From ccd69532091af7fcf0c1f45421a78a91ceda8968 Mon Sep 17 00:00:00 2001
From: Dmitri Soshnikov <dmitri@soshnikov.com>
Date: Mon, 23 May 2022 12:12:45 +0300
Subject: [PATCH] Add NER lesson and lab
---
README.md | 2 +-
lessons/5-NLP/19-NER/NER-TF.ipynb | 481 ++++++++++++++++++++++++
lessons/5-NLP/19-NER/README.md | 69 +++-
lessons/5-NLP/19-NER/images/bot-ner.png | Bin 0 -> 38663 bytes
lessons/5-NLP/19-NER/lab/README.md | 47 +++
lessons/6-Other/22-DeepRL/README.md | 2 +-
6 files changed, 598 insertions(+), 3 deletions(-)
create mode 100644 lessons/5-NLP/19-NER/NER-TF.ipynb
create mode 100644 lessons/5-NLP/19-NER/images/bot-ner.png
create mode 100644 lessons/5-NLP/19-NER/lab/README.md
diff --git a/README.md b/README.md
index f10e688..b931f87 100644
--- a/README.md
+++ b/README.md
@@ -80,7 +80,7 @@ For a gentle introduction to *AI in the Cloud* topics you may consider taking th
<tr><td>16</td><td>Recurrent Neural Networks</td><td><a href="lessons/5-NLP/16-RNN/README.md">Text</a></td><td><a href="lessons/5-NLP/16-RNN/RNNPyTorch.ipynb">PyTorch</a></td><td><a href="lessons/5-NLP/16-RNN/RNNTF.ipynb">TensorFlow</a></td><td></td></tr>
<tr><td>17</td><td>Generative Recurrent Networks</td><td><a href="lessons/5-NLP/17-GenerativeNetworks/README.md">Text</a></td><td><a href="lessons/5-NLP/17-GenerativeNetworks/GenerativePyTorch.md">PyTorch</a></td><td><a href="lessons/5-NLP/17-GenerativeNetworks/GenerativeTF.md">TensorFlow</a></td><td><a href="lessons/5-NLP/17-GenerativeNetworks/lab/README.md">Lab</a></td></tr>
<tr><td>18</td><td>Transformers. BERT.</td><td><a href="lessons/5-NLP/18-Transformers/README.md">Text</a></td><td><a href="lessons/5-NLP/18-Transformers/TransformersPyTorch.md">PyTorch</a></td><td><a href="lessons/5-NLP/18-Transformers/TransformersTF.md">TensorFlow</a></td><td></td></tr>
-<tr><td>19</td><td>Named Entity Recognition</td><td>Text</td><td>PyTorch</td><td>TensorFlow</td><td></td></tr>
+<tr><td>19</td><td>Named Entity Recognition</td><td><a href="lessons/5-NLP/19-NER/README.md">Text</a></td><td></td><td><a href="lessons/5-NLP/19-NER/NER-TF.ipynb">TensorFlow</a></td><a href="lessons/5-NLP/19-NER/lab/README.md">Lab</a><td></td></tr>
<tr><td>20</td><td>Large Language Models, Prompt Programming and Few-Shot Tasks</td><td>Text</td><td>PyTorch</td><td>TensorFlow</td><td></td></tr>
<tr><td>VI</td><td colspan="4"><b>Other AI Techniques</b></td><td></td></tr>
<tr><td>21</td><td>Genetic Algorithms</td><td><a href="lessons/6-Other/21-GeneticAlgorithms/README.md">Text</a><td colspan="2"><a href="lessons/6-Other/21-GeneticAlgorithms/Genetic.ipynb">Notebook</a></td><td></td></tr>
diff --git a/lessons/5-NLP/19-NER/NER-TF.ipynb b/lessons/5-NLP/19-NER/NER-TF.ipynb
new file mode 100644
index 0000000..46ed324
--- /dev/null
+++ b/lessons/5-NLP/19-NER/NER-TF.ipynb
@@ -0,0 +1,481 @@
+{
+ "cells": [
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "# Named Entity Recognition (NER)\n",
+ "\n",
+ "This notebook is from [AI for Beginners Curriculum](http://aka.ms/ai-beginners).\n",
+ "\n",
+ "In this example, we will learn how to train NER model on [Annotated Corpus for Named Entity Recognition](https://www.kaggle.com/datasets/abhinavwalia95/entity-annotated-corpus) Dataset from Kaggle. Before procedding, please donwload [ner_dataset.csv](https://www.kaggle.com/datasets/abhinavwalia95/entity-annotated-corpus?resource=download&select=ner_dataset.csv) file into current directory."
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 62,
+ "metadata": {},
+ "outputs": [],
+ "source": [
+ "import pandas as pd\n",
+ "from tensorflow import keras\n",
+ "import numpy as np"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "## Preparing the Dataset \n",
+ "\n",
+ "We will start by reading the dataset into a dataframe. If you want to learn more about using Pandas, visit a [lesson on data processing](https://github.com/microsoft/Data-Science-For-Beginners/tree/main/2-Working-With-Data/07-python) in our [Data Science for Beginners](http://aka.ms/datascience-beginners)"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 3,
+ "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>Sentence #</th>\n",
+ " <th>Word</th>\n",
+ " <th>POS</th>\n",
+ " <th>Tag</th>\n",
+ " </tr>\n",
+ " </thead>\n",
+ " <tbody>\n",
+ " <tr>\n",
+ " <th>0</th>\n",
+ " <td>Sentence: 1</td>\n",
+ " <td>Thousands</td>\n",
+ " <td>NNS</td>\n",
+ " <td>O</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <th>1</th>\n",
+ " <td>NaN</td>\n",
+ " <td>of</td>\n",
+ " <td>IN</td>\n",
+ " <td>O</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <th>2</th>\n",
+ " <td>NaN</td>\n",
+ " <td>demonstrators</td>\n",
+ " <td>NNS</td>\n",
+ " <td>O</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <th>3</th>\n",
+ " <td>NaN</td>\n",
+ " <td>have</td>\n",
+ " <td>VBP</td>\n",
+ " <td>O</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <th>4</th>\n",
+ " <td>NaN</td>\n",
+ " <td>marched</td>\n",
+ " <td>VBN</td>\n",
+ " <td>O</td>\n",
+ " </tr>\n",
+ " </tbody>\n",
+ "</table>\n",
+ "</div>"
+ ],
+ "text/plain": [
+ " Sentence # Word POS Tag\n",
+ "0 Sentence: 1 Thousands NNS O\n",
+ "1 NaN of IN O\n",
+ "2 NaN demonstrators NNS O\n",
+ "3 NaN have VBP O\n",
+ "4 NaN marched VBN O"
+ ]
+ },
+ "execution_count": 3,
+ "metadata": {},
+ "output_type": "execute_result"
+ }
+ ],
+ "source": [
+ "df = pd.read_csv('ner_dataset.csv',encoding='unicode-escape')\n",
+ "df.head()"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "Let's get unique tags and create lookup dictionaries that we can use to convert tags into class numbers:"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 4,
+ "metadata": {},
+ "outputs": [
+ {
+ "data": {
+ "text/plain": [
+ "array(['O', 'B-geo', 'B-gpe', 'B-per', 'I-geo', 'B-org', 'I-org', 'B-tim',\n",
+ " 'B-art', 'I-art', 'I-per', 'I-gpe', 'I-tim', 'B-nat', 'B-eve',\n",
+ " 'I-eve', 'I-nat'], dtype=object)"
+ ]
+ },
+ "execution_count": 4,
+ "metadata": {},
+ "output_type": "execute_result"
+ }
+ ],
+ "source": [
+ "tags = df.Tag.unique()\n",
+ "tags"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 8,
+ "metadata": {},
+ "outputs": [
+ {
+ "data": {
+ "text/plain": [
+ "'O'"
+ ]
+ },
+ "execution_count": 8,
+ "metadata": {},
+ "output_type": "execute_result"
+ }
+ ],
+ "source": [
+ "id2tag = dict(enumerate(tags))\n",
+ "tag2id = { v : k for k,v in id2tag.items() }\n",
+ "\n",
+ "id2tag[0]"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "Now we need to do the same with vocabulary. For simplicity, we will create vocabulary without taking word frequency into account; in real life you might want to use Keras vectorizer, and limit the number of words."
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 14,
+ "metadata": {},
+ "outputs": [],
+ "source": [
+ "vocab = set(df['Word'].apply(lambda x: x.lower()))\n",
+ "id2word = { i+1 : v for i,v in enumerate(vocab) }\n",
+ "id2word[0] = '<UNK>'\n",
+ "vocab.add('<UNK>')\n",
+ "word2id = { v : k for k,v in id2word.items() }"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "We need to create a dataset of sentences for training. Let's loop through the original dataset and separate all individual sentences into `X` (lists of words) and `Y` (list of tokens):"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 41,
+ "metadata": {},
+ "outputs": [],
+ "source": [
+ "X,Y = [],[]\n",
+ "s,t = [],[]\n",
+ "for i,row in df[['Sentence #','Word','Tag']].iterrows():\n",
+ " if pd.isna(row['Sentence #']):\n",
+ " s.append(row['Word'])\n",
+ " t.append(row['Tag'])\n",
+ " else:\n",
+ " if len(s)>0:\n",
+ " X.append(s)\n",
+ " Y.append(t)\n",
+ " s,t = [row['Word']],[row['Tag']]\n",
+ "X.append(s)\n",
+ "Y.append(t)\n"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "We will now vectorize all words and tokens:"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 93,
+ "metadata": {},
+ "outputs": [
+ {
+ "data": {
+ "text/plain": [
+ "([10386,\n",
+ " 23515,\n",
+ " 4134,\n",
+ " 29620,\n",
+ " 7954,\n",
+ " 13583,\n",
+ " 21193,\n",
+ " 12222,\n",
+ " 27322,\n",
+ " 18258,\n",
+ " 5815,\n",
+ " 15880,\n",
+ " 5355,\n",
+ " 25242,\n",
+ " 31327,\n",
+ " 18258,\n",
+ " 27067,\n",
+ " 23515,\n",
+ " 26444,\n",
+ " 14412,\n",
+ " 358,\n",
+ " 26551,\n",
+ " 5011,\n",
+ " 30558],\n",
+ " [0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 2, 0, 0, 0, 0, 0])"
+ ]
+ },
+ "execution_count": 93,
+ "metadata": {},
+ "output_type": "execute_result"
+ }
+ ],
+ "source": [
+ "def vectorize(seq):\n",
+ " return [word2id[x.lower()] for x in seq]\n",
+ "\n",
+ "def tagify(seq):\n",
+ " return [tag2id[x] for x in seq]\n",
+ "\n",
+ "Xv = list(map(vectorize,X))\n",
+ "Yv = list(map(tagify,Y))\n",
+ "\n",
+ "Xv[0], Yv[0]"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "For simplicity, we will pad all sentences with 0 tokens to the maximum length. In real life, we might want to use more clever strategy, and pad sequences only within one minibatch."
