw5x/bGu43T9NkkhSZIkSRoV99RY/UwBHmgJje2BvYD1gP2STE3yNODVXf1vpLM0BOCYMYrpOuBVSdZLsiFwyFAdq+pXwN1JXg2Qjl1GmP9K4K+SrN/GDLr8RJIkSZKkbiY1Vj+XA2slWQj8M3AT8HNgFvAd4Co6e1r0ezvwliS30EmIDCnJ59scz03y0yTHLUtAVTUPuBCYD3wRuH6EIccAxyVZANwJHDrC/JfT2V9jbpL5wJCbkEqSJEmS1C9VVv1rdJLMom0iOl4xbLXNlPr7k/car9tLkiStkLe87orxDkGSVmtJbq2qvpH6WakhSZIkSZJ6kpUak1ySj9P5BpVuH62qc8dy7Irq6+uruXPnjvVtJEmSJEnjYFkrNfz2k0muqt4yHmMlSZIkSVpRLj+RJEmSJEk9yUoN9aR771/Mhz5/4HiHIUmSBMAJR7vxpySNBys1JEmSJElSTzKpIUmSJEmSepJJDY2pJCcnOWCQ9v2SXNaOD02yMMn8JHOTvGjVRypJkiRJ6jXuqaExVVUzl6Hb1cClVVVJdga+AGw/tpFJkiRJknqdSQ2tFEmmAZdV1Y7t/ARgA6C//eIkBwFnAEuAef1jq+rXXVM9HahVE7UkSZIkqZe5/ESrRJJ1gbOBQ4B9gC0GXH9VkkXAHOCvVn2EkiRJkqReY1JDq8r2wN1VtbiqCji/+2JVfamqtgcOA/55sAmSzGh7bsz99UOPj33EkiRJkqTVmkkNrSxLeerfp3UH6TPispKqug6YnmTTQa7Nrqq+qurbYMO1lz9SSZIkSdKEYFJDK8u9wGZJpiZZB3jlgOuLgK2TTG/nR/dfSLJtkrTj3YG1gV+sgpglSZIkST3MjUK1UlTVb5OcDNwM3E0nidF9/dEkM4A5SZYANwA7tstHAMcm+S3wCPDatkRFkiRJkqQhmdTQSlNVZwJnDnP9cgb5qtaqOh04fQxDkyRJkiRNQC4/kSRJkiRJPcmkhiRJkiRJ6kkuP1FP2nyT7Tjh6CvGOwxJkiRJ0jiyUkOSJEmSJPUkkxqSJEmSJKknmdSQJEmSJEk9yT011JN+9sBiZn3hwPEOQ5IkTVCzXuPeXZLUC6zUkCRJkiRJPcmkhiRJkiRJ6kkmNSRJkiRJUk/q6aRGkpOTHDBI+35JLmvHxyRZ2H6+nWSXIeaammR++/mfJPd0na891s+yLJIcnmT7rvNTk7xkBeeck+T65Ri3RpITl/Oe70yy7hDXjk9yxvLMK0mSJEmaXHp6o9CqmrkM3e4GXlxVDyQ5GJgN7DnIXL8AdgVIMgv4dVV9aGC/JAFSVU+uSOzL6XDgSWARQFWdtCKTJZkK7AQ8mmSrqvq/oxi+BnAicNpy3PqdwKeBR5djrCRJkiRJQI9UaiSZluSOrvMTksxK8pkkR7a2g5IsSnIDnQ//AFTVt6vqgXZ6E/Cc5bj/tknuSPJJYB7wrCSzk8xNcmeSmV19f9piu61Vh/xJa98/yYJW+TEvydOTbJTkm+18YZJXds3zpta2IMm5SfYBXgF8pM0xLcn5SQ5r/V/W2m9PcnZ/dclQ8TRHAl8GLgRe23Xv85N8PMk1SX6YZN8kn23v95zW7TRgw3bP84Z4bxsm+Xp7hjuSHJnk74DNgOuTXNX6HZ/k+0muBfYa7Z+PJEmSJGly6omkxkjaUoazgUOAfYAthuh6HPD15bzNDsA5VbVbVd0DnFhVfcAuwMuS7NDV996q2g34FJ2qBIB3AzOqaldgXzpVCo8Ah1bV7sABwEfa8+wC/D2wX1XtAryrqq4Hvgb8XVXtWlU/7nr+9elUPhxRVTsB6wMzRogH4Gjg8+3n6AHPO6WqXgK8B/gqcHp7B3sk2ZFOlcZDLZZjh3hnrwB+XFW7VNWOwDeq6iPAfcA+VXVAkucA/wT8GfByYMch5iLJjJZImvvwrx4fqpskSZIkaZKYEEkNYHvg7qpaXFUFnD+wQ9t74jg6yYLl8cOquqXr/Ogk8+hUbjyPzgf+fpe037cC09rxjcAZSf4W2KiqngACnJ5kIXAlsGWSTYH9gQur6n6A/t/DeB6wuKp+2M7Po5M4GTKeJM8GtgJuqqq7gDW79+ugk8gAuB34WVXd1Zbc3NX1TCNZCByU5LQke1fVLwfpsxdwdVX9oqoeB74w1GRVNbuq+qqqb/2NVottTiRJkiRJ46hXkhpLeWqsg20yWUMNTrIznSqFQ9veGcvjN13zbQe8Hdi/qnYGLh8Q02Pt9xO0fUuq6hTgr4ENgFvaHMcCU4DdWwXHkjZPhnueQWSE638QD53lJlOBu5P8mE6C46hBxjzZddx/vkx7sVTVfwF9wJ3AB5P8w1Bdl2U+SZIkSZK69UpS415gs/YNJesArxxwfRGwdZLp7fx3SymSbEWnUuH1VfX9lRTPRsBDwK+SPAs4cKQBSaZX1cKqej9wG/BcOgmN+6pqaZKXAc9u3a8CjkqySRu7SWt/CNhwkOnvArZLsk07fx3wrRFCOho4oKqmVdU04AX84RKUIVXV0hbbkAmOVg3y66r6HPBhYPdBnuMm4KVJNmn7gBy5rDFIkiRJkia3nvj2k6r6bZKTgZvpfJvJogHXH00yA5iTZAlwA7/fm2EmnYqEf+t8cQlL214YK2IenUTCHcCP6CwtGckJbbPPJ+ksy7gS+C7w1SRz25yL2/MsTPIB4LokS+ksGzmOzt4X/57kXcBh/RNX1cNJjgMuSbImnfd09lCBtOTPFsDcrjkWJ3ksyR7L+A4AzgEWJpk7xL4auwCnJXkSeBx4c2ufDVyV5CdtX41T6CQ3ftYdkyRJkiRJw0lnCwqpt/zR9Ck14/1+UYokSRobs15zxXiHIEmTWpJbl6UgoVeWn0iSJEmSJD1FTyw/WdmSTAWuHuTSS1dgI9FJK8lmdJbTDLRfVT04Fvf8o2ds5/+gSJIkSdIkNymTGi1xset4xzFRVNV9+D4lSZIkSauYy08kSZIkSVJPmpSVGup9P3lgMe/44kHjHYYkSZqAzjji8vEOQZK0jKzUkCRJkiRJPcmkhiRJkiRJ6kkmNTQuklybpC/J+knmJFmU5M4kp413bJIkSZKk3mBSQ2MuyUh7t3yoqrYHdgP2TnLwKghLkiRJktTj3ChUo5LkWOAEoICFwBeAfwTWBn4BHFNV9yaZBfwRMA1YkuQ44FxgB+C/gPUAquph4Jp2/HiSecBzVuEjSZIkSZJ6lEkNLbMkzwdOAvauqiVJNqGT3NirqirJ8cB7gHe1IXsAL6qqR5K8E3i4qnZOsjMwb5D5NwYOAT66Kp5HkiRJktTbTGpoNPYHLq6qJQBVdX+SnYALkzyLTrXG3V39L62qR9rxvsCZbdzCJAu7J25LVD4PnFlVPxrs5klmADMANtx03ZX3VJIkSZKknuSeGhqN0KnM6PYx4Kyq2gn4a6A72/CbAX0Hju02G1hcVWcM1aGqZldVX1X1rbfR2qMIW5IkSZI0EZnU0GhcDbwmyVSAtvxkCnBPu/6GYcZeBxzTxu0I7Nx/IckpbZ53jEHMkiRJkqQJyqSGlllV3QmcCnwryQLgw8As4KIk1wNLhhn+CWCDtuzkPcB3AZI8h84+HTsA85LMb3tzSJIkSZI0LPfU0KhU1WeBzw5o/sog/WYNOH8EOGqIabNSgpMkSZIkTSpWakiSJEmSpJ5kUkOSJEmSJPUkl5+oJ235jO0444jLxzsMSZIkSdI4slJDkiRJkiT1JJMakiRJkiSpJ7n8RD3pRw8u5jVfOWi8w5AkST3qC4e6jFWSJgIrNSRJkiRJUk8yqSFJkiRJknqSSQ1JkiRJktSTTGpoTCU5OckBg7Tvl+SydnxMkoXt59tJdln1kUqSJEmSeo0bhWpMVdXMZeh2N/DiqnogycHAbGDPsY1MkiRJktTrTGpopUgyDbisqnZs5ycAGwD97RcnOQg4A1gCzOsfW1Xf7prqJuA5qyZqSZIkSVIvc/mJVokk6wJnA4cA+wBbDNH1OODrQ8wxI8ncJHMf+9XjYxOoJEmSJKlnmNTQqrI9cHdVLa6qAs4f2CHJS+gkNf5+sAmqanZV9VVV3zobrT220UqSJEmSVnsuP9HKspSnJsnWHaRPDTU4yc7Ap4CDq+oXKzk2SZIkSdIEZKWGVpZ7gc2STE2yDvDKAdcXAVsnmd7Oj+6/kGQr4BLg9VX1/VUSrSRJkiSp51mpoZWiqn6b5GTgZjrfZrJowPVHk8wA5iRZAtwA7NguzwSmAv+WBGBpVfWtsuAlSZIkST3JpIZWmqo6EzhzmOuX09lbY2D78cDxYxiaJEmSJGkCcvmJJEmSJEnqSVZqqCdts/F2fOHQy8c7DEmSJEnSOLJSQ5IkSZIk9SSTGpIkSZIkqSeZ1JAkSZIkST3JPTXUkxY/+GMO/spx4x2GJEnqAV8/9JzxDkGSNEas1JAkSZIkST3JpIYkSZIkSepJJjUkSZIkSVJPmrBJjSQnJzlgkPb9klw2xJg3JZnffh5Pcns7Pm3sIx5ZkjWSnNh1vmaS61dwzs2TLE0y6g0qkuye5KDlGLdNkqOGuX5Dkl1HO68kSZIkaXKZsEmNqppZVVeNcsy5VbVrVe0K/Ax4STs/sbtfkvHaYHUN4HexVNUTVbXPCs75WuA7wNHLMXZ3YNRJDWAbYMikhiRJkiRJy6LnkxpJpiW5o+v8hCSzknwmyZGt7aAki5LcABy+nPc5Jcm/J/kGcG6S6UmuT3JbkluT7Nn6HZDk6iSXJPlekvO65vhgkruSLExyems7NMnNbZ4rk2zW2jdM8tlWLbIwyWHAacCGrXrkvCRrJXmw9V8jyYeT3NHGHDlSPM3RwDuAbZJs0casleTBFu+8JFck2TPJt5L8KMkrkqwHzASOafEcOcR72z/JgtZnXpKnt+d4SWt7W5L1k1zUnvMCYN3l+TOSJEmSJE0uE/4rXZOsC5wN7A/8ALhwBabbDdi3qh5Nsj7wsna8PfBZYM/Wb3dgB+A+4KYkewF3A68Anl9VlWTj1vc64NLW9mbgXcDfA7OA/62qnZIE2Bi4DDi+VZIMrBh5dbvnLsAzgVuSXDdUPFV1U5JpwDOq6tYkFwOvAc5sY6YAV1bVu5N8tcXz0jb/v1dVX5KTgR2r6h3DvLN3AzOq6uYkGwCP0qk2eWtVHdae4z3AA1W1c5LdgLmDTZRkBjADYN1nPn2YW0qSJEmSJoOer9RYBtsDd1fV4qoq4PwVmOsrVfVoO14HOKdViVxAJ2nQ76aq+nlVPQHMB6YB9wNPAmcneRXwm9Z3K+DKJLcD7wSe39oPAD4OUB0PjBDbi4D/bEtS/ge4AegbJh7oVGn0J3ku4KlLUB6pqm+049uBa6tqaTuexrK7ETgjyd8CG7UYBtqX9udSVbcBdw42UVXNrqq+qupbeyOLOSRJkiRpspsISY2lPPU5Bvu0WyvpXr/pOn4X8BNgJ+AFdJIc/R7rOn4CWKuqfksnyfBl4AhgTrv+ceAjVbUT8P/z+/gzyrgzzLU/iKcdHw0cn+THwCXAHkm2btce7xrzZNccTzKKCp+qOgX4a2ADOtUj2w3VdVnnlCRJkiQJJkZS415gsyRTk6wDvHLA9UXA1kmmt/Pl2RBzMFOAn7fqjzcwfFKBJBvSqVS4DPg7OktZ+ue5py0xeUPXkCuBt7axSfKMVikx1Eal1wFHpfONKJsDezPEMo42xw7AmlX17KqaVlXTgA8yug08HwI2HK5DkulVtbCq3g/cBjx3kHHXAce0/rvw+2oVSZIkSZKG1PNJjVYBcTJwM509JxYNuP4onX0Y5rSNQv97Jd36LDpVDjcBf8xTqyEGM6XFsAD4Jp2lJtDZq+JLwLfoJGj6vQ/YvC1vmQ/0f8vJOcDCQTb8vJjOsy8ArgLeWVX3DRPPX7b7dvtia19W3wR2aZucDrpRKHBC27x0IfAgnWTNbcCabQPRt9F5l1Nbn79jmGSMJEmSJEn90ik0kHrLlG03rRf+66HjHYYkSeoBXz/0nPEOQZI0Skluraq+kfr1fKWGJEmSJEmanCb8V7oOJcmbgLcPaL6xqt4yHvH0uiTH0/YA6XJdVb1tLO633cbT/F8XSZIkSZrkJm1So6rOBc4d7zgmiqr6FPCp8Y5DkiRJkjR5uPxEkiRJkiT1pElbqaHetvjBn/KKL//9eIchSZJWU1877PTxDkGStApYqSFJkiRJknqSSQ1JkiRJktSTTGpIkiRJkqSeZFJjgktybZK+JOsnmZNkUZI7k5w2wrhnJrk5yW1J9llV8UqSJEmStKxMakwgSUba+PVDVbU9sBuwd5KDh+n7UmBRVe1WVdevtCAlSZIkSVpJ/PaT1VSSY4ETgAIWAl8A/hFYG/gFcExV3ZtkFvBHwDRgSZLjgHOBHYD/AtYDqKqHgWva8eNJ5gHPGeLeuwIfANZLMh/4M+Ao4P8APwe+DzxWVW8dYvwzgU8CW7Wmd1TVjS3WrYBt2u8zqurMwZ63ql4/ujcmSZIkSZpsTGqshpI8HzgJ2LuqliTZhM6H/b2qqpIcD7wHeFcbsgfwoqp6JMk7gYerauckOwPzBpl/Y+AQ4KOD3b+q5ieZCfRV1VuTPAt4X7vPL+kkR24b5hE+Cnykqm5IshVwBfC8dm174CXAhsD3knwC+JNBnnew9zIDmAGw7jM3Gub2kiRJkqTJwKTG6ml/4OKqWgJQVfcn2Qm4sCUY1gbu7up/aVU90o73Bc5s4xYmWdg9cVui8nngzKr60TLGsydwbVX9b5vjQjqJiKEcAOyQpP98oyQbtuM5VfUY8FiS+4DNB3vewSatqtnAbIAp225Ryxi7JEmSJGmCMqmxegqdyoxuHwM+XFWXJtkPmNV17TcD+g73gX82sLiqzhhlTKNJIqwB/FlXogWAluR4rKvpCTp/Bwd7XkmSJEmShuVGoaunq4HXJJkK0JZjTAHuadffMMzY64Bj2rgdgZ37LyQ5pc3zjlHGczOwX5KpSZ4GvHqE/lcCv9tvo+3RMZzBnleSJEmSpGGZ1FgNVdWdwKnAt5IsAD5MpzLjoiTXA0uGGf4JYIO27OQ9wHcBkjyHzr4VOwDzksxve3MsSzw/b/f/DnAVg+zTMcDbgL4kC5PcBbx5hPkHe15JkiRJkoaVKqv+NTpJ3kjbRHS8Ypiy7Ra194eGK1iRJEmT2dcOO328Q5AkrYAkt1ZV30j9rNSQJEmSJEk9yUqNSS7JSfzhHhkXVdWpYzl2RfX19dXcuXPH+jaSJEmSpHGwrJUaJjXUk0xqSJIkSdLE5fITSZIkSZI0oZnUkCRJkiRJPWmt8Q5AWh6LH/w5r/jSKeMdhiRJWo187VX/ON4hSJJWMSs1JEmSJElSTzKpIUmSJEmSepJJDUmSJEmS1JN6IqmR5OQkBwzSvl+Sy4YY86Yk89vP40lub8enjX3EI0uyRpITu87XTHL9Cs65eZKlSY5bjrG7JzloRe4vSZIkSdKq1BNJjaqaWVVXjXLMuVW1a1XtCvwMeEk7P7G7X5Lx2ix1DeB3sVTVE1W1zwrO+VrgO8DRyzF2d8CkhiRJkiSpZ6xWSY0k05Lc0XV+QpJZST6T5MjWdlCSRUluAA5fzvuckuTfk3wDODfJ9CTXJ7ktya1J9mz9DkhydZJLknwvyXldc3wwyV1JFiY5vbUdmuTmNs+VSTZr7Rsm+WyrFlmY5DDgNGDDVj1yXpK1kjzY+q+R5MNJ7mhjjhwpnuZo4B3ANkm2aGPWSvJgi3dekiuS7JnkW0l+lOQVSdYDZgLHtHiOHOa9faY924+THJbkX1ucc/oTREn+tM1/a5KvJ9m8tb85yS1JFiS5qN2XJOcn+WiSb7eYXrU8f66SJEmSpMlltUpqjCTJusDZwCHAPsAWKzDdbsAhVfV64OfAy6pqN+AY4MyufrsDbwF2AJ6XZK/2If0VwPOramfg/a3vdcBebZ5LgHe19lnA/1bVTsAuwLfoVGk81KpHjh0Q26vb/XYBXgZ8pD9BMlg87d1MA55RVbcCFwOv6ZpvCnBlVe0OPN7ieWm7z8lV9QhwMvAfLZ6Lh3lvW7dnPwL4T+DyqtoReBI4KMk6wEeBI6pqD+B84J/b2Iuq6k+rahfgh8Abu+bdDNgbOKzrfT5FkhlJ5iaZ+/ivfjNMiJIkSZKkyWC8ll4sr+2Bu6tqMXT+hx+YsZxzfaWqHm3H6wBnJdkFWApM7+p3U1X9vN1vPjANuJXOh/izk8wB+vf12Ar4QquSWAf4fms/gM6HdaqqgAdGWPbyIuA/q+oJ4H9aVUofnYTEYPHcRKdK48I2/gLg4/w+OfNIVX2jHd8O/LKqlia5vY0fja91jWXAvNOA5wHPB65KArAm8NPWZ+ckJwMbAxvy+/cG8OX2bhYmefZgN66q2cBsgCnbPrtGGbckSZIkaYJZ3ZIaS3lq9ci6g/RZWR9mu/+r/13AT4DXAU8Dft117bGu4yeAtarqt0n66FRRHAX8DfByOomEf6mqr6WzsWn/nhkZZdwZ5tofxNOOjwamJnlDO/+jJFu353q8a8yTXXM8yej/DnSPHTjvWi32hUPsD3IecHBV3ZHkeGCvQeaF4Z9fkiRJkiRg9Vt+ci+wWZKpbRnDKwdcXwRsnaS/kmJ5NsQczBTg561S4A2M8KE6yYbARlV1GfB3dJay9M9zTzolCm/oGnIl8NY2NkmeUVVL2/lgSYXrgKPS+UaUzeksy5g7TDw7AGtW1bOralpVTQM+SCfhsqweolM9saLuAp6d5AUttrWTPL9dezqdypOnAX+5Eu4lSZIkSZrEVqukRlX9ls7eDjfTWZqwaMD1R+ksN5nTlmT890q69VnA8UluAv6Yp1YNDGZKi2EB8E3gna19FvAlOntm3NvV/33A5ulsgjqfzn4gAOfQWW4xcMPPi+k8+wLgKuCdVXXfMPH8Zbtvty8yusTBN4Fd2iang24Uuiyq6jHgSODD7f3cBuzZLs8Evgt8g07yQ5IkSZKk5ZZOcYLUW6Zs++za+4N/M95hSJKk1cjXXvWP4x2CJGklSXJrVfWN1G+1qtSQJEmSJElaVqvbRqHLJcmbgLcPaL6xqt4yHvH0uraJ51sHNF9XVW8bj3gGs93Gz/J/YyRJkiRpkpsQSY2qOhc4d7zjmCiq6lPAp8Y7DkmSJEmShuPyE0mSJEmS1JMmRKWGJp/FD97Ln1/yr+MdhiRJWgnmHP6u8Q5BktSjrNSQJEmSJEk9yaSGJEmSJEnqSSY1JEmSJElST+qJpEaSa5P0JVk/yZwki5LcmeS0YcaclGR++3mi63i1+FrSJJskeXPX+ZZJLlzBOf80SSV56XKM3T/JXityf0mSJEmSVqXVLqmRZKTNSz9UVdsDuwF7Jzl4sE5VdWpV7VpVuwKP9B9X1ZmjvN9Y2QT4XVKjqn5SVa9dwTmPBm5ov0drf8CkhiRJkiSpZ4xpUiPJsUkWJlmQ5HNJDklyc5LbklyVZPPWb1aS2UmuBM5Lsl6SC9rYC4H1AKrq4aq6ph0/DswDnrMccZ2f5F+TXAP8S5K9knynxXVjku1av+OTXJzkiiSLk7y/ta/Vnuf2JHf0V38keXOSW9rzXpRkvda+RZKvdL2LPYHTgOe26pHTkmybZH7rv16Sz7b55yXZd7h42rU1gCOANwAHJ1m7tW/bYvx0q245L8mBSb6d5PutAmY6cDzw7hbPC4d5bx9Pck2SHybZt8W5KMk5Xf0Obu9zXpILkzy9tb+vvZ87knwySVr7De0dfDfJ94a6vyRJkiRJ3casSiHJ84GTgL2rakmSTYAC9qqqSnI88B6g/zu89gBeVFWPJHkn8HBV7ZxkZzrJi4HzbwwcAnx0OUOcDry0qp5MMqXd+4kkBwGnAP1VE7sAuwNLge8n+RiwJbBpVe3UFQvARVX1ydZ2GvBG4BPAx4FvVNVZrTJkfeBEYNtWSUKSbbtiexvweFXt1N7j1/oTLYPFU1U/A/YFFlXVj5LcCBwEXNrGPBd4DbCIzrt8rKpemOQI4MSqOjLJp4AlVXXGCO9tSlW9pI39KvBn/fMm2RG4rz3bS6vq4SQnAW8H/gX4aFW9tyUz/rPF+PU2b6rqBUn+ApjZrj1FkhnADIB1N33GCGFKkiRJkia6sVx6sT9wcVUtAaiq+5PsBFyY5FnA2sDdXf0vrapH2vG+wJlt3MIkC7snbomBzwNnVtWPljO+i6rqyXa8MZ0KkemD9Luqqh5q910EbAUsplNl8VHga8CVre/OSU5u820IXNba9wOOas+zFPhVks2Gie1FwAdb/zuT/AzoT3oMFs/P6Cw5uaD1uaCd9yc1flBVd7UxdwFXtfbbgf8zTByD+WrX2J8NmHdai3MH4NutsL/L1gAAIABJREFUEGNtOktiAF6a5N3AusCmwK38PqlxSft9a5vnD1TVbGA2wJRtt6xRxi1JkiRJmmDGMqkROpUZ3T4GfLiqLk2yHzCr69pvBvQd7kPrbGDxMlQVDKf7fqcCV1TVv7WKicu7rj3WdfwEsFZV/aJVkBxMp6riCDoVBOcBB1fVHa0SpXuPitF8CM8w1/4gniRPA14FvCLJe+ksK9q4f9nHgDFPdp0/yej/DnSPHTjvWi32y6vq9d2DkqwPnAXsXlX3JDmFTnJj4LxPLEdMkiRJkqRJaCz31LgaeE2SqdD5tg9gCnBPu/6GYcZeBxzTxu0I7Nx/oX0YngK8YyXG2h3XG0fqnOSZdJZLXAS8l85yEICnA//Tkgx/2TXkGtqmoEnWTLIR8BCdao7BdD//84BnAT8YJqSXA7dU1ZZVNa2qtqJTUfEXIz1Ll+HiGY1vAy9Osg1Akqe3pTPr0Ul8LEmyIZ1EkCRJkiRJy23MkhpVdSedCohvJVkAfJhOZcZFSa4Hlgwz/BPABm3ZyXuA7wIkeQ6dfTp2oLOHw/xWEbGiTgc+2PaiWBZbAte1jT3PBv6htc9ssX4DuKur/1uBA5PcDswFtq+qe4G5bTPQgV9N+zFgvdb/P4Bj28aoQzka+NKAti/y1MTKSL5CJwl124ps1Nme6zg6y4wW0Ely/ElV/QL4LHBHi/Xm5b2HJEmSJEnQqTYY7xikUZuy7Zb1og+szGIdSZI0XuYc/q6RO0mSJpUkt1ZV30j9xvQrXSVJkiRJksbKhNiQsX1t6KsHNF9UVaeORzy9LslM4PABzRdU1cBlMuNmu4039391JEmSJGmSc/mJelJfX1/NnTt3vMOQJEmSJI0Bl59IkiRJkqQJbUIsP9Hks/jB+/jzS84a7zAkSeoZcw5/63iHIEnSSmelhiRJkiRJ6kkmNSRJkiRJUk8yqSFJkiRJknqSSQ2tNEmeSDI/yZ1JFiR5Z5I12rWpSa5J8uskZw0Yd22S77Wx85NsNj5PIEmSJEnqJW4UqpXpkaraFaAlJv4TmAK8F3gU+Cdgx/Yz0DFV5Xe0SpIkSZKWmZUaGhNVdR8wA3hrklTVb6rqBjrJDUmSJEmSVphJDY2ZqvoRnb9jy7Kc5Ny29OSfkmSwDklmJJmbZO7jv/z1So1VkiRJktR7TGporA2aoBjgmKraCdin/bx+sE5VNbuq+qqqb+0pG6zMGCVJkiRJPcikhsZMkm2AJ4D7hutXVfe03w/R2YfjBWMfnSRJkiSp15nU0JhI8kzgk8BZVVXD9Fsryabt+GnAK4E7Vk2UkiRJkqRe5refaGVaL8l84GnAUuBzwIf7Lyb5MbARsHaSw4CXA/8NXNESGmsCVwFnr+K4JUmSJEk9yKSGVpqqWnOE69OGuLTHyo9GkiRJkjTRufxEkiRJkiT1JCs11JO223gz5hz+1vEOQ5IkSZI0jqzUkCRJkiRJPcmkhiRJkiRJ6kkmNSRJkiRJUk9yTw31pMUP/C9//sXZ4x2GJEmrnTlHzBjvECRJWmWs1JAkSZIkST3JpIYkSZIkSepJJjUkSZIkSVJPmpBJjSTXJulLsn6SOUkWJbkzyWnDjDkpyfz280TX8dtWZexDSbJJkjd3nW+Z5MIVnPNPk1SSly7H2P2T7LUc43ZPctAw13+aZOPRzitJkiRJmnx6PqmRZKTNTj9UVdsDuwF7Jzl4sE5VdWpV7VpVuwKP9B9X1ZmjvN9Y2QT4XVKjqn5SVa9dwTmPBm5ov0drf2DUSQ1gd2DIpIYkSZIkSctqtUpqJDk2ycIkC5J8LskhSW5OcluSq5Js3vrNSjI7yZXAeUnWS3JBG3shsB5AVT1cVde048eBecBzliOu85P8a5JrgH9JsleS77S4bkyyXet3fJKLk1yRZHGS97f2tdrz3J7kjv7qjyRvTnJLe96LkqzX2rdI8pWud7EncBrw3FY9clqSbZPMb/3XS/LZNv+8JPsOF0+7tgZwBPAG4OAka7f2bVuMn27VLeclOTDJt5N8v1XATAeOB97d4nnhEO/tqDbXgiTXtOebCRzTxh2Z5JlJvtHi/gSQ0f75SJIkSZImp9XmK12TPB84Cdi7qpYk2QQoYK+qqiTHA+8B3tWG7AG8qKoeSfJO4OGq2jnJznSSFwPn3xg4BPjocoY4HXhpVT2ZZEq79xNtKcUpQH/VxC50qhGWAt9P8jFgS2DTqtqpKxaAi6rqk63tNOCNwCeAjwPfqKqzWmXI+sCJwLatkoQk23bF9jbg8araqb3Hr/UnWgaLp6p+BuwLLKqqHyW5kU71xKVtzHOB1wCL6LzLx6rqhUmOAE6sqiOTfApYUlVnDPPO3gvsV1X3Jtm4/VmdDOxYVe9oz/FvwDVV9S9JDqWrGmWgJDOAGQDrbrrJMLeVJEmSJE0Gq1Olxv7AxVW1BKCq7qdTVXFFktuBdwPP7+p/aVU90o73Bc5v4xYCC7snbomBzwNnVtWPljO+i6rqyXa8MXBJkjuADw2I66qqeqjFtgjYCvgBnSqLjyY5EPhl67tzkuvb8x3VNc9+wL+351laVb8aIbYXAZ9r/e8Efgb0Jz0Giwc6S04uaMcX8NQlKD+oqrva894FXNXabwemjRBLtxvpVNIcz9B/17r/7L4CPDTUZFU1u6r6qqpv7Y02GEUYkiRJkqSJaHVKaoROZUa3jwFntQqHvwbW7br2mwF9B47tNhtYPEJVwUi673cqcEVV7QgcNiCux7qOnwDWqqpfADvT2b/ibbSEBXAe8Dft+U4ZMM9wzzPQcEs2/iCeJE8DXgWcnOTHwBnAnyd5+iBjnuw6f5LRVff8f3SqNaYBC5I8Y4h+o3lWSZIkSZKA1SupcTXwmiRTofNtH8AU4J52/Q3DjL0OOKaN25FOAoF2fkqb5x0rMdbuuN44UuckzwRSVRfR+ZC/e7v0dOB/WpLhL7uGXENbhpFkzSQb0alg2HCIW3Q///OAZ9GpDhnKy4FbqmrLqppWVVsBXwX+YqRn6TJcPP22qaqbgH8CHgCePci47tgPWYY5JUmSJEkCVqOkRls2cSrwrSQLgA8Ds4CLklwPLBlm+CeADZIspLPvxncBkjyHzj4dOwDz2uaUx6+EcE8HPtj2olgWWwLXtY09zwb+obXPbLF+g84yj35vBQ5sy1LmAttX1b3A3LYZ6MCvpv0YsF7r/x/AsW1j1KEcDXxpQNsXeWpiZSRfoZOEum2ojUKBj7SYbqezDOYO4JvALm3ckXSSPAckmUdn2c09Q8wlSZIkSdJTpMrKf/WeKdP/uF70gZPGOwxJklY7c46YMd4hSJK0wpLcWlV9I/VbbSo1JEmSJEmSRmO1+UrXVSnJScCrBzRfVFWnjkc8vS7JTODwAc0XVP0/9u49zM6yvPv+9wdECZsECAkFLbJTUBFCM0ZFoCogrx7VEolbFOVtTdun1Cry1Ne6Aa22BVtbkSrmQYH64A6QSsECBUECVDBASAhCRUBFqFEMhG0gyfn+se7oOMw2mZmVe+b7OY455l7X7j7XyvpnzpzXdVffbTKj5rnbz/R/oiRJkiRpknP7iVqpp6enFi9e3O0wJEmSJEljwO0nkiRJkiRpQhty+0mSnYC/BXapqtckeQHwsqr64phHJw3gzpUP8Afnn9XtMCRJGlcXHfWubocgSdImZTiVGmcBlwK7NK//G3jvWAUkSZIkSZI0HMNJauxYVd8A1gFU1Rpg7ZhGJUmSJEmSNIThJDUeTTIDKIAkLwUeGtOoJEmSJEmShjCcR7oeD1wI7JnkWmAmMH9Mo9KYS7IWWAZMAdYAZwP/DBwOnNwM2wv4GfA4sLSqjulnnXcBPVV13DiELUmSJEnSrw2a1EiyGbAl8PvA3kCAO6rqqXGITWPr8aqaDZBkFvAVYHpVnUjnDBWSXAWcUFXj/uzUJFs0W50kSZIkSerXoNtPqmod8I9VtaaqllfVrSY0Jp6qWgEsAI5Lkg1YYpcklyT5YZJT1jcmeWuSZUluTXJyr/ZHel3PT3JWc31Wkk8nuZLfVItIkiRJktSv4Ww/uSzJUcA3q6rGOiB1R1Xd1VTmzAJ+PsLps4EDgNXAHUk+S+cw2ZOBOcBKOt+jI6vq34ZY63nAYVX1tMNokyygk3xh6o4zRhiiJEmSJGmiGe6ZGlsDa5I8QWcLSlXVtDGNTN2wIVUaAFdU1UMASW4DngPMAK6qql807ecAhwBDJTXO7S+hAVBVC4GFANvtubsJNkmSJEma5IZMalTVtuMRiLoryR50qitWbMD01b2u19L5Xg2WIOmdkNiyT9+jG3B/SZIkSdIkNGRSI8kh/bVX1dWjH466IclM4HTgtFHcYnQ98JkkO9LZfvJW4LNN38+TPB+4A5gHPDxK95QkSZIkTSLD2X7yv3tdbwnMBW4EXjUmEWm8TE2yhN880vXLwKdHa/Gquj/JB4Er6VRtfLuqvtV0/3/ARcBPgVuBbUbrvpIkSZKkySMj/Y/5JL8LnFJVbx2bkKShbbfn7nXQKSd2OwxJksbVRUe9q9shSJI0LpLcWFU9Q40b9JGuA7gX2HcD5kmSJEmSJI2a4Zyp8Vl+c7DjZnQe33nLWAalTU+SI+g8orW3u6tqXjfi2Wv