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 51,
+ "metadata": {},
+ "outputs": [],
+ "source": [
+ "X_data = keras.preprocessing.sequence.pad_sequences(Xv,padding='post')\n",
+ "Y_data = keras.preprocessing.sequence.pad_sequences(Yv,padding='post')"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "## Defining Token Classification Network\n",
+ "\n",
+ "We will use two-layer bidirectional LSTM network for token classification. In order to apply dense classifier to each of the output of the last LSTM layer, we will use `TimeDistributed` construction, which replicates the same dense layer to each of the outputs of LSTM at each step: "
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 94,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "Model: \"sequential_3\"\n",
+ "_________________________________________________________________\n",
+ " Layer (type) Output Shape Param # \n",
+ "=================================================================\n",
+ " embedding_4 (Embedding) (None, 104, 300) 9545400 \n",
+ " \n",
+ " bidirectional_6 (Bidirectio (None, 104, 200) 320800 \n",
+ " nal) \n",
+ " \n",
+ " bidirectional_7 (Bidirectio (None, 104, 200) 240800 \n",
+ " nal) \n",
+ " \n",
+ " time_distributed_3 (TimeDis (None, 104, 17) 3417 \n",
+ " tributed) \n",
+ " \n",
+ "=================================================================\n",
+ "Total params: 10,110,417\n",
+ "Trainable params: 10,110,417\n",
+ "Non-trainable params: 0\n",
+ "_________________________________________________________________\n"
+ ]
+ }
+ ],
+ "source": [
+ "maxlen = X_data.shape[1]\n",
+ "vocab_size = len(vocab)\n",
+ "num_tags = len(tags)\n",
+ "model = keras.models.Sequential([\n",
+ " keras.layers.Embedding(vocab_size, 300, input_length=maxlen),\n",
+ " keras.layers.Bidirectional(keras.layers.LSTM(units=100, activation='tanh', return_sequences=True)),\n",
+ " keras.layers.Bidirectional(keras.layers.LSTM(units=100, activation='tanh', return_sequences=True)),\n",
+ " keras.layers.TimeDistributed(keras.layers.Dense(num_tags, activation='softmax'))\n",
+ "])\n",
+ "model.compile(loss='sparse_categorical_crossentropy',optimizer='adam',metrics=['acc'])\n",
+ "model.summary()"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "Note here that we are explicity specifying `maxlen` for our dataset - in case we want the network to be able to handle variable length sequences, we need to be a bit more clever when defining the network.\n",
+ "\n",
+ "Let's now train the model. For speed, we will only train for one epoch, but you may try training for longer time. Also, you may want to separate some part of the dataset as training dataset, to observe validation accuracy."
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 57,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "1499/1499 [==============================] - 740s 488ms/step - loss: 0.0667 - acc: 0.9841\n"
+ ]
+ },
+ {
+ "data": {
+ "text/plain": [
+ "<keras.callbacks.History at 0x16f0bb2a310>"
+ ]
+ },
+ "execution_count": 57,
+ "metadata": {},
+ "output_type": "execute_result"
+ }
+ ],
+ "source": [
+ "model.fit(X_data,Y_data)"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "## Testing the Result\n",
+ "\n",
+ "Let's now see how our entity recognition model works on a sample sentence: "
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 91,
+ "metadata": {},
+ "outputs": [],
+ "source": [
+ "sent = 'John Smith went to Paris to attend a conference in cancer development institute'\n",
+ "words = sent.lower().split()\n",
+ "v = keras.preprocessing.sequence.pad_sequences([[word2id[x] for x in words]],padding='post',maxlen=maxlen)\n",
+ "res = model(v)[0]"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 92,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "john -> B-per\n",
+ "smith -> I-per\n",
+ "went -> O\n",
+ "to -> O\n",
+ "paris -> B-geo\n",
+ "to -> O\n",
+ "attend -> O\n",
+ "a -> O\n",
+ "conference -> O\n",
+ "in -> O\n",
+ "cancer -> B-org\n",
+ "development -> I-org\n",
+ "institute -> I-org\n"
+ ]
+ }
+ ],
+ "source": [
+ "r = np.argmax(res.numpy(),axis=1)\n",
+ "for i,w in zip(r,words):\n",
+ " print(f\"{w} -> {id2tag[i]}\")"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "## Takeaway\n",
+ "\n",
+ "Even simple LSTM model shows reasonable results at NER. However, to get much better results, you may want to use large pre-trained language models such as BERT. Training BERT for NER using Huggingface Transformers library is described [here](https://huggingface.co/course/chapter7/2?fw=pt)."
+ ]
+ }
+ ],
+ "metadata": {
+ "interpreter": {
+ "hash": "16af2a8bbb083ea23e5e41c7f5787656b2ce26968575d8763f2c4b17f9cd711f"
+ },
+ "kernelspec": {
+ "display_name": "Python 3.8.12 ('py38')",
+ "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": "3.8.12"
+ },
+ "orig_nbformat": 4
+ },
+ "nbformat": 4,
+ "nbformat_minor": 2
+}
diff --git a/lessons/5-NLP/19-NER/README.md b/lessons/5-NLP/19-NER/README.md
index f6e0339..380b458 100644
--- a/lessons/5-NLP/19-NER/README.md
+++ b/lessons/5-NLP/19-NER/README.md
@@ -1 +1,68 @@
-Placeholder
\ No newline at end of file
+# Named Entity Recognition
+
+Up to now, we have mostly been concentrating on one NLP task - classification. However, there are also other NLP tasks that can be accomplished with neural networks. One of those tasks is **[Named Entity Recognition](https://en.wikipedia.org/wiki/Named-entity_recognition)** (NER), which deals with recognizing specific entities within text, such as places, person names, date-time intervals, chemical formulae and so on.
+
+## [Pre-lecture quiz](https://black-ground-0cc93280f.1.azurestaticapps.net/quiz/119)
+
+## Example of Using NER
+
+Suppose you want to develop a natural language chat bot, similar to Amazon Alexa or Google Assistant. The way intelligent chat bots work is to *understand* what the user want by doing text classification on the input sentence. The result of this classification is so-called **intent**, which determines what a chat bot should do.
+
+
+
+> Image by author
+
+However, a user may provide some parameters as part of the phrase. For example, when asking for a weather, she may specify a location or date. A bot should be able to understand those entities, and fill in the parameter slots accordingly before performing the action. This is exactly when NER comes in.
+
+Another example would be [analyzing scientific medical papers](https://soshnikov.com/science/analyzing-medical-papers-with-azure-and-text-analytics-for-health/), one of the main things we need to look for are specific medical terms, such as diseases and medical substances. While a small number of diseases can probably be extracted using substring search, more complex entities, such as chemical compounds and medication names, need more complex approach.
+
+## NER as Token Classification
+
+NER models are essentially **token classification models**, because for each of the input tokens we need to decide whether it belongs to an entity or not, and if it does - to which entity class.
+
+Consider the following paper title:
+
+**Tricuspid valve regurgitation** and **lithium carbonate** **toxicity** in a newborn infant.
+
+Entities here are:
+* Tricuspid valve regurgitation is a disease (`DIS`)
+* Lithium carbonate is a chemical substance (`CHEM`)
+* Toxicity is also a disease (`DIS`)
+
+Notice that one entity can span several tokens. And, as in this case, we need to distinguish between two consecutive entities. Thus it is common to use two classes for each entity - one specifying the first token of the entity (often `B-` prefix is used, for beginning), and another - the continuation of an entity (`I-`, for inner token). We also use `O` as a class to represent all other tokens. Such token tagging is called [BIO tagging](https://en.wikipedia.org/wiki/Inside%E2%80%93outside%E2%80%93beginning_(tagging)) (or IOB). When tagged, our title will look like this:
+
+Token | Tag
+------|-----
+Tricuspid | B-DIS
+valve | I-DIS
+regurgitation | I-DIS
+and | O
+lithium | B-CHEM
+carbonate | I-CHEM
+toxicity | B-DIS
+in | O
+a | O
+newborn | O
+infant | O
+. | O
+
+Since we need to build one-to-one correspondence between tokens and this classes, we can train rightmost **many-to-many** neural network model from this picture:
+
+
+
+> *Image from blog post [Unreasonable Effectiveness of Recurrent Neural Networks](http://karpathy.github.io/2015/05/21/rnn-effectiveness/) by [Andrej Karpaty](http://karpathy.github.io/). NER token classification models correspond to right-most network architecture on this picture.*
+
+## Training NER models
+
+Since NER model is essentially a token classification model, we can use RNNs that we are already familiar with for this task. In this case, each block of recurrent network will return the token ID. Example notebook shows how to train LSTM for token classification.
+
+## ✍️ Example Notebooks: NER
+
+Continue your learning in the following notebooks:
+
+* [NER with TensorFlow](NER-TF.ipynb)
+
+## [Assignment](lab/README.md)
+
+In the assignment for this lesson, you will have to train medical entity recognition model. You can start with training LSTM model as described in this lesson, and proceed with using BERT transformer model. Read [the instructions](lab/README.md) to get all the details.
+
diff --git a/lessons/5-NLP/19-NER/images/bot-ner.png b/lessons/5-NLP/19-NER/images/bot-ner.png
new file mode 100644
index 0000000000000000000000000000000000000000..2b5c982566076065559362a5af407a758a5a4b35
GIT binary patch
literal 38663
zcmeAS@N?(olHy`uVBq!ia0y~y;H_t1V5?_gW?*2LbG<^Afq{V~-O<;Pfnj4m_n$;o
z1_lPk;vjb?hIQv;UNSH+a29w(7Besy9ROiQjg+X{3=9k^0X`wF?gc(33|PQF28RD&
z;r|Q_X3M78F)%Q2l?3?(GuZ5@d=x1+bAQ9bq#v9N3<?aME{-7;x8B@6n=gD_ianu6
zHacnQCH^IAG8&>!u5P##_Nj|oL}mRYv92jo7lwGO$<WuG^k0gDbK;-ydnS+9&#XM3
z^2zbit@E25j~q8n?ys$TKDYm|_jIt`gP4ej$~V5p!SFBEt-WlI;D3j+IRzhgt@5?b
z{=B@LTOi?2a?H81bGuJ5oWAv@aytX#vHgYz>?*c1Ft%~=I5ZSW$%;xeFa$R;au~2k
zG%z?fFvgTLnC8bG__p1goqKn_6%)?`8KyORD+TwyIQ=i9konCP(TLw?i`eWH4)i^}
z{*QGEW3M^;ZM_3RjAgkdt8T?I@dzYD{w!rYzH!}GcP1Utgx-f5k_o+MTkGyKFqZNB
z?Y0p7taSU^(YGD`#e8pWJ)A$?_y;3n*_)+viW~MuePOV6W!+}x#eU-uyS>5zmK#ZI
zH?$eqI1Cuo4luCHV9<WWVZgw+j7d15fk8NdiN~RV(}0mp;ec3AaRLLA53AY%21d05
zAfG^Fb)}wz^l|w#Ff#cxtOT(ko)w+Jz|1lO#MW_Xu>Jk%xqAFus{`|FfBrn#|JC~3
zx9R@-)@G~&+q0T)!?*9|{Lyd!Sz5R@n7;e6d1=jO@3p$WOKqdGHdThIuVu0>Z7}6)
z<M$8q|1HJGrg&U>OHJ8^YrD?Q|7&~b=c^YBuTO5&x6<)#sQv%#f{N#^#hr6|4_)5e
zmVdl|-(0Ql68^t0FYS5f{x?az`q{M|zwgX3{lEFbr>JX1Gi<G*UuKKmn8|wM`OO)r
zCGT%X&HaCQ_FwkbOPBS2Jn=liaC`RNNB94ouv6`?TXTO~mdlkot9uRZ+iR~TdNSo$
z{69Xu_@uE^uFKu@$c@_B`OoC;-PFqW%&ueYeOU8Z{*&%@VeQ)GOuN6Boc@$2^6uw@
z|I3tqgzqZ;{c3l{-kZkt=gt4m=wEODto&TvhSU0MpO`;b9{E|?{mHXavwybQ-|FiA
zk@wY|FX98kYmuO&*Dj3Lv~A={)IT#`lY2J1e(w~<vPoQP!av?ybG7#Vr{rmecH95o
zdGqw~SiJ-5o{AgYx~rkTbf#^I`HgS<Q#rqF`5;tQ&oA|NO8)1$`PMp-4W_$V*CqW|
zVGMg0_MJVHF|2C^FJoC9<F`#W9=yKt{8???mXC&!Me=QD!b%%N&Hg$cU0oMbCm1h3
zHU8*g_5+95BX&C1RQzG6mfP}Uq0)i4X`lTX=0-P|p4K&6rL`e)))}TX5l-pyD;z@`
zVs*{_`+suHXXDn7bu8KH*TlpvlQ6f)f8*||I}F)vzGjmcuf3Zd9Up&)VRwxp<Fyyz
z?>{FsoPB$@f9g+Bo~UO`Ya)}Ep3LB4()o4a?eBeug;*nYOP4V7_e&=f=5L$Nvq68+
z#ob(Q<}j@ZSC@Onr<ljixNT;{<uI!QZKh$fs-ss)x6i-RwTLfb2Sc^sSEk*^c{cbh
zI{VtC!L%<+_kdkugX*_yZmXDPuP?EC=6(ED>ZQ9!f)5Gl^eua|Xg`OvYPVIwN2WE^
zF4L}XZOB`6wod6loXk~6&JF(J37r)hu|bBhZ*LVct&#BD%)lM~+o0;`?NYM>)s6qY
z@B6qxhtW*3_x-x{-w*4o6i)cO_vWu(%}-X;NhM4Uy?fepUi1<M>&S+ymG|S8e0$;c
zNFeEd1Y?=GU_#+tabJPUGFLe_T)Dk~rFkjmhBeYREax13aC7zpiKMv<v-{ST^QUKJ
z-0_t8I+Nwbj_<p_mOuV>dBr>7g!yZjX78_icJJ5OFN(WP$`_v$X1&pTXF;mHV8Y>o
zxffgC?-Z}Nut%OX!ua~LgMqHH8!8y5&F#50pC@Yl)ZbITHJ`CqDCfi5wfDopM{1`d
zXMVBV-K=qduX@GZld;P;TPmix*UP29me5bzc4;@$n=OoH+6VG-{{H=ces|~M`pH+<
zxiVgxb#6l8?*j4vWyXb)$5jp-`(IbQqi?760{bx01nd3sdVg-tEowM>ul0ZG;{|o0
zf$Ta74O_X&W$gXFX^V#Ue(m~MdV%xn;?ow#_TGDY=)&%qZ>r9?$+O%@bKP;B<6mZ+
zZ~C?yJ$WfJ-hO<;*}w1ao73-qA7!}xbLkw7>xUVtr$#@|k=yvUi|Nh3lH=z0M3{1_
z*mA7+X6Y6kE|;68bfE3%Z2R_=o<f%xw+Y1-|FE&rF@5*-oyhXgig~-#-d<gF-mU)M
z<ez0;Kd0V$`S|L(Gh6e%e_|}F&*Qx;efZlU2J4=K`<K<v^Zjq+Q9H+N`?LQw>iaA8
zPi0<xq1J7cz{tF&=DE-E8pXeXjAk#r|4FY+Q0LfCm$_?IUCqtEkN(d&^^-k9f5+pB
zTVMH`Z10^)JsJOb_opz4UhBYyyZZZDo~@{wKmA?jzFFKGPKx~(b(x>svB88ftYX3J
z9g3#135N^B7W!?N6ZY!P-|izfB3a9>F<#?%UQ;F4da`0`_QSmVsh_f^^)`D428jB1
z@I9`zlRYzK*`+scr`pH<oK<N3@g}R7>4A0AtCwXY%;DKE^@s?^-A|9AS9)KSTKu^D
z64M(M#xl!$d+$DwTlF!kJyB^()8_E$iw}Aln|ivb-+Mmy$;>t1dW&Xm6Ri_j|NGM6
zB9<F@O8xh)@@@!w^lf(d)ATJt!6hy1GNB3c6%JH=l)lS#O!V>9W8c~0IqPy3$Ub#9
zw!D~cerL+tz`6JTy*u8!?X`_~#0&=QQop-D-k)5zaC$|`(mnUWv`p1)-+8cp%Vb_N
zaYdli!HfGpuIu7I8Qb~q26vpVab^9rsFEXvRV;;DU+HG5t=iHzIblNxW0+OY=CUJ4
zoH`TBGc6~F@-#8r_WtUzT5E&9RKnaVIih=uQe%E@jtyTVIeQ&T^8B=-S@&P)WWG9(
z^yk;N;HftcY?qs+c;Hm?)zcrDSJo|^Df4CB)Jbbguk`zi$_F&gEo@l3v@ChG4S&ch
zoy=?d9$h_Roc3mUo89`wn{rlv%9UPbmUDY<!Uh+{X>&Ldx2H91ofN!co7Y|+?NS!Q
zJx`oj%f2yQ>-Bf^`~K^<#(}Wc`({m5ST*aVgbdqc^|JndYoA4aS-Z?FzP$Ezq9oIs
zs!in*SGYI44ViB7a@W17%sMaTZ96>AowaNm<264%1-+@8pR?>2ys2M*W!6iHJ9_(n
zxUGH7G5aKMw4t)CPGG}YweE#$s~E5O&M(!lT{?B%r0<~-0;>+3GdnQr<*&4eu5A}T
zs&9KSZTrfZ`*vGcq+VMY`eo}S>9FMu%woz17QNnDC2GHJ<MbD?-!I?2*<RGZ`t;V*
zf)_d<{ZTKKbF!B%pOcVvFwW}0w@JTi*6(@f&Ju8R=DLNa1^%DEck|DdwQCA3`X0L^
zx#oGVKeEa-MlsbjGjr)0iz)NGA}{7^uD%dgq8j!kfG_Bul*eVIpX>M7a;Wz`uRU(O
zO;fdUo~yX$$>TRF=l?w0bDV)&dBgjvN{cd~hz!ZihFzY@VUMG{Yie1-)*krvi>tHk
zV~9+({>#_1>jjUkGKo%mzWlXG`_+s4tLpEamtbHq6THE{_t#F&HF+~PzesmqD)mdm
zM`fQ~al%!uoE86l?^?WKEfcdbVrOu+J<j?}`pX=r`Kn*#E7w^wHtg0mSh;$)ij}9}
zjfqPiNu+LB7(UNI##*T{RC~j`xS!6CR~-NM+0#4i5V!8;sADrTBU6&f%8p#QpJ49N
z&?<Oi<2~_m;fVW2XHrBaw>?z~y_l=A!HZQa;+DOUooD0h$uF3M=UzE&Cu+5}Iw$g&
z<m+FRybQ%^8)~2bT5x6G!lkG0oLFXj`jw%j%!C74qBGWhGVe_ePDqc5O?c$6r0~;~
zxpQuASn_Ye+ZskOw*=<BY&$sDB+pb|sm&zJox$FCcgKMo;c2GHG3yTe?JaHWWS)3;
z&4Foiz7(fOs`c!cB$<%N5~i(faNDNSaL(loSJYX`J~Ksc(mYu)oyq+AuJv4RoWvt?
zYL?5S-)dW{+|I_@uv6g1jx`6EZl7acyJme3ld%4SM7}jajW-va{wnd#+&O{y`1+y`
zWwE=8bhh+xNoC*L?U&7=J9qt)mOls2%Ox1HidAyxL<(HXliwnr^8aV}w^pfXl4XGj
zpdd->)_bOs^5));v(@!qmzl4dB6MTt<O6Mb-2!V5U-M;Q+n~WDJykT~Ro!l_wgYCK
z3EP!7ylyB~_-b}bHZtMo^hVj5KeH?!tz$CY=XCz1?)uQ%dE9rGnax&YQZR6D+^QuU
zF`cFCF_ZK&`89#xI-cuf+TVFKLw{2ySI%+IguL*1mUD{lOqNXe$Ep^#+(BnmQQ`{a
z4c00f{H`)t$B4(fcqH)0vRo_K<S*RFs9>Pm7^=JB^O^(S1aG|MSQGA$FwNr1ouIgs
z%Yg|u4@+D0{%7UjVQ37UctFaP<=MOgGniM0Bor=X;=bbWeAa<YOx<(B5?r7v@*)yc
zHpJ_0D7+_F{vrG!i<njHk*7)qc^qsGOv;Ua?U;7|*voQl%NmZIk&Ue&JrfV)&A*U%
zL1#y3W2SJ>`$w->uNB>5=W1kB*sy!T0WsGE@%=6dXSrC_B2)HUbWLDzU~*2_zx_a4
z+n#OHkDe71lwkPd)aZJq@oyMY@U>@mUkN;V!}@0Ta``LFZ<IG!F)=s9&pS{yODs!e
zgN<cvxyFs>`&|+?&OWe9vpSbQhkXVM8^ivU2k!hjqhio|VBtNj*@mKGH&&Q!c%ZW3
z<<=vMk6W_uXgF|g?vnbCo3l+Mo|MgEU}(JCci@>_qp9q-?5vYq8j1$<KR-KEr11F1
zQHA#m_vak=k;)8;qpvx$f6i0h5X7|EYH8fe168an-(u|}3jY}WXAl>daobNOXB%U~
z0WHB9kKaxgm{C-4j5TWQhqQf9g(HGzXKHC~xS$yg33kN|7qTtdEe)Fe{+WloTO1RS
zki7o0+b0(ZZT3y?%B@9H^<B2#eWCv|m04j!5YuPrbBc|UJi5|f&d<u)x#vLJHfy75
zl?`oWSG?6W9LVKVWMXzmsJq^Hl8Jjy7bmCB&s#;4&-|MnRoPgYDq5w!VV~O4rA3@;
zUT!;eBwvf4!6RYQfj?r~CLXvGYv3yRLt9}(UEQ5xuY{}|ot?t%r+X64X5J8G5Qx}!