7Gf5vlSRJkiRNcsM5U2Nxr+s1wFer6toxikebqKq6FLi023FIkiRJkrTecJIa21XVZ3o3JPnLvm2SJEmSJEnjaThnaryzn7Z3jXIckiRJkiRJIzJgpUaStwJvA3ZPcmGvrm2BB8Y6MGkwd678FX9w3jndDkOStJEumn90t0OQJEktNtj2k+uA+4EdgX/s1f4wsHQsg5IkSZIkSRrKgEmNqvox8GPgZeMXjiRJkiRJ0vAMeaZGkpcm+X6SR5I8mWRtklXjEZwmjuZ7syTJ8iS3JDk+yWZN34wkVzbfsdO6HaskSZIkqR2G8/ST04C3AOcCPcAxwF5jGZQmpMerajZAklnAV4DpwInAE8BHgH2bH0mSJEmShjScp59QVXcCm1fV2qo6E3jl2IaliayqVgALgOOSpKoerapr6CQ3JEmSJEkaluFUajyW5BnAkiSn0Dk8dOuxDUsTXVXd1Ww/mQX8vNvxSJIkSZLaZziVGu9oxh0HPAr8LnDUWAalSSMjGpwsSLI4yeInV3msiyRJkiRNdkNWalTVj5NMBXauqo+NQ0yaBJLsAawFVgx3TlUtBBYCbLfnHjVGoUmSJEmSWmI4Tz95HbAEuKR5PTvJhWMdmCauJDOB04HTqsrkhCRJkiRpgwznTI2TgLnAVQBVtSTJbmMWkSaqqUmWAFOANcCXgU+v70xyDzANeEaSI4FXV9Vt3QhUkiRJktQOw0lqrKmqh5IRHX8g/Zaq2nyI/t3GKRRJkiRJ0gQxnKTGrUneBmye5LnAe4DrxjYsSZIkSZKkwQ14pkaSLzeXPwJeCKwGvgqsAt479qFJkiRJkiQNbLBKjTlJngO8GXgl8I+9+rYCnhjLwKTB7LX9Dlw0/+huhyFJkiRJ6qLBkhqn03niyR7A4l7tAapplyRJkiRJ6ooBt59U1alV9XzgS1W1R6+f3avKhIYkSZIkSeqqIQ8Krao/G49ApJG4c+VK/uC8b3Q7DElSHxfNf1O3Q5AkSZPIgJUakiRJkiRJmzKTGpIkSZIkqZVMamhUJVmbZEmS5UluSXJ8ks2avhlJrkzySJLT+sybk2RZkjuTnJok3XkHkiRJkqS2MKmh0fZ4Vc2uqhcChwOvBU5s+p4APgKc0M+8zwMLgOc2P//POMQqSZIkSWoxkxoaM1W1gk6i4rgkqapHq+oaOsmNX0uyMzCtqv6rqgr4V+DI8Y9YkiRJktQmJjU0pqrqLjrfs1mDDHsWcG+v1/c2bZIkSZIkDcikhsbDUOdj9NdfTxuULEiyOMniJ1etGp3IJEmSJEmtZVJDYyrJHsBaYMUgw+4Fnt3r9bOB+/oOqqqFVdVTVT3PmDZtdAOVJEmSJLWOSQ2NmSQzgdOB05qzMvpVVfcDDyd5afPUk2OAb41TmJIkSZKkltqi2wFowpmaZAkwBVgDfBn49PrOJPcA04BnJDkSeHVV3Qb8GXAWMBX4j+ZHkiRJkqQBmdTQqKqqzYfo322A9sXAvmMRkyRJkiRpYnL7iSRJkiRJaiWTGpIkSZIkqZXcfqJW2mv77blo/pu6HYYkSZIkqYus1JAkSZIkSa1kUkOSJEmSJLWS20/USneufJDXnfdv3Q5DkiaNf59/ZLdDkCRJehorNSRJkiRJUiuZ1JAkSZIkSa1kUkOSJEmSJLWSSQ2NiyRrkyxJsjzJLUmOT7JZnzG7JnkkyQndilOSJEmS1B4eFKrx8nhVzQZIMgv4CjAdOLHXmH8C/qMLsUmSJEmSWshKDY27qloBLACOSxKAJEcCdwHLuxmbJEmSJKk9TGqoK6rqLjrfv1lJtgY+AHxssDlJFiRZnGTxk6tWjUeYkiRJkqRNmEkNdVOa3x8D/qmqHhlscFUtrKqequp5xrRpYx+dJEmSJGmT5pka6ookewBrgRXAS4D5SU4BtgPWJXmiqk7rZoySJEmSpE2bSQ2NuyQzgdOB06qqgIN79Z0EPGJCQ5IkSZI0FJMaGi9TkywBpgBrgC8Dn+5uSJIkSZKkNjOpoXFRVZsPc9xJYxyKJEmSJGmC8KBQSZIkSZLUSlZqqJX22n47/n3+kd0OQ5IkSZLURVZqSJIkSZKkVjKpIUmSJEmSWsmkhiRJkiRJaiXP1FAr3bnyIf7wvG93OwxJ2uR8a/5rux2CJEnSuLFSQ5IkSZIktZJJDUmSJEmS1EomNTQukqxNsiTJ8iS3JDk+yWZN3+FJbkyyrPn9qm7HK0mSJEna9HmmhsbL41U1GyDJLOArwHTgROCXwOuq6r4k+wKXAs/qWqSSJEmSpFawUkPjrqpWAAuA45Kkqm6uqvua7uXAlkme2b0IJUmSJEltYFJDXVFVd9H5/s3q03UUcHNVrR7/qCRJkiRJbeL2E3VTfutF8kLgZODV/Q5OFtCp8GDqjjPHPDhJkiRJ0qbNSg11RZI9gLXAiub1s4ELgGOq6kf9zamqhVXVU1U9z5g2ffyClSRJkiRtkkxqaNwlmQmcDpxWVZVkO+Bi4INVdW13o5MkSZIktYVJDY2Xqesf6QpcDlwGfKzpOw7YC/hIM2ZJ84QUSZIkSZIG5JkaGhdVtfkgfZ8APjGO4UiSJEmSJgArNSRJkiRJUiuZ1JAkSZIkSa3k9hO10l7bT+db81/b7TAkSZIkSV1kpYYkSZIkSWolkxqSJEmSJKmV3H6iVrpz5SqOPO/ybochSRvs3+Yf1u0QJEmSWs9KDUmSJEmS1EomNSRJkiRJUiuZ1JAkSZIkSa00YZMaSdYmWZLk1iTnJtlqnO+/S5LzmutXJLlogHH3JNlxDOPoSXLqBs59T5IfJDlntOOSJEmSJGljTdikBvB4Vc2uqn2BJ4E/Hc+bV9V9VTV/PO85QByLq+o9Gzj9fwGvraqjRzMmSZIkSZJGw0ROavS2CNhroM4kb09yQ1PZ8YUkmzftjyQ5OcmNSS5PMjfJVUnuSvL6ZsxuSRYluan5ObBX+6393GtGksuS3JzkC0B69R3fVJbcmuS9vda5PckZTfs5SQ5Lcm2SHyaZ24ybm+S6Zt3rkuzdtP+6SiTJSUm+1Os9DJjsSHI6sAdwYZL39Y07yY8HqzAZ4jP9ZJJbknwvyU5N+05JLmjab1n/OUqSJEmSNJAJn9RIsgXwGmDZAP3PB94MvLyqZgNrgfWVCVsDV1XVHOBh4BPA4cA84OPNmBXA4VX1e806Q231OBG4pqoOAC4Edm3imAMcC7wEeCnw7iQHNHP2Aj4D7AfsA7wNOAg4AfjrZsztwCHNuh8F/naA++8DHAHMBU5MMqW/QVX1p8B9wCur6p8Girs/w/hMv1dV+wNXA+9u2k8Fvtu0/x6wvJ91FyRZnGTxk6seGuj2kiRJkqRJYotuBzCGpiZZ0lwvAr44wLhDgTnA95MATKWTqIDOtpVLmutlwOqqeirJMmC3pn0KcFqS9X+8P2+IuA4B3gBQVRcnWdm0HwRcUFWPAiT5JnAwnQTC3VW1rGlfDlxRVdUnjunA2UmeC1QTV38urqrVwOokK4CdgHuHiHmwuPsz1Ge6/nyRG+kkiQBeBRzTrL8WeFrWoqoWAgsBttvzeTWMmCVJkiRJE9hETmo83lQJDCXA2VX1wX76nqqq9X88rwNWA1TVuqYCBOB9wM+B/elUvjwxjHv29wd5+mlbb3Wv63W9Xq/jN/+GfwNcWVXzkuwGXDWMtdYysu/AcBMJw/1MR3p/SZIkSZJ+bcJvPxmGK4D5SWYBJNkhyXNGMH86cH9VrQPeAWw+xPirabZiJHkNsH2v9iOTbJVkazpbXBaNMI6fNdfvGsG84Roo7v5syGd6BfBnzfjNk0zb+JAlSZIkSRPZpE9qVNVtwIeBy5IsBf4T2HkES3wOeGeS79HZevLoEOM/BhyS5Cbg1cBPmjhuAs4CbgCuB86oqptHEMcpwN8luZahEysbot+4+7OBn+lfAq9sttTcCLxwVKKWJEmSJE1Y+c1OAGn4ktwD9FTVL7tx/+32fF694uTPdePWkjQq/m3+Yd0OQZIkaZOV5Maq6hlq3KSv1JAkSZIkSe00aQ5pTDKDzrkNfR1aVQ+Mdzybig39XKpqtyQzej1hZthzR8Ne20/zfzklSZIkaZKbNEmN5o/s4TwNZVLZmM/Fz1SSJEmS1E1uP5EkSZIkSa1kUkOSJEmSJLXSpNl+oonlRysfYd7513Q7DEka0AVHHdTtECRJkiY8KzUkSZIkSVIrmdSQJEmSJEmtZFJDkiRJkiS1kkmNcZZkbZIlSW5Ncm6Srcb5/rskOa+5fkWSiwYYd0+SHccwjp4kp47V+pIkSZKkic+kxvh7vKpmV9W+wJPAn47nzavqvqqaP573HCCOxVX1nm7HIUmSJElqL5Ma3bUI2GugziRvT3JDU9nxhSSbN+2PJDk5yY1JLk8yN8lVSe5K8vpmzG5JFiW5qfk5sFf7rf3ca0aSy5LcnOQLQHr1Hd9Ultya5L291rk9yRlN+zlJDktybZIfJpnbjJub5Lpm3euS7N20/7pKJMlJSb7U6z2Y7JAkSZIkDcmkRpck2QJ4DbBsgP7nA28GXl5Vs4G1wNFN99bAVVU1B3gY+ARwODAP+HgzZgVweFX9XrPOUFs9TgSuqaoDgAuBXZs45gDHAi8BXgq8O8kBzZy9gM8A+wH7AG8DDgJOAP66GXM7cEiz7keBvx3g/vsARwBzgROTTOnnM1mQZHGSxatXPTjE25EkSZIkTXRbdDuASWhqkiXN9SLgiwOMOxSYA3w/CcBUOokK6GxbuaS5XgasrqqnkiwDdmvapwCnJVmfEHneEHEdArwBoKouTrKyaT8IuKCqHgVI8k3gYDqJj7uralnTvhy4oqqqTxzTgbOTPBeoJq7+XFxVq4HVSVYAOwH39h5QVQuBhQDb77lPDfF+JEmSJEkTnEmN8fd4U3kxlABnV9UH++l7qqrW/1G/DlgNUFXrmgoQgPcBPwf2p1OR88Qw7tlfoiD9tK23utf1ul6v1/Gb79bfAFdW1bwkuwFXDWOttfjdlCRJkiQNwe0nm64rgPlJZgEk2SHJc0Ywfzpwf1WtA94BbD7E+KtptrckeQ2wfa/2I5NslWRrOltcFo0wjp811+8awTxJkiRJkgZlUmMTVVW3AR8GLkuyFPhPYOcRLPE54J1Jvkdn68mjQ4z/GHBIkpuAVwM/aeK4CTgLuAG4Hjijqm4eQRynAH+X5FqGTqxIkiRJkjRs+c0uBqk9tt9zn3rFKWd0OwxJGtAFRx3U7RAkSZJaK8mNVdUz1DgrNSRJkiRJUit5GGOXJZlB5/yMvg6tqgfGO5622HP7bfxfUEmSJEma5ExqdFmTuBjO01AkSZIkSVIvbj+RJEmSJEmtZKWGWulHKx/jqPMXdzsMSZPI+UcNeU6VJEmSxpmVGpIkSZIkqZVMakiSJEmSpFYyqaFRl+R3knwtyY+S3Jbk20mel+SSJA8muajP+C8muSXJ0iTnJdmmW7FLkiRJktrDpIZGVZIAFwBXVdWeVfUC4K+BnYBPAe/oZ9r7qmr/qtoP+Alw3LgFLEmSJElqLQ8K1Wh7JfBUVZ2+vqGqlqy/TvKKvhOqalXTF2AqUGMfpiRJkiSp7azU0GjbF7hxpJOSnAn8D7AP8NnRDkqSJEmSNPGY1NAmoaqOBXYBfgC8ub8xSRYkWZxk8epVK8c1PkmSJEnSpsekhkbbcmDOhkysqrXA14GjBuhfWFU9VdXzzGnbb0SIkiRJkqSJwKSGRtt3gGcmeff6hiQvTvL7/Q1Ox17rr4HXAbePS6SSJEmSpFYzqaFRVVUFzAMObx7puhw4CbgvySLgXODQJPcmOQIIcHaSZcAyYGfg492JXpIkSZLUJj79RKOuqu4D3tRP18EDTHn5GIYjSZIkSZqgrNSQJEmSJEmtZFJDkiRJkiS1kttP1Ep7br8V5x/V0+0wJEmSJEldZKWGJEmSJElqJZMakiRJkiSpldx+ola6a+UTvOn827odhqQJ6htHvaDbIUiSJGkYrNSQJEmSJEmtZFJDkiRJkiS1kkkNSZIkSZLUSiY1NGqSnJTkhCQfT3LYIOOOTPKCXq/fmGR5knVJfE6rJEmSJGlYTGpo1FXVR6vq8kGGHAn0PoXvVuANwNVjGpgkSZIkaUIxqaGNkuRDSe5Icjmwd9N2VpL5zfXfJ7ktydIk/5DkQOD1wKeSLEmyZ1X9oKru6OLbkCRJkiS1kI901QZLMgd4C3AAne/STcCNvfp3AOYB+1RVJdmuqh5MciFwUVWdN8L7LQAWAGy1486j9C4kSZIkSW1lpYY2xsHABVX1WFWtAi7s078KeAI4I8kbgMc25mZVtbCqeqqq55nTdtiYpSRJkiQ/5hohAAAgAElEQVRJE4BJDW2sGrCjag0wFzifzjkal4xXUJIkSZKkic+khjbG1cC8JFOTbAu8rndnkm2A6VX1beC9wOym62Fg23GNVJIkSZI04ZjU0AarqpuArwNL6FRjLOozZFvgoiRLge8C72vavwb87yQ3J9kzybwk9wIvAy5Ocun4vANJkiRJUpt5UKg2SlV9EvjkIEPm9jPnWn77ka4/Ai4Y5dAkSZIkSROclRqSJEmSJKmVrNRQK+2x/ZZ846gXDD1QkiRJkjRhWakhSZIkSZJayaSGJEmSJElqJZMakiRJkiSplTxTQ610z4NPcuw3f9LtMCS11Jlv2LXbIUiSJGkUWKkhSZIkSZJayaSGJEmSJElqJZMakiRJkiSplUxqaMSSXDdA+1lJ5g8yb/ck1yf5YZKvJ3lG0358ktuSLE1yRZLnjFXskiRJkqSJw6SGRqyqDtzAqScD/1RVzwVWAn/UtN8M9FTVfsB5wCkbH6UkSZIkaaIzqaERS/JI8ztJTmuqLC4GZg0yJ8Cr6CQtAM4GjgSoqiur6rGm/XvAs8cseEmSJEnShGFSQxtjHrA38CLg3cBgFRwzgAerak3z+l7gWf2M+yPgP/pbIMmCJIuTLH7ioV9teNSSJEmSpAnBpIY2xiHAV6tqbVXdB3xnkLHpp61+a0DydqAH+FR/C1TVwqrqqaqeLafvsKExS5IkSZImiC26HYBar4YeAsAvge2SbNFUazwbuG99Z5LDgA8Bv19Vq0c/TEmSJEnSRGOlhjbG1cBbkmyeZGfglQMNrKoCrgTWPx3lncC3AJIcAHwBeH1VrRjbkCVJkiRJE4VJDW2MC4AfAsuAzwPfHWL8B4Djk9xJ54yNLzbtnwK2Ac5NsiTJhWMUryRJkiRpAnH7iUasqrZpfhdw3Ajm3QXM7af9sNGLTpIkSZI0WVipIUmSJEmSWslKDY26JBcAu/dp/kBVXTpa99htu2dw5ht2Ha3lJEmSJEktZFJDo66q5nU7BkmSJEnSxOf2E0mSJEmS1EpWaqiV7nvwKU664L5uhyGpJU6at0u3Q5AkSdIYsFJDkiRJkiS1kkkNSZIkSZLUSiY1JEmSJElSK5nU2MQkWZtkSZLlSW5JcnySzZIc0bQvSfJIkjua638d5rpXJekZ6/g3VJKzkszvdhySJEmSpPbwoNBNz+NVNRsgySzgK8D0qjoRuLRpvwo4oaoWdy1KSZIkSZK6zEqNTVhVrQAWAMclyUjmJpma5GtJlib5OjC1V9/nkyxuqkE+1rQdmuSCXmMOT/LNAdZ+U5JPN9d/meSu5nrPJNc013OSfDfJjUkuTbJzrzGXNO2LkuzTz/p/01Ru+P2UJEmSJA3IPxo3cVV1F51/p1kjnPpnwGNVtR/wSWBOr74PVVUPsB/w+0n2A74DPD/JzGbMscCZA6x9NXBwc30w8ECSZwEHAYuSTAE+C8yvqjnAl5oYABYCf9G0nwB8rvfCSU5p3uuxVbWuT9+CJhmz+LFVD4zks5AkSZIkTUBuP2mHEVVpNA4BTgWoqqVJlvbqe1OSBXT+/XcGXtCM+TLw9iRnAi8Djulv4ar6nyTbJNkW+F06W2QOoZPg+CawN7Av8J9NgcnmwP1JtgEOBM7tVXjyzF5LfwS4vqoWDHDfhXSSIuyy1/41kg9DkiRJkjTxmNTYxCXZA1gLrNiA6U/7wz/J7nQqJF5cVSuTnAVs2XSfCfw78ARwblWtGWTt/6JTzXEHsAj4f+kkQt4P7Aosr6qX9bn3NODB9WeG9OP7wJwkO1TVr4b3FiVJkiRJk5XbTzZhzVaQ04HTqmqklQlXA0c36+xLZ6sJwDTgUeChJDsBr1k/oaruA+4DPgycNYz1T2h+3wy8ElhdVQ/RSXTMTPKy5v5TkrywqlYBdyd5Y9OeJPv3WvMS4O+Bi5sqEEmSJEmSBmSlxqZnapIlwBRgDfBl4NMbsM7ngTObbSdLgBsAquqWJDcDy4G7gGv7zDsHmFlVtw2x/iI6W0+urqq1SX4K3N7c48nm8aynJplO53v2z809jwY+n+TDzXv8GnDL+kWr6twmoXFhktdW1eMb8N4lSZIkSZNARl4AoIksyWnAzVX1xW7HMphd9tq/FnzqP7odhqSWOGneLt0OQZIkSSOQ5MbmAReDslJDv5bkRjpbU97f7VgkSZIkSRqKSY2WS3IEcHKf5rurat5I12oes9p3/ev57SeUALyjqpaNdP3RtMt2U/yfV0mSJEma5ExqtFxVXQpcOobrv2Ss1pYkSZIkaWP49BNJkiRJktRKJjUkSZIkSVIruf1ErbTiwaf4lwt+3u0wJG2C/nzeTt0OQZIkSePESg1JkiRJktRKJjUkSZIkSVIrmdSQJEmSJEmtZFKjhZKclOSEJB9Pctgg445M8oJer9+YZHmSdUl6hnGfryZZmuR9oxW7JEmSJEmjxYNCW6yqPjrEkCOBi4Dbmte3Am8AvjDU2kl+Bziwqp6zUUFKkiRJkjRGrNRoiSQfSnJHksuBvZu2s5LMb67/PsltTWXFPyQ5EHg98KkkS5LsWVU/qKo7hnnLy4BZzdyDk8xJckuS/0ryqSS3DhLr5s2Y7zfx/EnT/ookVyU5L8ntSc5JkqbvxUmua+5xQ5JtN+LjkiRJkiRNAlZqtECSOcBbgAPo/JvdBNzYq38HYB6wT1VVku2q6sEkFwIXVdV5G3Db1zdzZzf3WAr8RVV9N8mnhpj7R8BDVfXiJM8Erk1yWdN3APBC4D7gWuDlSW4Avg68uaq+n2Qa8Hg/n8MCYAHA9jOfvQFvSZIkSZI0kVip0Q4HAxdU1WNVtQq4sE//KuAJ4IwkbwAeG82bJ5kObFdV322avjzElFcDxyRZAlwPzACe2/TdUFX3VtU6YAmwG53Kk/ur6vsAVbWqqtb0XbSqFlZVT1X1bDNth41+X5IkSZKkdjOp0R41YEcnATAXOJ/OORqXjPK9M9j9Bxj/F1U1u/nZvarWV2qs7jVuLZ3Kk5GuL0mSJEmSSY2WuBqYl2Rqc9bE63p3JtkGmF5V3wbeC8xuuh4GNvpsiqp6EHgoyUFN09FDTLkU+LMkU5r4npdk60HG3w7skuTFzfhtk7g1SpIkSZI0KP9wbIGquinJ1+ls1/gxsKjPkG2BbyXZkk7Vw/pHsH4N+D9J3gPMB/YDPgvMBC5OsqSqjhhmGMcCX0ryGJ2kxWDOoLOt5KbmINBf0KkgGej9PZnkzcBnk0ylc57GYcAjw4xNkiRJkjQJpcqqf41Mkt3oHCK6b7di2HWv/esDn7ps6IGSJp0/n7dTt0OQJEnSRkpyY1X1DDXO7SeSJEmSJKmV3H4yySU5Aji5T/PdVTVvoDlVdQ+w74bMHS2ztpvi/8ZKkiRJ0iRnUmOSq6pLGfqMjFGfK0mSJEnSxnL7iSRJkiRJaiUrNdRKv1q5hnPO/0W3w5A0To4+ama3Q5AkSdImyEoNSZIkSZLUSiY1JEmSJElSK5nUkCRJkiRJrWRSo8WS/E6SryX5UZLbknw7yfOSXJLkwSQX9Rn/xSS3JFma5Lwk2wyy9swk1ye5OcnBY/9uJEmSJEkaGZMaLZUkwAXAVVW1Z1W9APhrYCfgU8A7+pn2vqrav6r2A34CHDfILQ4Fbq+qA6pq0SiHL0mSJEnSRvPpJ+31SuCpqjp9fUNVLVl/neQVfSdU1aqmL8BUoPpbOMls4BRgapIlwMuAtwAfBO4H/htYXVX9JkWSzAROB3Ztmt5bVdcmOalp26P5/c9VdWoz5xjghCampVXVX1JGkiRJkqRfM6nRXvsCN450UpIzgdcCtwHv729MVS1J8lGgp6qOS7Iz8DFgDvAQcCVw8yC3+QzwT1V1TZJdgUuB5zd9+9BJyGwL3JHk88DzgA8BL6+qXybZYYDYFwALAGbs+OwRvGtJkiRJ0kTk9pNJpqqOBXYBfgC8eZjTXkJnm8svqupJ4OtDjD8MOK2p8rgQmJZk26bv4qpaXVW/BFbQ2S7zKuC8po2q+tUAsS+sqp6q6pk2bcYwQ5ckSZIkTVQmNdprOZ3KiRGrqrV0EhNHjWTaCMZuBrysqmY3P8+qqoebvtW9xq2lUy2UEa4vSZIkSZJJjRb7DvDMJO9e35DkxUl+v7/B6dhr/TXwOuD2Yd7reuAVSWYkmQK8cYjxl9HrENLmjI7BXAG8KcmMZny/208kSZIkSerNpEZLVVUB84DDm0e6LgdOAu5Lsgg4Fzg0yb1JjqBTDXF2kmXAMmBn4OPDvNf9zdr/BVwO3DTElPcAPc2jY28D/nSI9ZcDnwS+m+QW4NPDiUuSJEmSNLml87exNHxJ3kVziGi3Ythjz9n1N6f8Z7duL2mcHX3UzG6HIEmSpHGU5Maq6hlqnJUakiRJkiSplazUmOSSfIinn5FxblV9ciznbqyenp5avHjxWN9GkiRJktQFw63UMKmhVjKpIUmSJEkTl9tPJEmSJEnShLZFtwOQNsSDK9dw4bm/7HYYkkbJ69+4Y7dDkCRJUgtZqSFJkiRJklrJpIYkSZIkSWolkxqSJEmSJKmVTGpo1CRZm2RJkuVJbklyfJLNmr4ZSa5M8kiS0/rM+2SSnyZ5pDuRS5IkSZLayKSGRtPjVTW7ql4IHA68Fjix6XsC+AhwQj/z/h2YOz4hSpIkSZImCpMaGhNVtQJYAByXJFX1aFVdQye50Xfs96rq/nEPUpIkSZLUaiY1NGaq6i4637FZo7FekgVJFidZvGrVA6OxpCRJkiSpxUxqaKxltBaqqoVV1VNVPdOmzRitZSVJkiRJLWVSQ2MmyR7AWmBFt2ORJEmSJE08JjU0JpLMBE4HTquq6nY8kiRJkqSJZ4tuB6AJZWqSJcAUYA3wZeDT6zuT3ANMA56R5Ejg1VV1W5JTgLcBWyW5Fzijqk4a7+AlSZIkSe1iUkOjpqo2H6J/twHa/wr4q7GISZIkSZI0cbn9RJIkSZIktZKVGmql7bbfgte/ccduhyFJkiRJ6iIrNSRJkiRJUiuZ1JAkSZIkSa1kUkOSJEmSJLWSZ2qolVb9ag2Xf+UX3Q5DUj8Oe9vMbocgSZKkScJKDUmSJEmS1EomNSRJkiRJUiuZ1JAkSZIkSa1kUkMjluS6AdrPSjJ/kHnnJLkjya1JvpRkStOeJKcmuTPJ0iS/N1axS5IkSZImDpMaGrGqOnADp54D7AO8CJgK/HHT/hrguc3PAuDzGxujJEmSJGni8+knGrEkj1TVNkkCfBZ4FXA3kMHmVdW3e61xA/Ds5uUfAv9aVQV8L8l2SXauqvvH5h1IkiRJkiYCKzW0MeYBe9OpvHg3MKwKjmbbyTuAS5qmZwE/7TXk3qat77wFSRYnWfzQww9sTNySJEmSpAnApIY2xiHAV6tqbVXdB3xnmPM+B1xdVYua1/1VeNTTGqoWVlVPVfVM33bGhkUsSZIkSZow3H6ijfW05MNgkpwIzAT+pFfzvcDv9nr9bOC+jQ9NkiRJkjSRWamhjXE18JYkmyfZGXjlYIOT/DFwBPDWqlrXq+tC4JjmKSgvBR7yPA1JkiRJ0lCs1NDGuIDOIaHLgP8GvjvE+NOBHwP/1TljlG9W1ceBbwOvBe4EHgOOHauAJUmSJEkTh0kNjVhVbdP8LuC4Eczr9/vWrPPnoxOdJEmSJGmycPuJJEmSJElqJSs1NOqSXADs3qf5A1V16WjdY9oOW3DY22aO1nKSJEmSpBYyqaFRV1Xzuh2DJEmSJGnic/uJJEmSJElqJSs11EqPPLCG6/71F90OQ1IfBx7jtjBJkiSNHys1JEmSJElSK5nUkCRJkiRJrWRSQ12R5KokPb2u70iypPmZ1e34JEmSJEmbPs/U0JhLskVVrRli2NFVtXhcApIkSZIkTQgmNTQiSY4BTgAKWAp8A/gw8AzgATrJiZ8nOQnYBdgN+GWSPwLOBF4A/ACYOu7BS5IkSZImFJMaGrYkLwQ+BLy8qn6ZZAc6yY2XVlUl+WPgr4D3N1PmAAdV1eNJjgceq6r9kuwH3NRn+TOTrAXOBz5RVTUub0qSJEmS1FomNTQSrwLOq6pfAlTVr5K8CPh6kp3pVGvc3Wv8hVX1eHN9CHBqM29pkqW9xh1dVT9Lsi2dpMY7gH/te/MkC4AFADvNePbovjNJkiRJUut4UKhGInQqM3r7LHBaVb0I+BNgy159j/YZ22/1RVX9rPn9MPAVYO4A4xZWVU9V9Wy37YwNCF+SJEmSNJGY1NBIXAG8KckMgGb7yXTgZ03/OweZezVwdDNvX2C/5nqLJDs211OAPwBuHZPoJUmSJEkTittPNGxVtTzJJ4HvNudf3AycBJyb5GfA94DdB5j+eTrnZiwFlgA3NO3PBC5tEhqbA5cD/2fs3oUkSZIkaaIwqaERqaqzgbP7NH+rn3En9Xn9OPCWAZadMyrBSZIkSZImFbefSJIkSZKkVjKpIUmSJEmSWsntJ2qlbWZswYHHzOx2GJIkSZKkLrJSQ5IkSZIktZJJDUmSJEmS1EomNSRJkiRJUit5poZa6bFfruHmM1Z0OwxpQjvgj2d1OwRJkiRpUFZqSJIkSZKkVjKpIUmSJEmSWsmkhsZFkrVJliRZnuSWJMcn2azpm5Lk7CTLkvwgyQe7Ha8kSZIkadPnmRoaL49X1WyAJLOArwDTgROBNwLPrKoXJdkKuC3JV6vqnq5FK0mSJEna5FmpoXFXVSuABcBxSQIUsHWSLYCpwJPAqi6GKEmSJElqAZMa6oqquovO928WcB7wKHA/8BPgH6rqV10MT5IkSZLUAiY11E1pfs8F1gK7ALsD70+yx9MGJwuSLE6yeOXDD4xjmJIkSZKkTZFJDXVFk7RYC6wA3gZcUlVPNVtTrgV6+s6pqoVV1VNVPdtvO2N8A5YkSZIkbXJMamjcJZkJnA6cVlVFZ8vJq9KxNfBS4PZuxihJkiRJ2vT59BONl6lJlgBTgDXAl4FPN33/ApwJ3EpnS8qZVbW0K1FKkiRJklrDpIbGRVVtPkjfI3Qe6ypJkiRJ0rC5/USSJEmSJLWSSQ1JkiRJktRKbj9RK2214xYc8Mezuh2GJEmSJKmLrNSQJEmSJEmtZFJDkiRJkiS1kttP1EpPrHiKO/7l590OQ2q9vf98p26HIEmSJG0wKzUkSZIkSVIrmdSQJEmSJEmtZFJDkiRJkiS10qRJaiRZm2RJkluTnJtkq3G+/y5JzmuuX5HkogHG3ZNkxzGMoyfJqWO1/hD3npnk+iQ3Jzk4ybeTbNeNWCRJkiRJ7TdpkhrA41U1u6r2BZ4E/nQ8b15V91XV/PG85wBxLK6q94zV+kk2H6T7UOD2qjqgqhZV1Wur6sGxikWSJEmSNLFNpqRGb4uAvQbqTPL2JDc0lR1fWP+HepJHkpyc5MYklyeZm+SqJHcleX0zZrcki5Lc1Pwc2Kv91n7uNSPJZU31wheA9Oo7vqksuTXJe3utc3uSM5r2c5IcluTaJD9MMrcZNzfJdc261yXZu2n/dZVIkpOSfKnXexgw2dHrvmcnWZrkvPXVLk11yUeTXAO8McmeSS5pPqdFSfZJMhs4BXht87lOXV+VkuTFzZpbJtk6yfIk+47oX1SSJEmSNOlMuqRGki2A1wDLBuh/PvBm4OVVNRtYCxzddG8NXFVVc4CHgU8AhwPzgI83Y1YAh1fV7zXrDLXV40Tgmqo6ALgQ2LWJYw5wLPAS4KXAu5Mc0MzZC/gMsB+wD/A24CDgBOCvmzG3A4c0634U+NsB7r8PcAQwFzgxyZRBYt0bWFhV+wGrgP/Vq++Jqjqoqr4GLAT+ovmcTgA+V1VLmji+3lTMPL5+YlV9v3nvn6CT+Pi/VdVfAmhBksVJFq985FeDhClJkiRJmgy26HYA42hqkiXN9SLgiwOMOxSYA3w/CcBUOokK6GxbuaS5XgasrqqnkiwDdmvapwCnNZUJa4HnDRHXIcAbAKrq4iQrm/aDgAuq6lGAJN8EDqbzx//dVbWsaV8OXFFV1SeO6cDZSZ4LVBNXfy6uqtXA6iQrgJ2AewcY+9Oqura5/r/Ae4B/aF5/vYlnG+BA4Nzm8wN45hCfAXSSQt8HnmjWfZqqWkgnYcK+u+5fw1hTkiRJkjSBTaakxuNN5cVQApxdVR/sp++pqlr/x/Q6YDVAVa1rKkAA3gf8HNifTiXME8O4Z39/oKeftvVW97pe1+v1On7zb/o3wJVVNS/JbsBVw1hrLYN/J/rG2fv1o83vzYAHh/lZ97YDsA2d5MuWvdaTJEmSJKlfk277yTBcAcxPMgsgyQ5JnjOC+dOB+6tqHfAOYLCDMwGuptnekuQ1wPa92o9MslWSrelscVk0wjh+1ly/awTzBrNrkpc1128Fruk7oKpWAXcneSNAOvYfxtoLgY8A5wAnj1K8kiRJkqQJzKRGH1V1G/Bh4LIkS4H/BHYewRKfA96Z5Ht0tp4MVXHwMeCQJDcBrwZ+0sRxE3AWcANwPXBGVd08gjhOAf4uybUMnVgZrh/QeW9L6VRWfH6AcUcDf5TkFmA58IeDLZrkGGBNVX0F+HvgxUleNUoxS5IkSZImqPxmN4U0sGYLy0XNI3G7bt9d96/zP3BZt8OQWm/vP9+p2yFIkiRJT5PkxqrqGWqclRqSJEmSJKmVJtNBob8lyQw652f0dWhVPTDe8WwqhvhcNokqDYAtZ03xf5glSZIkaZKbtEmNJnEx0id0THh+LpIkSZKktnD7iSRJkiRJaqVJW6mhdnvy50/x03/8n26HIbXS777/d7odgiRJkjQqrNSQJEmSJEmtZFJDkiRJkiS1kkkNSZIkSZLUSiY1NGJJrhug/awk8weZ98UktyRZmuS8JNs07YckuSnJmsHmS5IkSZLUm0kNjVhVHbiBU99XVftX1X7AT4DjmvafAO8CvjIK4UmSJEmSJgmffqIRS/JIVW2TJMBngVcBdwMZbF5VrWrmB5gKVNN+T9O+bgzDliRJkiRNMFZqaGPMA/YGXgS8GxiygiPJmcD/APvQSYgMW5IFSRYnWfyrRx/YgHAlSZIkSROJSQ1tjEOAr1bV2qq6D/jOUBOq6lhgF+AHwJtHcrOqWlhVPVX/f3v3HmZXVeZ5/PubJCAXxQAmoKAwzUURIUqCIIIgN2XsUZDW0LaCjUQdQB2c9vY4greZRp3uEWnRiAj2KNDaxHaQFhAMIHJLICQEoUFFoUGi3ARULuHtP86uprqsSqWuJ/vU9/M89Zx91l5rnfdU1rNPnTdrrV1zN91os1EFLEmSJEnqHSY1NFY14gZVq4FzgTeOfziSJEmSpKnCpIbG4nJgfpJpSbYE9huqYjq26zsG/hS4ZXLClCRJkiT1IjcK1VgsorNJ6ArgX4DL1lA3wFlJntUc3wi8GyDJvKavmcCfJvl4Vb14IgOXJEmSJLWfSQ2NWFVt3DwWT9+Wdbg2TwF7DXHuOmCrcQtQkiRJkjQluPxEkiRJkiS1kjM1NO6SLAK2HVD8waq6cLxeY73ZM9j6/VuMV3eSJEmSpBYyqaFxV1WHdjsGSZIkSVLvc/mJJEmSJElqJZMakiRJkiSplVx+olZ64leP86vP3tHtMKRW2OKvtul2CJIkSdKEcKaGJEmSJElqJZMakiRJkiSplUxqSJIkSZKkVjKpoRFL8uMhys9Mcvga2h2X5PYklWTzfuUzkyxKsjzJtUl2noi4JUmSJEm9xaSGRqyqXjHKplcCBwC/GFD+EWBZVe0CvA34/BjCkyRJkiRNESY1NGJJHmkek+TUJDcn+R4wa03tquqGqrpjkFM7AZc0dW4Btkkye5zDliRJkiT1GJMaGotDgR2BlwDHAKOdwXEjcBhAkt2BFwBbDayUZEGSJUmW3PfofaN8KUmSJElSrzCpobHYBzi7qlZX1d3ApaPs56+BmUmWAccDNwBPDqxUVQuram5Vzd1so81GHbQkSZIkqTdM73YAar0acwdVvwXeDp0lLcDPmx9JkiRJkobkTA2NxeXA/CTTkmwJ7DeaTpI8O8l6zdN3AJc3iQ5JkiRJkoZkUkNjsQi4DVgBnAZctqbKSd6T5C46+2UsT3J6c+pFwMoktwCvBd47cSFLkiRJknqFy080YlW1cfNYwHEjaHcKcMog5VcB249bgJIkSZKkKcGZGpIkSZIkqZWcqaFxl2QRsO2A4g9W1YXj9RoztliPLf5qm/HqTpIkSZLUQiY1NO6q6tBuxyBJkiRJ6n0uP5EkSZIkSa3kTA210hP3/oFf/c3N3Q5DWidtccJO3Q5BkiRJmhTO1JAkSZIkSa1kUkOSJEmSJLWSSQ1NmiRbJDknyU+T3JzkgiS7J7kqycoky5O8udtxSpIkSZLawT01NCmSBFgEnFVV85uyOcAmwNuq6rYkzwWWJrmwqh7sYriSJEmSpBYwqaHJsh/wRFV9qa+gqpb1r1BVdydZBTwHMKkhSZIkSVojl59osuwMLF1ThSS7A+sBP52UiCRJkiRJrWZSQ+uEJFsCfw+8vaqeGqLOgiRLkiy579H7JzdASZIkSdI6x6SGJstKYLfBTiR5FvA94KNVdfVQHVTVwqqaW1VzN9to0wkKU5IkSZLUFiY1NFkuBdZPckxfQZJ5SV5FZwPRr1fVt7oWnSRJkiSpdUxqaFJUVQGHAgc2t3RdCZwE7NP8HJVkWfMzp4uhSpIkSZJawrufaNJU1d3AmwY59cnJjkWSJEmS1H7O1JAkSZIkSa1kUkOSJEmSJLWSy0/USjNmP4MtTtip22FIkiRJkrrImRqSJEmSJKmVTGpIkiRJkqRWMqkhSZIkSZJayT011EpP3Ps77v2/S7sdhjQpZr9vt26HIEmSJK2TnKkhSZIkSZJayaSGJEmSJElqJZMakiRJkiSplUxqrEGS1UmWJbkpybeSbDjJr//cJN9ujvdNcv4Q9e5IsvkExjE3ySmjaLdNkpuGOHd6kp3GHp0kSZIkaaoyqbFmv6+qOVW1M/A48DPjRoMAABR8SURBVK7JfPGquruqDp/M1xwijiVV9Z5x7vMdVXXzePYpSZIkSZpaTGqsvSuA7YY6meQvklzbzOz4cpJpTfkjSU5OsjTJD5LsnmRxkp8l+a9NnW2SXJHk+ubnFf3K/2imQ5LNklyU5IYkXwbS79wJzcySm5K8r18/tzSzI25K8o0kByS5MsltSXZv6u2e5MdNvz9OsmNT/u+zRJKclOSMfu9huGTH9CRnJVme5Nt9s12a9nP7/Y4+neTGJFcnmb2W/yaSJEmSpCnMpMZaSDIdeC2wYojzLwLeDOxVVXOA1cBbmtMbAYurajfgYeBTwIHAocAnmjqrgAOr6mVNP8Mt9TgR+FFVvRT4LvD8Jo7dgLcDLwf2AI5J8tKmzXbA54FdgBcCfw68EvgfwEeaOrcA+zT9fgz4X0O8/guBg4HdgROTzFhDrDsCC6tqF+C3wH8bpM5GwNVVtStwOXDMYB0lWZBkSZIl9z/6wBpeUpIkSZI0FUzvdgDruA2SLGuOrwC+OkS9/YHdgOuSAGxAJ1EBnWUr32+OVwCPVdUTSVYA2zTlM4BTk/QlRHYYJq59gMMAqup7Sfq+4b8SWFRVjwIkOQ/Ym07i4+dVtaIpXwlcUlU1II5NgLOSbA9UE9dgvldVjwGPJVkFzAbuGqLunVV1ZXP8/4D3AJ8bUOdxoG+/kKV0kj5/pKoWAgsBdt16pxri9SRJkiRJU4RJjTX7fTPzYjgBzqqqDw9y7omq6vsC/hTwGEBVPdXMAAH478C9wK50Zs/8YS1ec7Av9RmkrM9j/Y6f6vf8KZ4eB58EflhVhybZBli8Fn2tZs3jaGCcg8Xd/3c0XH+SJEmSJAEuPxkvlwCHJ5kFkGTTJC8YQftNgHuq6ingrcC0YepfTrO8JclrgZn9yt+QZMMkG9FZ4nLFCOP41+b4qBG0W5PnJ9mzOT4C+NE49StJkiRJmuJMaoyD5i4eHwUuSrIcuBjYcgRdfBE4MsnVdJaePDpM/Y8D+yS5HjgI+GUTx/XAmcC1wDXA6VV1wwji+Azwv5NcyfCJlbX1EzrvbTmwKXDaOPUrSZIkSZri8vSsf6k9dt16p7ro/X/f7TCkSTH7fbt1OwRJkiRpUiVZWlVzh6vn3gVqpRmzN/SLniRJkiRNcSY1RiDJZnT2zxho/6q6b7LjWVf4e5EkSZIkdYNJjRFovqCvzd1QphR/L5IkSZKkbnCjUEmSJEmS1ErO1FArPbHqEe79/JXdDkOaFLPfu1e3Q5AkSZLWSc7UkCRJkiRJrWRSQ5IkSZIktZJJDUmSJEmS1EomNXpQki2SnJPkp0luTnJBkh2SfD/Jg0nOX4s+9k6yMsmyJBtMRtySJEmSJI2ESY0ekyTAImBxVf1JVe0EfASYDXwWeOtadvUW4HNVNaeqfj8x0UqSJEmSNHomNXrPfsATVfWlvoKqWlZVV1TVJcDDw3WQ5B3Am4CPJflGOk5tZn18r5n5cfga2u+W5LIkS5NcmGTLpnxxkpOTXJvkX5Ls3ZRPS/K5JCuSLE9y/Fh/CZIkSZKk3uctXXvPzsDSsXRQVacneSVwflV9O8lhwI7AS+jM+LgZOGOwtklmAF8AXl9Vv07yZuDTwF82VaZX1e5JDgFOBA4AFgDbAi+tqieTbDpE3wuaumw1c/ZY3qIkSZIkqQeY1NDa2Ac4u6pWA3cnuXQNdXekk1i5uLMShmnAPf3On9c8LgW2aY4PAL5UVU8CVNX9g3VcVQuBhQC7Pv+FNap3IkmSJEnqGSY1es9KYMilIWOwtkmEACuras8hzj/WPK7m6fGXEfQvSZIkSRLgnhq96FJg/STH9BUkmZfkVWPo83JgfrP3xZZ09u0Yyq3Ac5Ls2bz2jCQvHqb/i4B3JZnetBl0+YkkSZIkSf2Z1OgxVVXAocCBzS1dVwIn0Vk2cgXwLWD/JHclOXgtu10E3AasAE4DLlvD6z9OZ6bIyUluBJYBrxim/9OBXwLLmzZ/vpZxSZIkSZKmMJef9KCqupvO3UsG2nsEfRzV77iA4/qeJzlzmLbL6OzDMbB8337Hv6HZU6PZS+OE5keSJEmSpLXiTA1JkiRJktRKztSY4pIsonM71f4+WFUXDtWmbxbHaNqOlxmzNmb2e/ea6JeRJEmSJK3DTGpMcVV1aDfaSpIkSZI0Vi4/kSRJkiRJreRMDbXSk6seZtUXLu12GNKkmHX8q7sdgiRJkrROcqaGJEmSJElqJZMakiRJkiSplUxqaNIk2SLJOUl+muTmJBck2SHJ95M8mOT8bscoSZIkSWoP99TQpEgSYBFwVlXNb8rmALOBzwIbAu/sXoSSJEmSpLYxqaHJsh/wRFV9qa+gqpb1HSfZtxtBSZIkSZLay+Unmiw7A0u7HYQkSZIkqXeY1FBrJFmQZEmSJfc98mC3w5EkSZIkdZlJDU2WlcBuY+mgqhZW1dyqmrvZxs8ep7AkSZIkSW1lUkOT5VJg/STH9BUkmZfkVV2MSZIkSZLUYiY1NCmqqoBDgQObW7quBE4C7k5yBfAtYP8kdyU5uIuhSpIkSZJawrufaNJU1d3AmwY5tfdkxyJJkiRJaj9nakiSJEmSpFYyqSFJkiRJklrJ5Sdqpemznsms41/d7TAkSZIkSV3kTA1JkiRJktRKJjUkSZIkSVIrmdSQJEmSJEmt5J4aaqUnVz3EqlMv6HYY0oSbddwh3Q5BkiRJWmc5U0OSJEmSJLWSSQ1JkiRJktRKJjU0YkkWJzl4QNn7knwtydIky5KsTPKufuc/neTOJI8MaLd+knOT3J7kmiTbTM67kCRJkiS1nUkNjcbZwPwBZfOBM4FXVNUc4OXAh5I8tzn//4HdB+nraOCBqtoO+Fvg5AmJWJIkSZLUc0xqaDS+DbwuyfoAzeyK5wKXV9VjTZ316Te+qurqqrpnkL5eD5zVr9/9k2SC4pYkSZIk9RCTGhqxqroPuBZ4TVM0Hzi3qirJ1kmWA3cCJ1fV3cN097ymLlX1JPAQsNnERC5JkiRJ6iUmNTRa/ZegzG+eU1V3VtUuwHbAkUlmD9PPYLMyatCKyYIkS5Isue+Rh0YZtiRJkiSpV5jU0Gh9h85SkZcBG1TV9f1PNjM0VgJ7D9PPXcDWAEmmA5sA9w9WsaoWVtXcqpq72cabjDV+SZIkSVLLmdTQqFTVI8Bi4AyaWRpJtkqyQXM8E9gLuHWYrr4LHNkcHw5cWlWDztSQJEmSJKk/kxoai7OBXYFzmucvAq5JciNwGfC5qloBkOQzSe4CNkxyV5KTmjZfBTZLcjtwAvChyXwDkiRJkqT2mt7tANReVbWIfntiVNXFwC5D1P0A8IFByv8A/NlExShJkiRJ6l3O1JAkSZIkSa1kUkOSJEmSJLWSy0/UStNnbcKs4w7pdhiSJEmSpC5ypoYkSZIkSWolkxqSJEmSJKmVXH6iVnpy1YOs+rvzuh2GNOFmHXtYt0OQJEmS1lnO1JAkSZIkSa1kUkOSJEmSJLWSSY0ekmR1kmVJVia5MckJSUb1b5xkbpJTRtl2cZK5w9R5TpJrktyQZO/RvI4kSZIkaWpzT43e8vuqmgOQZBbwTWAT4MSRdlRVS4Al4xvef7A/cEtVHTmBryFJkiRJ6mHO1OhRVbUKWAAcl45pST6b5Loky5O8EyDJuUkO6WuX5Mwkb0yyb5Lzm7KNk3wtyYqm7Rub8oOSXJXk+iTfSrLxwDiSPJLk083MkauTzE4yB/gMcEgzs2SDJEc0/d+U5OTJ+B1JkiRJktrNpEYPq6qf0fk3ngUcDTxUVfOAecAxSbYFzgHeDJBkPTozKC4Y0NX/bNq+pKp2AS5NsjnwUeCAqnoZnVkdJwwSxkbA1VW1K3A5cExVLQM+BpzbzCyZCZwMvBqYA8xL8obx+j1IkiRJknqTy096X5rHg4BdkhzePN8E2B74Z+CUJOsDrwEur6rfJ+nfxwHA/L4nVfVAktcBOwFXNnXXA64a5PUfB85vjpcCBw5SZx6wuKp+DZDkG8A+wHf+wxtJFtCZfcJWMzcf9o1LkiRJknqbSY0eluQ/A6uBVXSSG8dX1YWD1FsMHExnxsbZg3UF1CBlF1fVEcOE8URV9bVdzeBjLoOU/ZGqWggsBJjz/O0GxiNJkiRJmmJcftKjkjwH+BJwapNUuBB4d5IZzfkdkmzUVD8HeDuwd1NvoIuA4/r1PRO4GtgryXZN2YZJdhhluNcAr0qyeZJpwBHAZaPsS5IkSZI0RZjU6C0b9N3SFfgBnWTEx5tzpwM3A9cnuQn4Mk/PmriIznKPH1TV44P0+ylgZrOJ543Afs1SkaOAs5Msp5PkeOFogq6qe4APAz8EbgSur6p/Gk1fkiRJkqSpI0+vDJDaY87zt6uLPviZbochTbhZxx7W7RAkSZKkSZdkaVXNHa6eMzUkSZIkSVIrmdSQJEmSJEmt5N1P1ErTZz3bafmSJEmSNMU5U0OSJEmSJLWSG4WqlZI8DNza7Tg0ZW0O/KbbQWjKcvypmxx/6ibHn7rJ8Tf5XlBVzxmukstP1Fa3rs1OuNJESLLE8aducfypmxx/6ibHn7rJ8bfucvmJJEmSJElqJZMakiRJkiSplUxqqK0WdjsATWmOP3WT40/d5PhTNzn+1E2Ov3WUG4VKkiRJkqRWcqaGJEmSJElqJZMaapUkr0lya5Lbk3yo2/GodyS5I8mKJMuSLGnKNk1ycZLbmseZTXmSnNKMw+VJXtavnyOb+rclObJb70frtiRnJFmV5KZ+ZeM23pLs1ozn25u2mdx3qHXZEOPvpCT/2lwDlyU5pN+5Dzdj6dYkB/crH/QzOcm2Sa5pxuW5SdabvHendV2SrZP8MMlPkqxM8t6m3GugJtwaxp/XwBYzqaHWSDIN+DvgtcBOwBFJdupuVOox+1XVnH636/oQcElVbQ9c0jyHzhjcvvlZAJwGnT/IgBOBlwO7Ayf2/VEmDXAm8JoBZeM53k5r6va1G/hamtrOZPAx8bfNNXBOVV0A0HzOzgde3LT5YpJpw3wmn9z0tT3wAHD0hL4btc2TwPur6kXAHsCxzdjxGqjJMNT4A6+BrWVSQ22yO3B7Vf2sqh4HzgFe3+WY1NteD5zVHJ8FvKFf+der42rg2Um2BA4GLq6q+6vqAeBi/ENKg6iqy4H7BxSPy3hrzj2rqq6qzsZZX+/XlzTU+BvK64Fzquqxqvo5cDudz+NBP5Ob/xF/NfDtpn3/sSxRVfdU1fXN8cPAT4Dn4TVQk2AN428oXgNbwKSG2uR5wJ39nt/Fmi9C0kgUcFGSpUkWNGWzq+oe6HwIArOa8qHGomNUYzFe4+15zfHAcmk4xzXT+8/o9z/eIx1/mwEPVtWTA8qlP5JkG+ClwDV4DdQkGzD+wGtga5nUUJsMth7S2/dovOxVVS+jM43w2CT7rKHuUGPRMaqJMNLx5jjUaJwG/AkwB7gH+D9NueNPEyLJxsA/Au+rqt+uqeogZY5Bjckg489rYIuZ1FCb3AVs3e/5VsDdXYpFPaaq7m4eVwGL6EwrvLeZxkrzuKqpPtRYdIxqLMZrvN3VHA8sl4ZUVfdW1eqqegr4Cp1rIIx8/P2GzvKA6QPKpX+XZAadL5TfqKrzmmKvgZoUg40/r4HtZlJDbXIdsH2zo/B6dDbt+W6XY1IPSLJRkmf2HQMHATfRGV99u6kfCfxTc/xd4G3Njux7AA81U2UvBA5KMrOZtnhQUyatjXEZb825h5Ps0aztfVu/vqRB9X2ZbBxK5xoInfE3P8n6Sbals+nitQzxmdzsYfBD4PCmff+xLNFcl74K/KSq/qbfKa+BmnBDjT+vge02ffgq0rqhqp5MchydD7FpwBlVtbLLYak3zAYWNXd8mw58s6q+n+Q64B+SHA38Evizpv4FwCF0Nov6HfB2gKq6P8kn6XzQAXyiqtZ2Mz5NIUnOBvYFNk9yF50d/P+a8Rtv76Zzh4sNgH9ufiRgyPG3b5I5dKZJ3wG8E6CqVib5B+BmOncNOLaqVjf9DPWZ/EHgnCSfAm6g8wVC6rMX8FZgRZJlTdlH8BqoyTHU+DvCa2B7pZNMkiRJkiRJaheXn0iSJEmSpFYyqSFJkiRJklrJpIYkSZIkSWolkxqSJEmSJKmVTGpIkiRJkqRWMqkhSZIkSZJayaSGJElSSySZ1u0YJElal5jUkCRJmiBJvpNkaZKVSRYkeXeSz/Q7f1SSLzTHf5Hk2iTLkny5L4GR5JEkn0hyDbBnko8luS7JTUkWJklTb16S5UmuSvLZJDc15dOa59c159/ZhV+FJEkTwqSGJEnSxPnLqtoNmAu8BzgPOKzf+TcD5yZ5UXO8V1XNAVYDb2nqbATcVFUvr6ofAadW1byq2hnYAHhdU+9rwLuqas+mfZ+jgYeqah4wDzgmybYT8WYlSZps07sdgCRJUg97T5JDm+OtgW2BnyXZA7gN2BG4EjgW2A24rpl4sQGwqmm3GvjHfn3ul+QDwIbApsDKJFcAz6yqHzd1vsnTyY6DgF2SHN483wTYHvj5eL5RSZK6waSGJEnSBEiyL3AAsGdV/S7JYuAZwLnAm4BbgEVVVc0SkrOq6sODdPWHqlrd9PkM4IvA3Kq6M8lJTZ9ZUyjA8VV14fi8M0mS1h0uP5EkSZoYmwAPNAmNFwJ7NOXnAW8AjqCT4AC4BDg8ySyAJJsmecEgfT6jefxNko2BwwGq6gHg4WYGCMD8fm0uBN6dZEbT9w5JNhqXdyhJUpc5U0OSJGlifB94V5LlwK3A1dBJQCS5Gdipqq5tym5O8lHgoiT/CXiCzpKUX/TvsKoeTPIVYAVwB3Bdv9NHA19J8iiwGHioKT8d2Aa4vpkR8ms6SRVJklovVdXtGCRJkjRGSTauqkea4w8BW1bVe7scliRJE8qZGpIkSb3hvyT5MJ2/734BHNXdcCRJmnjO1JAkSZIkSa3kRqGSJEmSJKmVTGpIkiRJkqRWMqkhSZIkSZJayaSGJEmSJElqJZMakiRJkiSplUxqSJIkSZKkVvo3WqHP0UtMoJIAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<Figure size 1152x1152 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "feature_importances['average'] = feature_importances[[f'fold_{fold_n + 1}' for fold_n in range(NFOLDS)]].mean(axis=1)\n",
    "feature_importances.to_csv('feature_importances.csv')\n",
    "\n",
    "plt.figure(figsize=(16, 16))\n",
    "sns.barplot(data=feature_importances.sort_values(by='average', ascending=False).head(50), x='average', y='feature');\n",
    "plt.title('50 TOP feature importance over {} folds average'.format(NFOLDS));"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}