zpvvwjOJk<#ANO}3cdv0sxXv#>-8UhN#Y$`X0U=Y~H}W|zL{%6W8a0`uf0euvoG~p@
zXL83alVrw5-KWQG#T#F)*7&6=e8ZsLq4byD@jqJ_1tK&XOKZD0nT%IF?>m?kt|cN6
zargBi_EL_({gZY?8#R8N%d`8{(?kY`1lw%B8;%J_J;Ijoe=@q|$I6slnJcnF&EW68
z$#W+$MVA`AnRNZ?XNHCYB8{ds*Di{S-bkq9_Y-*DJ1bLid&2>x^{VEtCSKhh-|5Y@
z=3&YWt8gP`g$-IP*FLT6<nUShv#RB-rneXu#~R7sTb5e|tjLp+pCTG@I#V)9VsoDy
zL*vhQ_HVvF&*ym`>f*|&?CU>gpkCq<<y$T=<6@rdd`WN9=U28`t_X{4eER(2m;X{9
zZnHTg$Z@`zEhZqRv?1wB@v1`S=ug|Tep;`+yUT91#UIwPkg$Z4H@77JPqCc_%JA}y
zrBke|zRZ~LLF~q)>r=OvPkM4rCCc<&mU88`hoNil_SKmA1-1)Zt$xAd&AsNe3Foz^
zZCVT*XW|o-bvZ2F3Pgt_%r34eiTYN}p?)}(e{S$J--OcnjQ97jl+CNPnVF=mxIuY2
zf5j0M2?nOm;x{5zr};g$G;8dXUt#z^*>&=4>C>jxnfGdagJw@_GrOhRWy@CcIZQDw
zLr+g}L)aUmwiV`F3<?I03Ag;7O*D06x$mvIq3t^F{P$`ncifiWmVP|zl+`wlGhaf^
zKla+NC$!^O_m|V$I-G^y9M3(O6a4HGQ{86<4j+LYh53DJt?!hI&G^yyire63p)H$q
zKfCEw(Mg3;(+rlL3o2cEbE$oSSHeP_tDe(z6b%Zc)-Ib{d}>E&`+XY*rpddmr|o?c
zt#o7gju@uTx2+!KP3mdy6wFz{ztU7k(?zs?Q;N$Qp_Yxa4=CMi>eXWTcK^oH3Ey*_
z7tTNS=NY3|Si+~$1ox<?v(MDXc*lP24&v1L6rorYK9g;F?Ly`S{u6FKk;`H(T=*|;
z>jGs?pMcolZCT5?8e2C?9bCkHc&hTp-O2_GLNjjcms|N%M=o`B&W&zwwGC-D|4$33
zAFleCvEb#ip!exLxj`}#=S!}%T$VZ3$Eh>ZbN7_gXs?9H{23;0YpWBz?)$6_WCxXf
zJqKQ`@>XND2o|yb{%8}Yj^@<jy>IW9ym(tQq1Wa5pUjC@merkU4X}yK__+Onet2VP
z>vaCAs~kF!9`gIo{CQsfvN&_a(SP-~ctGX*#>4h6%kImkT%D$QK>qBn>G~Ts{&<mc
zMB8fq;m5+h+W%9nj>whiG<OGkb{2&nz2EiJaMy!u9d4gly3wLRm!@86d&+8-+pBNf
zzklXW4R^0M{Z@ud%n>P#CyNt2_x8lcgs!b$-gS4&d)umMHrp-<zM5apG+XE5ZmowG
z3*=bMdN<hbf1Y`1%~|uy?oUN|WxMYfcmE7J@4LpLF74DM!$#j-I_uT=ZTD84T>6R0
zd+V{qA$E6W*-l?JOM2<ky7k9d&oQVNI3`REoPF`+x5?{YCT+LU-TLv}<m>#m-k(=X
z-IT&Kd)e;Zx7(xlR_L)_^V%A3bMHo6=K4K<wg2AO_#^%AOd+|_bD<yEH9s!ryH-ER
zSt7@?IBnN~Wv~6>mVJ)?vzM!JY1HGZpL*i;UC#FISDF9ogQj%}!-NAajY~~Uqu(?|
zEKgzLHr6+v&g8xJzHs2WSGf^S)^*!^o&NNu_IJY5ld}qZvbQN7$xK(1o4e00;-**G
z^|ek3#%nijyla<XA8Bsy74>WJ<m+#nzn{PUSfY_}!U31?E1`{&dES9WjZ53qpGPF<
z?mqB@A#G~)68r1h_e@S^n(gO!)?ihr(uQfl=_izGS=bzyPTSmhv+}6ki@PCAr>EWi
zDH5S|aR<A><AyoAJMC9FBz(&L6WM#<%C#K@Q|v$myrrI}_-_BamkS-3vUOVytP#wK
z|E)dyPWad7zv7s77nr@PxEY>s%GT0ZyfJb~{{aSp2*0VaJ3=PAZ=J-cV;o|#oN2S?
z(}}CUozoTauwVZCX};-k7B#8w{ZX!*YrdUxO-O7MV{UL2jW`x$%*6dfvs2vEzj5aD
z4>eI7ji%OlK{HdYxmd4zbp2CY+Gp<9rcazdu3zn$5S5*B=)qqmT{ebptqlR<H?D3A
zto)}n<v^PE(>+{k#56Z*Y_@FoeP=zbr{vv)g0yujR^4nDXpP)ZzwO%!j|8u5nde?6
z+mA9N1hR%zFXr}HWx3omzmapc#+em`d8}&R8Pc{tJ2CNh-|io~C+#;o`q}OFQSI6F
z>(+6s_Db0LpHtpo_s^*^zl)=I7{o#oUatGi{rJ3*?4yhAv-Tfw`*vv_FK6-8gDpjW
zk{%@PmG$mdDmIwyFUwVb&TNW*>z&}42eRh(sw$m7^IwE#Z!`~s*!2FBYgznt{o~%A
z__*WHrGKV7*Ob=JyS;7a)U`L`i~c{`Y{xWX+wF?`Qp=WZnPc2C!|=xT#%s=+8XMM}
zkMBIFb)uu<&y4vR3@<{zz5Tp(Yj5t(`L|;%qwn9oeb@H(x(b=+A5O++ORMug+gr$g
zS=yu5KSruh>wD_oo#9PC`1k+XH+AXZqAKC^TgO*6l!s2eAMLOCXUY#&hJt98_vf!&
zudd#mt-gNl+V6js-|?6K(f$9y&s+EI-TT${J9_3Xi{C$mvO~+mjs7m|oy@=O|JI_f
zrq|5o**y5d$Uq11!@j!w^W|f6*$X5-a0mH&-ttwQui+VeQhm}sg+i~WTT^Qm>EE1c
zvGZ=4@>7@DCCWe7Gdeqbir?9AJk92_^zG!?4<zmTzg=M4&-1v%?)O}C(BuJ)oPDS9
zS9g0D6GLn8%db`O5xfi{x&N}_*w&OMxH3%m_G$Xv^7r>@Hc#38%543wtJWVC87Hi{
zemN}v!JE_hx4gGiwEt!OY{_V_=-W2)S2-tz#IBYecV~4tbAIoZciT^IX}Rbm`*RJ0
zMAZ8)O?L5%TzkYaPcLUZV8C}auI>Hp)0=v>W^~T`xr1TGjrFQlxqI`Nm=z|r`K@PQ
zl%8{>@Ch^Xg7>fE6q&?Qzs=+4X~^FHeVS3i%7D=I)1O~qXFISpDm&*u*}Jrr*Q}4P
zWG&_3X_&G1$JW#Fc2yty#dp7#yZ@EtTju(585z<0zYj3%*e1C<@8`+03|nsRIn24H
ztaOLA?1ruJ+y3YMKEkl$+L>J$KOaAf^-cJ@Ozw9M!wEB+=EpjwS!?TS-Gy(sGqD|D
zdA;^b-P&0B&RyM{J_i_{eJT7|f7HGzIe+5sQ!H!&F1?@Pi%l$;zS%r}&hL}J;K~#6
za{a8k9luH^{_OCF2;Ahp_EPupbQdm(1AiE`%MWS=?w|JNRRSY(gu?-^*bQzaY>#Xi
znKd4M+k1pHyU)Ja`2a(g?Sd)W5}J2==-y;dOJLyUipb=?7I8{Bfr)LyhK8wI6V9ss
z?Mh~3jtDt$Dsn?q3D@yUx-%LWb#xMz>O}mP?yFX3VhecGFm+=>_2~-1%?*q?77404
z5!r`gm!yT(-<bC-asO6v`3?X5zFrH~Q*iKP3ft9m_1T@~i2HZ%)qPj_xm^E$Rn5;u
z_iHN+zU_RyJ752E<+i^m6B(<2e$hJlyZ>Ep?UnuiZ|#pdU@pbB#_zzXunjNV_C;IH
z`}A@4-SzisuI2|XwLdVozwPDAf?Qjv;0}NG`%c$(|G0jFziww$^WA^@&R%|9zwgie
z1x2&puah|%|NNK#qA<SuzuufToG4w#So;0;7jK3sCJCzA5z7BV&$OE_zssMu*g|MS
z_}Uw{UZsU5?iG$-Isa(=m6xn@()iLgK3%!L*Y<Jun)nBqe@g%R>|t@JWDZ-`^fme7
z+0RpEul@UB_MFFztGBpa3;TX=hq>IZ`+FE~@BDRq@5JRPEH4kf;C{~`vB8K@JM>`G
z&-}xMo43kOQxn?|WxR2l-mQh=KTSWb*gTijC!y1%HQ)cws_Au88W^>OHVBF4*#1}C
zFnjIaG=13(RlDrYtNS~KPmZd7$GK)&jo*taQ@mbHtrsvzC}j$p*W}7;HA6F@y0M{F
zC?fi$;|j5gE8@le3Dv=SK1Uh<y~NH03P`Wu4KM6ucFr@taqD>6%+E($%(eD57wLak
zF<Z9I*e_vf=DkYg9W`L3r-C-P{VTR;i~jAVkkEc`>QtEzS_j>2=NB_^hkVSL#rUQu
zK-eIml__je)7!HjcJplecDCd|QtGd~-BT?&?uxw1ss8^=^X*Qx4F~g1{=Ff+e3ppx
zfphz+Z7Lr0GqJ4+I&dm*!!w@m4?DJf`)K&8W=Z?>!XLl2-1P6au`hghQvUxQsk5(k
zH`m{tmCclWC&u#c?b|(<Enc0yn7#b}<^SKg8JWYh5|(O2WVih`b&dXg{d(rFXNiB0
z-cv|;eRB2mdR{)K>dw{E*~9yI*|sjf9sVo#WuE&zpIHai$^XxN)Vul8&+iK|{-$Q?
zz2xm>%u`QL)rhEV+x&u^@%5c4uN*}-yqGd!b92gSw{ME;mcQHgQEMZ=;o^{F=5J|V
zieC7A_;5S%)!N6)|IR+pFqJdnBX`*Br@KQIH5~n1kiSG~qkzlzR}zvfW|y?z8wgyQ
z!<gOo&S1?~CjH2Mx4rKluU=qU<j<GDT6-nh)m1-2_5WOf$HM2+56bkdnPPt__r;R*
z#Zd_XrzJN$nQP9+T%OtUO<`WIPu2d}2OJKFtf??d-@&8fbvXXz=J)aJ-)^}&pWhs^
z#?62I4e$Bq=G!NnXU?_r)vtVi{9^c<8DJd+W@-Ep*-N~x&su0&G-J-%RSTPgr+#tV
z@P{#*ecCe}*~?FKh0J{t6cV_?EUxyPu3gR%QRE+Utya{0^5)PVmo|rGv7M6LaKwAr
zevM_53fqrN<K$uSIoS1c<<s4Q#|<RjH$RRs%Kn<-9(7PyUizeM?@wkio%EVZtJ6wW
zDS}Mh7R1PHu2K5?;JFv?c1C5C-(OJo{y9g)F^7sNH3`4AujP>7n9(Ha`+rM!_f4iU
z{fJd-w|?9lYP#o?*2*1EcyDA@oKMK9G|IQWBxE39z^d(e;F`Sp+a*mCYr<?FPx;e(
z;ceagrT2_VP4w~)^c`M&zp#k0|I+Hjd{gFcpX>7f2maxo7x(Yr^KGdhTQm}0F7>b1
z_OeQN-uXNy?Z^hBUBw$_eq>EK`()nHJ&d<G`|H$C+P1#^lC@%8$^o-Uw(HMr7Q0q{
ze{W@ybw`8d4fe@}H?j=Y^zhyY+rXqQX}CwC_URhU7YW}QwjS=UnICcB0^|O(Q+Xmx
zcdEsoFNojx@cxXjS6(&(2CTcg8+?r$-?GGO@69jBT_U}8&EmY?CpHs|6Ylo#e)-+G
z|ILT%q8n~9n_U)dRY>46%Rk_kKI_1(gMwNDjcGNPx6KTl*gq-y{2WH>vp1H_Q<M9y
zvqg?2=Uzi-`c6$rju}n5b_u~g2`g96w~mb7u=fZ5sz}kB9(jJnY&v`2Z<_SusnyNx
zw==kAG$<x;h3Q15EBYBWE}eJf)U9@tq7v<5<((64j8)PP1o8X4U7M4ZpJeUIapTwD
z`EqH0R@D7p{+o%7$!vGt$A7QS+Vs{H9k|CnTinX^k-<Nv_Q>Q5X+?dfBsZ)*%BA-J
zwici#{*ltg`5X35yLsa&-;L(gu74DNEPkEOa`x|a{{s#OxbCdan0$3^DwFk0gZC*c
zYbF<p{awqt#`@R#hrjLQ#1s<NzHu^1(>k9!=g7rvsxAk<yl$?`x*Mss;c-I4z4W}e
z{ztPP9w=Kksiy8+gKTQZ4(q5Jd5O%|PJ8{oaXFZUhh@#2zo+&{{}mCwvH5}6n;iSK
zQ~4qu1peMGyWy-MyOwk-=Z(IG#)g;IcQmZMyZ<5&dye)V*3(-=@ALQ<EBTyjc<aM;
z!|ho|#FB=Y2OJK3(cX7c^7jl)=CHY1319axR-08HxaP!Yt@%UGMOcDkM%~e})xGxR
zZ<^WOJo?5I<~whBzWAbs#s<-C@y)+)zdIwaVey=L#$U6qmof?)2pIg0YKY4JWR`uZ
zVedSxgsm@`!(s~#81=2>yrIOL=G@Wn@_M__+l$uo6cens3vGCp%dWF$YB^8Ny@p&a
z38rQ7@)6l*X7c`Ix;8^AVd`$~8?OzcD}_PP`YY6AuA#WQZW)vHzSo>Nciw8T_#6;0
z_^UN*?-^+a&KV7yAno33CYGueziasVpJ~pW^Oxd-FGhCE?B7*?n~^*JsVuXwfyPp|
z&flgxs$M4qGFs37Q1|+B{&Tx0h6aq$J_pX_=q=D-)UMml_WyWg_P>Tu_4$08%xvC%
ze8*I#lfcBr6xLS9+vhp=jQobe?NUCuZMVw>HWXfOe9L)b4&!PL2@%bN)ECziHs6Sy
z*)W%TM#I5V>}zEHcAZ-+Uv7RxlX;uv&z3*hYfqZT?^)UqdR$~1s4=oyc*CSG*&lAp
z>S`p2PrkR_vQ#TiJwd!!Z)yXlgotfI>dWg7OpbiM!|1KH-t4Z}8snAIw~KA)HMI7w
zW>QPopv9=|yPxfQ*{gH5jaQdj?=t<nt?VM>>Xi8mk-{5}vaHE2X7rZd@aaqT57t=g
z&wdFPFRpFoeE+>_)#;;o$_ZQFEUp!06gG&M*f4cPe6#R}V{<(0pR$PAzROs(_SyfN
zyQk0Fw$YB^G3UX)E1?<ES{oN_cRZz)yI|_3S&iRhPp(dxA)>kZOv$Z@n-&E~Y0h8P
zS~B5}X1vC>$EW`vIc}k7`R?=VNS~c7!ZU1azgZr?Z_1Z*k8$(W@ZZ<CWDIuKS=@SE
zcSd@{{#R=QSMOeFW7XZj=*F@p;O{5>CSSHn_k`TFi3etV<G#Cx`5CX^{#ErySa~9T
zF8%KPyIR-(z&-On_y3Bs@EBMo9DNmD&^-C^9>&|3q@qJ(?pA)0WG<T{wPE8YhU2ch
zH<Z(VRack2ZeSJt^RZ^1azevKM(v3Ge;PZhSg%JICpfRI%$T;?nzt}k&fsnx%eC;m
z$Ckyjo0uegSKhxfKVQ;dwcua-1)c{OoSDOR)Un==H-65s=2h0&M+GJ6g(*A{zc_MM
zNc{=rlG;$8{G3njzE;B1!Z(R(zD#Tqk{cd<)voxt?~Emr_U@?3H*QsN&-=SN_5ay?
zmTk+d4kQ{fPuuhR>qlLtw0>cJ<Cg*k2M#odZixRU;5}3BuHcOCjhUHJuXwNgxmPp$
zKF67$x!SoR#tFTV+qtju-I#yK=mdA@0lmtd3}I#ot#85&7%b(VNHtdGi(gLh{lD2N
z`BvSNmupWI>*d)qeO~|VXnjZM_6dx?qczT#J!?3*{eqvI>&ynm)v_CszGeTC`Wn4Y
z@Ww7H-`*eEb1OEzoc}f@T<^I#)92}Xa!)_hjW}?sZ%eb!vxcol8)AhF4s2);-4eer
z{KN5`ZVC5Hr+a?N{B_xvCD0-<vU8vB#@%s!f;R$kmOm9UO1Qkg{H=ZEV-IJ8xecKr
z--WEy5*h*-wR846tq+Qy^uFf(E4DLbPhQ@-p1ktO=b09Xnf;5`&rSW=lX6Y)hKk*?
z?(Li7Pc+<|_%!~Y*ShW=M((r%pR#jR>;12UALNkOaEehoZ@-g%@X03IqOHtgOBAln
zd0pz_r1w}|r~Gm1ME5L9w}iSx<+pzdE&LubZ4>()`NR8A)=Va`9zXu9gZ)AKCU_iZ
z5ao-Q_}g^uu?M@(E8k!K_tIH`4GB(lPYiX`x3SkacP-sJyTZqPPsFrj7BP`;3SSOY
z*P2YOd&qRH&;Iz|Blp=&n6}B;Z<p~|bzd!^!H`kAWd9*&gL0$0MJM>)x-1dipx`@w
zq2jq}^*UkU`fHC>z6g|A|D17=cOKv8{sYSv_U%=u*N!=`=kxme!udN-h-@f)G+D=e
z-Z94m4O4j|4*nM18&<!s_}1M&tFQlBGs9vn-;I)er?ubf%KJ?UHe9w@?cAOVH{JU_
z%=owS_PMX}a&c!8S;DxQ{``5S9es~y>LFI28}a`W|I}r?WZc}+J8$);%B1RVbtw$f
z#5OGYu5B~f(NQ~LA>X_0RnqP2=l^KdJz9}pwBh>N?S6A4tzBbkFI>KNG2L7@!FO&$
z>ok`wyS^Tt{x9RuhX&PrmJ9xRi@fh_HMUK7ed+3{jTL+o8yK~vHaz<N`dLuGy9UiL
z-lNQ6rCZg~!;jpRFIvMKtb6*dnMQ(f3$xnYIk__2{#6HVU25UYFZeFc$n3*$!~ep1
z$v{t^>KUR2-r5PL&K;FA-X!w<w~brEBadPujhF+cE^YqjpEI|z>VV8juWf(AnAp}V
z`n`4j<J)t--ukg^iMYYll_zd^uYC6D#@foB1B>E!>C{|3zVs058m+gTL7ited+^S9
zbj`f>rMuSw2A#m@zn6aOSt4yv$h<g3>etJeJ>_|J$_DjX&K2iQ9q><GZonwK{b?`%
z=^*B3b9H2n@TJ7gI_h|UL1*IKYp<`LX;6LVsH)#DroU4){_nBR``FZKpQe2fILa9j
z{ijaqqTEup9F6}GI)!5FjLaK+c0YQZ^hoZLF|*p-E|F~!TP7)Qeevn}5rG^(<qb!|
z>uhQ}q7s>&b=>wiW;9o7!^Rr#K>e2Z*2BNg{dZxQ_QXS~j@N4@Q`j%%y1%cxzXVRL
zI5>|lP2Kc2$C+v6%<DLo>)t-fHRDmuy|6ExUc56FS0vrOBx=UQmaucp`HkX7=lXqW
z4q^(k>e}*hlX88CYSHrhCzo$=e5ii)bY9zBi#(eQm2-XDj&|D~V>>geRC{ag-G-a1
ze(%op<o7?opmV!D;Ly4APYh2!xYcu<<D1=nu7vQviTq6$-p=O@%Br|G^YM3yHy?KV
zUlnqqA#~}~sDDd(*6(86{3yEI#s1@@2F3`}vRV@>$+_>on&tnOkbk}Q;<*)9SD*do
zCEdQ>-)-&Qa%1}y*Rx8^pZ~F)_WF8!#jh2~N4>W$WQ?Bte#bVkw^30CR;_x=yZX}3
zNG3Lh(*YYAc%L?eZf4E#HQl&DiT5mr1cOeqO?JL0^R+4WMDL#oFPv($ZI7jH!uJs0
zO-p2#$uO}ol%F?y7^fh~bZwgb7Af7Jr~})cXlHial<hMx7G&6vFpYC{ficUQXFV^@
zWqkSdYku(m`;&7Rh4Us$T8c2SF?g%z$!63X5ZS_fZQA-JKR0&CaxpSDs73_mSSDs2
zNV_NfT&ezm6blbS*x`VmL4S?ebJoOc(7T)4)4+H@XiM-nG0l(zTl^JQF`KC+Fl>;L
zVK!TQUdSLl?7)`g7tN={7wGgvNOx&=Gc+9C7{gpPX;$=sE#gMP1`H9U*<G>qy}UQh
z@LJE>#aKKgKEUd!*8zqH@4oilV9WGV-q3t$k6Qu@55u%GWmB}vdiid2_r7!6F88ba
zu~$ZP%H9T9ZRTq#2@D%zu1aP`>})74VY+r<)|5%*Qnnxuyp}d>SN?7C%awhN^8tnj
z>pFV1cB|z)ot?>Ic1<ei+1*ZOjvL{1#~Bg?-*~S~$h~fw#GChQufTKR4e?XkgSz+R
z`W#?LC}b>dZ3>Mw)@w|aZ<Kw`6Jh_fOp6&5=3*OC4xYNV&HK4_{g;{V*|*I#PKaJL
z@#Wb=?{ECImzVmuoq@r?EaBZx?Tzv8{_}9oxW<v=ogM2Cb)aq4(?ZdYixT&4{oZ%q
zkRc-CK+N~b41qM(+xN=4L^f2fJ2&OPmWH*f)wZvZ-1tG7flJ(=yeR5(TfBbrUroL6
z|1XT^I9<{9i8`=nb@NGonO#0J*&9+>-kkb(?&XV<n=i0=-(JSzlg$!#wW}!VfYz$@
zH_Vfs3o&S<9C*Yd(U8hH!{W43@qs<8MaMh+b7owsVbG{MV0DaT%}QZ|=-0cwxFb$Z
z^%uSM%qE>7;7h|^(}ypNdk*ZmTeH2G?M>Q}Qo;XwR$cqS$FSi?!(9F|AYYw{=J=+Y
zen4clOWw&?W(FNU=Xc@9JDSfMF-gDDIp@cFV^@vfnu97X3_Ct6-bgS0JykOybnfYp
zm5iU4DIQz)I9#OIn$O=&Cxxw{wQ}a0@Us(~&lhoRT(jA+Fzdjd;B^<%SLp8D7;<ZG
z-DB~(!wd<5b}sCRlQ%bNy3b2xc{X3i_gLiw#e{py4qVLLrc?Ql(b$?H;o&QFKZ9Sl
zq-PZ0naOlb{QSSEZ|<H{a6YhY&#CfB%o~2(oA`Z3G(SW6#SL+*ufEFV>#=&Fvs|Eg
z_7Ru!FC!m+U|zz$W>U=a+W3t{>)vkKCCM<Mp5@tEmSXjX4`eMTo)mw{$h|CA?#J5t
z^|y8%vA+Fd>U?1ah4UO|?0<{zI`Msh;<-ulB{w7%Z2Eef<?yF<|DVS{Zog14GlrjG
z#$k>%Msw3DcU^zG)$g9lHr;C+5vnG4pL%`}2(MouXuxp5iRsz=%ib2}gbc1O{xi*j
z#XbFz9m9j?yb;0X30HTrtXZ`CyO^>8!+{#NgnxSu>_}VI@bmM>#dVq>y^EOAEQ6|@
zynDAYso8dwd^br`WMW8TTeDJj!=?M0&FX6#rvAHnyQaG|fuX^&_kh<@_mg$CTB*Dd
z8%iuC4tKU2q^R>TG@J<(Px9uASd?(F_xtt7^Gp5ZT>c8aoo>OFqVK}caH_vg=ET>6
zBU>YvUTRN&ShKfd-)5QrYr2%r8`KEw>c4e))}s%%-nj8IOsIYFbyrvU)b;mF1@3=-
zaI=5n?UefE+BIRT!)O0wyMOVl)H8GcE9HMS#m|59=A$zc!<qiQ7yr)qA82>YoZW`5
z%f^4$SEpZ|ZMKc=LBm0hH5o;hy&_95mAEBMHOp1rcxuU=l}|)7XNM)ZF57M#6rOuo
zQsw`wwh0P9!|yfNZ>YUjY@plrz@~p!hQR~>d&l2@tUgDVNd)2s{AV1!ccEBj`N2j8
zhK8?;UiLaBFfcS6U|?VnkYHe7U}9!qU}0lmIC^-V#lhVMdU6u2MV$v27!rP3xN(No
z?$4a|i+jDG1OtPPe7REV!nDkr<srvYw!i1#VPKeczR&RZs^8xB-Q3yBK<ii-ik9Y`
z+UT@bwv$`?Sih111H*~Qn+$u4eEWCWNlrEg*;5&0`ZU6^v%OqZ?~RoFd_|BN(cnH&
zk(=4CUfiu&{UR5nip{6=i(}QYOtq4twTu2g_yTh3i%iL)HANa5+U#t9&-4HF_3USm
zsD#1Pq$|#iZ@Ia3>n|R*F?)5;Vz+Yw14F`sb>}x8(yF={^z_fF7yEtpNj5SvG%yC2
zr<jY}DBt_@(~V_nOs6})=dS;|g2OuLt>xixSwD>fcaMd*zi(mMeQ8<x&X9}B8}=O9
zxAOdvsE0b`GMNlJ=Dv9S;;>fLd1+4X=%1k%l{d`M(69XQ^p1i5WCn)v)@5eLHz_sp
zZk@mTT>fN%2tTIX4hakaJYPz`1d8rem~2q|yYGO&jG3E%bMP>Ha*TJk*Ssu|k;Ut6
z8Fo>5!y}G0RmIoZz;X)~tzA5SN@h%!P3Y-DohlZ!G^6Fq)om*SGoyEB{MlHoEfNtt
z{h+czgKFYi@$&mwxAvWyto2LV*E>=5_{JB4GY+J@d9rZF_xir<PbQh?Zht%;x6UWw
zYK>a2a{|N7mGNTN6Mk(w;JNZsvhSJI&n?7knJ&vsQU1GV_05TOKbclP`Mg4;k@1Y8
z!TJMims=(5J!&q7oGvex>6UBE&B#-C`8kEBm@T67ch6MLoQtbn5*R*ntU17T`R@MT
z8{fC8h3v3A=ES15VE*1?MrR}g;tJWq%+G4HA7GgF)mUfE!CkZVovPcCZBbNEQUo%Q
z^US+<7cO$8Y%ywNHC@+z?uwwqfjQqNx|+^-mpSc!Rl)Q7g32Ctp<g)8ENGgyOC@8O
zu(oQM5a)$cEYtki%<eEWtTEiTfNgb#Qu6ivCv!UwZk&}YYRhCiXIY<LP*J11wqe|<
zQ?mpr+1NH@HHuzIn6|9%-q+i8Wk()FoZi?a*Xa3TId4Ygua7#jJ6$RpnY3+lxA(`V
zTz%KyZ}BZ*eeIu&ladjo+wcBcTkO7m^Y`kv-|SodIB~6+!FG8on^oOy|JWV}(O{O1
zn|b6??*D$B&RSNse|oCxckaSjUoNwm@D{VF<ok(fY?zx<E}Z)QIopBjJ@e<UKQL<-
z<C^yjPi5qJ53FJh^JzX=`lakcRwPf~{4I;G<k#ljZ<+SEI5@|;dYAdj>eV}2&#P=$
zyZ-V1Tm5GYOJ2YJ5&UfJ$G-(TcHgcq|LuQyYhGi8*{YL?VPYB^!nUbBnN;WX-tyDA
z?x$RD^z0K>KZ%&bsx2O&$g389B>Ldm8LD=98JSLrrn6XzSLC+;*kCTJS}}#Aap~da
z&$H|6T9fNP{hwX`yXnB2<rk~uq}P1O-DR4!=~?~m|Nq&4_DF2}u*bn;vsarXbJj^#
zZt)v++qz#Lmk(1kU}57Hi4f#f`|hd7F1Sj??rqko>p!#gkF8K&d)9Byap|%vEs95-
zLld@1N3S!zu}t{BlqzR;slbgdd!p}`T)MO*@^WOp)GW4z7r9u|zTZB+-?Xm6lp{xV
zAz!0tSOTlbj2QQL{^Ui<tnQZg{?9Zy@2#m>6kOHS|8RO_!b$y~;rG_>pB!&|iF?hn
zwT%btG9SGvxi-7@Z2Gz|jScfz8Wjy9RvhR`->@md{XTE~(RBi=ypnvkPYpX?A6joJ
zYTx@z{6^X<S=r3(I>(KFn+VN#yQcX-$z`@HwtugcPkxvY%UafYKtMvrBcauJhHBIS
z-lvyDn-;5Du79QR?&qAS(3P4TudY|JF6#(dwO=^wYk=~G|CO&54DN1Ww7z(4b>eZ~
zccPLKOwI}AaS5)*GjvZK<2?JpbH|d?m-6>Kow;fCjOI(d9eXWVm#xc)7f%%CsakYk
z8Rw0)ViBR6J9V3n&wX++(!O!3pafGhEBBfVw&Ip^W@ROc-lbnk9sNxw*9fh;vQKmN
zffmu1ZA#akRTVad?%66ZBkxwjvVWTM=Vq0vFgqu3@HC2^O$d_RE6Q5p&CzFlNd3qg
zt(jjkSj=YmC#=)#^nMroE0@<Z;oU3_pI~?S+qRaO(&tX4=hd<uGd5sh>lTUF$g@rP
z%Ejr68A65X&abPQ>8~x&e!O~?+s+vW+IDX=FMTFBLv&s0_O<ml!X|Hj<{=PqHCnjw
zWa*csH=cbzve+Q1x^W^T8n{w7?79%X(=NBB=*oi^8V~RB32!(!`Q!ZPhx4aTI^d?S
zcK_Qw&)!GVJrn8zPd@%=UQ<$<+wt&KyZ^EGzb{OYy2;cSdd)Rq{t>;W(;fyUsFvmk
zfP$7|O$M7bckR^Om*qDaEbNlf&t7@&bYS@7HlgyGqrOe+CvS=Q`ud&3|EqT=oafZM
z`~CS%p15kY?eijE?z^J9@PVDcjg*TIwv>N6w)Za6<mJ2{{NsYZy!X?RQZ!&`J0K8o
zkw@)p?;}59v8{dkR<&IUtNPNrq33U+)jJ-ku5WEYzpmUC*u3hji>cd_lYOi1|5$bC
zyR0VD?3~Nm^{H#4bvMkqarfYUv)xH|GMAoXIDcRh_nF<N7xFdUWo#5Zk#MM3C0)wm
ziq)@KLiS#(+_SQz7p>)a`NYfdriOvJ<OQ9LAI}E{xp?O~f4QG>)Z5o8yI}su$P=5`
zN&|h*UI|oy#_qX<EZJJkZ#iDPN}{f8j@nP&XZ3V<p9pJH(eeIe$fXnOb3@22b3G^j
zv0Yx>%l2QNntJ4+E88{YgDT;vg*K<ZaC*%1tp5Gs&lQX7XKucH`14?<bJUZ6OAc(i
zFEelcnRzdc&1iJ(`}yP9pP#$uee6DW@64^6mMmrZw~zVr75LYFe6VEV)5XU3QuFt}
zl%Mlw>eKEP@!#y9E=J5b@aFQqUC(-Z_iy9>b>6?q<Ea?O8UwasQT7PK^$}hLf0k&h
zRTVkA+Bo%=*9mRU$t`DuZme%wqZb;f+WqY6pQD`8_t#x}Vq&_KGgM7KuDh5`<LhG1
zM%xeH|1Z^>b|bL&?BXXT#!D|t|M7k{Us(RHjmfubiT`(Q>e_JZ`BmZI7keXn?LU2f
z`f{7UzfgpM&CLT7?D8!Web4te|0&#)|F60^HrP~1ZhE7EWWdV%_Wauq&wrTkL-k|Z
z^}Y{__V@mM*~oRSH*xx+mlAQD?|)Wo7TtSeqi0@bNWxUBU*Ari-S^vm?&^Z)%i}9=
z*oX7q`&Rk?Pi&TYQ>FZ$=eLiqT-*Juy`tq%UgLqTqz$r9?Y7Gl`+IWsof4jYz`5Ro
z@w<lJhWk&Ktyo>jVm2l382c-4abumgEl>Gh7YRiN&3Y=ITzHOQDSNz6!ne&UR^Qtf
z8J}F&sbAP4-}FtOto73Wm%DiX_3%zyS{BjI{l;<2m+oh?mu-G^;$z%SX`|lzpA4_v
z51iSP7t}2s({<5rUzcwAR_562NxBmL>F+LG`=YYqGiT(*R3Tp5TLE8}%G4&lne{?E
zSJG<ro8@skX8h7F^-q;Q_Uod}&4#JmYZTamU%z<$@pN1!SIOnKb*UToIebjYToCy+
zL#{E_Bk`{BvbxJ<=TC1ra4q9q?8?4<A}(StSv20}8#W$YG__>@{COr1HWrmkmYKQv
zSjv?0bpKtO70X@aS1{c+;n^$s#{1BY=3{&8zw}*Lv*6~BS6`TQGeurbt6T8;)R$K!
zrrY^~&3cp5rZ2v|Cu~=}$DezOdzmlwo=kmoYNwd~iN#mA7xzZ2pLJyYhsbH3DhV^0
zwA&6|vHF!~b2aeQi@9Hx?o_sG3cR)TP3>9R_XTo|ok?BxcUE=BdGUJ-{!NfycRF(w
zN9HuqIkT4u9Gv=GGPp5UN`2PN9}ix*Kh*NJZ+>mba&4~d%Raq%Rb2;k-aaYnT%*x#
zc-1#}>kPRKzk<{^Y`<ilBgM9B9@m?<XYQEI)%d4WbJd`5Ro|1}lYa0`u{kVzcn{O>
z{w=fDr}KQ>8@0#UOOPilROZ1k=4kWMom116r+hTHe0syHxU$B~s!wH?EN^a=5tR07
z$yswca~0?9XuiJCaF3Q<MT=?{x*V{7;;hwwNR{i2((VKQ&fWN<@$XjE|Md4cRd>uL
z8Q6$;a$d;1#+T!uepB*zgWo5Mn~Wus_QvTSn4I)Jk71sObsyv0Ey?MuFRs5RW_b~w
zXCrj--Ky>O9)&r9x8KIBIT5ni{?*FuyLQcH=(%ILVPjy?={XF-QWN3}Ka}$r$cLW3
z-m2F8q<dNRn**+nEUcObtXL9W7<FtEJ<wFc@L4HfYk7_jugn3NgI;g;=ox;?K5FrG
zrO*oPbx${WhBCWF@*8An+q#yyiR&x&Y97!McJxWOpQ@ET^~=THeLsa8Dkpx^@bnk=
zndGN;F)pK-@!0eWX_h>vEEyxtzqqphwv_(Qi~7u+4Bfnp$5;|VzHSxtx-F3KCTP)h
zv4f$KZJY=0T@=Xo-+qL>|6rtOaDwiI8xOWEUCKN|IYP5l?#Tns$5VI3n`Zrx4v;E*
zx$NJ|M%x3oA|khP#)o&QF0o_1vB&Bzi<E-rB#s8xg!BeghTnOsO($*alWk{OlO|oa
zB`{&%r<e@c9&=aG8TnP83*H;mw6~ne=PFfZUbFpbov&hJ{ey*RFDu^e3$k9memQG|
z9P73TNA5&i-xkYm;J(3#i+9)0OBa?^e(|*`cIOge`Ndi{)$1~IPJ_0+^8Q_6YrdS-
zIB@Lh%Vm7wI(z@_W|D7u|LrdGnc!QRXI};1>Sb>eu-^8+EPErHndz(}(=tP~&r3;&
zWW=sJU?cMQ^Owf`wg+m~G93|`=CvV~-&6K#xV75pP1g^I39ULV^+SFt%WbQs(Cs}P
zXV-fs_dmP1eT8hV=t%+7zfAiRMBXqiyHshuq)KMCo#FI}OdZGHrLT!xd?x7g;inAu
z^X&S0JGm+*<Ug5evD!d^JtFIPnE!hHEex+`Y!=8qmNV&1Y@F@Bjvk?$C1;&-4z%m;
zWXe|bI9^r6zWe)ZDc0m=@8b5=h%Pj_&~SP0`ZAwO-!BWQ&1dR9aKh`MeysAvQifs{
z?S&T>o4S2EBx2ZSdusK{CYF76dsH5z<?wD;dg#?1jqAHoUT;!iSbb^XTIIA0uU2wC
z%<km9_UreJe5dB);uqejmsNfJd!GN*{{4S_c;-&`z8ui-R&e9ayGp-b_e(V#m8#yw
ze&b%ix=Eks@7(Giu<7={uNyXBwbu2BP2+Cq<lT4Ic53bv=}U^=bQpcE9q1NWda0c!
zK~wpEp{(YbnomF9o@0A2wC8)CS-X#&dVS@uHwQm_=bs+((AV^o)pVxphKUm`^mf0E
ze|c`tPEbPQY*pKQeACUBCys4o4z_dHwlG~#{RFo`=z^kkt8@*&aR<HjZQZ4raDw;p
zwbvWUcxISQdh>a@?WTL_3Etnoso!1tyG>%(_VDxSlVcy-&aKg$Z}+z_jeEn~sXoCs
z>jR}Xet7kB`N?&4-M8n@jEUt<X?Uys>-phV`P&SZZ&7~ErPj%EYQcfe{^8-1@0L%x
zUsL9ml+)iaBj@tF8;yBipZ@4BIs8%KoSl!kCXd8{Z(sNB)p?;96uz>0qR;;7Kasop
z?(gR6KIs~C-1gO^{5qA@75}EQ-T!`mU%dTXmro%lVQT@z0}@v6ef{y%C#BWlL9?GU
z+G~mZ0Ofctk$#_lD{Xcp&5XG6agFxzMMo86pR(NWd1!Y2sl4kUtx5mgL$=EF9jV}b
zsooqS_a*Z5Hc^J#bM;#C_I?cBck)znUc{~~j1hCWw)M~Y@Ob7!gLuczl-<j3EjVaq
z(zJ7fK`PgG#|c~HBY6x{vXeDckMK*M+Hrr~pKvuE3pL*pw|=NshxP8zl((|(c~daI
z_Ptz2kJ`MHN|75{(h1+KPbRz-Fr3M=XB{Y#GWTAen)5p-k;~cYp!3oTn>a2N8n0%L
zh)%qA-KfVrEm<mZ%>~(TPSJB_TuY3$J8#Re7np3V#vtr}p1Uzjm&2eg?XmM)w<6;M
z(~5eZ{OgHqJ}SmNpR2()a@Ub5Rw7IX;x25J|9E)DZWg8kau+pLeVF`PgfDSZ<(l#q
z_qR*Z|Hi+pwtbkd(iz5j;FE?yVZ+t>X)^8NIff<g7@l#2IZb@{VB$k9{`+j_7R9eN
z>3<jV{@343zlKs@hR;(iRL}S76sh!IIxlhAwS7(hwKF!niU+1$S@b>C=bxyq`&0%4
zh9$obJ2M<Ji_ri1Bf40A!F>4yMgNHW3;#GOpSAX1GOUTa&hVx8xNvHA%#J^9n&J<h
z9MP}+$fEnR^mJ&#tF0cH`xD=vQ2G(=zJW{F$Itcq$pcda5BA;Qp3<37)$q5flrgxq
zn5kz;*9!f;vrhd>TM|&>{yzN+XO!86wG5xOA2WVEL0#seamqnK1y)A0Q<COJOgfJ}
z=ccdx<&gLFOsA_v!V9OITk?t?duHTKm0zcDfQ^U4{L1u*rXTg<HmiKqe%?<o-uB+1
zvGk%r#oGRM>jujvQC;N*jaLzQOWvP*VLxxCX}?AJy*T$(d-b9kctseZXBBcWnys5R
z-%x$M$@yOyu5WL#G3A__75eAi;q)2ecMmgOJC(V`hv9a?T#F@X<(%_&oaEmiwu0+!
zolw)AUt+hungm@5<=f1-^crJxnU_US@Ck+g{5HyIPqht%7(O~QowbvwGBLaSarG${
zZiYqYM9<auw7<2p`_LucI{E3}BFCtA0u8O3L^HLrg-`N#tlGw_^HQQ}(LbLV4bPAK
zI4gKy$?Qu>-CNl5%($6!&h0Ed$g;t(-g?`Et8Uv$W=JQz=zTdUjw2)_dG)`71B~04
zgz_EZ=E(eYU3{ETEMdVGiRoTJ3CSuCIj=r?+|^)F6FM`2w?RWR=!Q?8WYp7`e`~+b
z{~7*<v1}c~jY;i>1uKmXFlC&-WpF6!@v;?dW%sXaTh{9JV6zBogns(FEsP1$xf;Cu
zD#d3Ln;(QkI~^9_5Lj~Rz}Hob0tqoI4?a!n3OwWVwIXWzx&s-;rWX$-7sN!$^u<LV
zPIW#Wld`|Wi#^eQ&KE&e&pGS~U%kJGRyXkSvqproMJ%ma<Cv4Ju_v(KBxSOB4(}U#
zuR|fun?KE-%Cg~;$c#7pF1z~9>CY<rqj;buLiiGoViaT7g?62#W(<#JtUkpMy3bEb
zCMVbU&kiFk_r|Y@FRB}&N**&j`w<%Towr4Edy4W_-sRnr36Bo=zVFy;+_2S8BH_!u
z*=--Liy7ML=9y$gKR9q*SLk$e5aTvo?u2gRw^0GiQL$g8G<jxZU)-4g#O&pK-VOIY
z9JXV=AyrlpJui#7<6`lv<uj_=jhD!<x~Ye@L~Y3v++D^{{AZK)BKH>?FKYQ!)g6gD
zFPFeoH&xQ9=QCqY#L7ISiOU@xF!Hj0{qg$hg2c@cKF`0so?yB`#;;i}Vg1DU_LnmQ
zmQFEZ;#u4*6E3TiD6!gpPa=b~!sYpUgW0trHa*tAacj}R&8!i2tlL`bH=FeEvKt&<
z|43DgZQc1p47`b#%tfU8au}Xz{`hLjXtvJifX9<e6$!0baqKy==MK!Cwb@WQm2u5f
z<IlnT8;Vo>pIh-(ExTQ|p7n-EpV%b>#i;5mr8yUN9qKvxCH;BORWpl2-gl&C>;8&i
znY!q}hVuo*%f$Qm68hqgdI!(YFilH!xw@1~Ju*YiG{%c1LSO#Cs;j;(DMt>?zHC#K
z%JoymZgVc%jXwnqy=QiBc*D4@!6<Rv<pnSI-+jU=am_A;U#UgeC0Lj_r%m?sR+ZFi
z=~c24H{1jcM&>Wo*id|<`>(Nf;A>C}F1XSXY4(WOTjGVTnUwW5hQ;MfYvulbNeD4J
zuxvSx0aL@u1lHF@Ms6HF_TLsqr#1ZKYM9yU&;C?=&AHFK-rfxtWxvg^eD?Y5195|I
z9tAv1YnCmW_0xo*@^OpPsR?y!Z|du94A*sT%r*FTgX6UJaoGbh`+3x|6f4})xt4vI
zQJpU9By#lPtC#-#n+)HJ%!pO|e$Qpm)`{C$O&U}WSxDSu%BkJKl)WJ4<0b~T=G%hX
z9v@pO&ay!uTVCn8!hwWWyWbX?8TE(+wQ)R=`T269>VYE74R&gw;_kJ#Ktb6qezAW3
zr%wGR2aEb{v<Yl(E(^Q=KRVfN|0}78t-Cz$tq-4mD^Ke$Z{^-;bz5u?6lrg`VCA@u
z;edy3=yEG@rWyI6f*c0%x~I=DOty|m{8Y=bK{|VN>^7$C#|OS2U_A5j^}DVI``v2}
zac0ebKj~)gdH4E2(VvR-)6}2eGK^<C5T}t{9GBYA8Z^gXN`k`ki_I1rnAp!gVmQ6_
zz@&oDhl-v)7HC-MSNd+oq;i1+A=A5~nR9v%Y%yW-+4pPVQsx`K_!z}xSPWdZwH-ZB
z3`+Vsk$SV(*|N9!yeYhWar2iYi}wC`CB+oc&bTJkT6YzPLFN89=PV6NeFe7Oj<p4!
zwY4@}>D;Ii!Pua?b>Z$s48am*pJF)-tRt7P-w1efW#hu?tmDfLh!~$`Jo9n&9--xn
zPqIJO>fXd$Z?uj1hFxp2slnO<Qy4eg>}2vewU}S!v$8sOgT)6!jSXK?9ITsUnKt}g
z)_C&C<+t-!KiPj`%HFkX67@4o4_rISpeEnEyI6F-pIUhQza>|9Op`Aw;mW!GNI2^4
z`EABm8P9w?9bxo;FWa3xIcZGsXM#TWY&%h}Bb%^zGTV$P?)H<;?5I(RV-aikq_yGK
z?bRp0K0YokzJ3PpJJFDY$WyOg*~hactePa8pnP}pyMV@%*7vroK5NNP_J8M%_2ugh
zY>|BkO4e>C|CDd3niR>C@J@&ET66cSgfss9Oc9r_{a>SUU<vPrjV+=td)8ZR`ty^a
z?EkEuFGeCWDlT20yqh&5VY*Z_(}8WSqBD*ztqRKZX1vjU+T-B;e5MW3pyEKt>)HO}
zCpIi^<7j!PYkFR+?_JnUt~Xm4)!uf!J&~{V?BZgJf^`Cq*z3|AG`Y?QuX&ypA=$8%
zk2S)PHH;<UmZ`z}AYFy`8Z8m)Sq}K^H1VGw-}L{U<FSR)FEFQw+*CY#c;W1@n_M~p
zYwpfxJ0P{_rP{Y-tsi0<d~@6z_cA8$6=5y=7FhA<m`ODsW5nh&0!gng2qYaeHK;6)
zf1bel&F+>M;|)iyGpk)VKK>U@P(Q4nxa`22$0q%9Zx*eLR6N}7#-7?Bv+RIW*IWDL
zj0T2lcljxZa(`RWnbNkG$@!0#d*eyzuRbefb~1=PW4vJ?e_+dQZt?OJ2b3-bUtl_;
z*86YowEg#45^9f6&u8Bt{3!g@{%R@h>ut*p^GTZS(On+CjVZX{lM91d+@3eW9UWdf
zT5ny5YP@^<7UP-@rpf*#_wLMP+I)Jq26Mt}>yQN5jSH1mN6e8*__s~r03VlPTU5hV
z+4y~u2mCtYGqv29f4O$Pb>3`fGOJ~c;PV-?RkaNAvpJV#F|C<=a^0=}H>`KxTQ*Uz
z?{?uMkG)#wgS4Km65XKvbCvgLrKhV7Z1ULW#hth`rj@m9X^g0b!qdxJI=xlvPc1j8
zn>qde85!%cj5{;$uAVggWW}>T=l-9WX?)0Hqn?v|WW_3`*GCF_xq9{oKYw|i@rJ8o
z!`D*zZ{fXMItlDnDpO|!zHemxcs?Lu?xvs{+o~9^Y1F0uv%X_{@<6K|=NC`oz=^K}
z!`2>{Gb1DVx^TkZI~h3~*V?bnW1O@*g!Nkg>yU(S{~lkr8>_h!b_pbztO#g)TDG7~
z$K{jUjXQ=~8+I4gJa+2fm%UvlbYRxe^*QR<OmFL>wr=VD66v!sO?btdJ3ERRSa-xY
zIy0D^^gZd)ad6`;VX3-BOs_v({&T!>p_k~5Q2*r#%q!&@V(;AC_3M4d&;Rby(u=nU
zBs>p)f2}qmw>n|=n~XY}N4qw>IBFZb=lv?>Jq@SlW#3gik`i}sr}CHR7u#}Ivql{L
zF?H=`!ELV}Cav-pkYzfRm&5Oux4x>uxAW%C*PI4wCpYN_shXFqGvmu)h)_GQY0~Mt
zS)w;Wtr-@s?_vyFps`yrVaqSQt*MGd47Zhi1uGf%_`CmkdNJZ_@auqtx;@jh*FI#(
zmfF8**OK5jV(KB1id{aMnH@X&qN6!CoZZAAEqGbrx^RNeF&_n~sW#j9a<k7~cHok3
z_f6^21KFOe&o)#l6+d3DJH54HGs8B}FvyRFvzxx|PU4K$v^eX(m0`CkL$=)hplIQw
z#hW4yHfzr67WNWjUpK=dh$$yPEFt&Nvkz5FJtt>O3<<M4l55do!ul=lT*3OdM$2<%
zsnYd7gM!45wnxNvCV%BRqt9s2%JjM6QLL}7%j~s<j~T8x)Tx@ijBmZaOQLylz}&lO
zAC4OyIP{9~n(DIW>(mbvWk2wGrm=mK^o5iQA~!<rU0UPVXqZ09Pb(<FTq@yo=;G&Y
zf0xJ{P*ZqYJay?3odbD%2WFjE?~=?bs$FW>TApx0Ju>xqOTT^d`zO!$9NN8{=}dp-
zv}|9-X>Xe=gq({e+cp&@%wl3b9Nx}!TJ_*hDGh_DtutS37O-7s?9G_-@!8ECHV4jm
zGwGOIwcD7xl_6VY|D?+O>nG@UPyHeuC*5c@|IBms@2rbveW<P2-ovxO#mVKw<KpY`
z2{%u^pZ4T>)y;o5M2<Cja;;%5^2!cOh>ZHaM~>A^@1zr>M6UDND%N9D`5tGjow4G-
zZ1R)cpEom@73A!eWXM+AKWXbHU&EV|K0T{o?VYl5LEnU{HJr*F90}G-8LtH&6SzD5
zHU9>k8Cw-43(YsXE1GOqVZ9b~Eox;{W2Wr<r=k(<QIlGkHcT_TzoduBn7QYFT$F;a
zwt@K3pW)0KG(RSE=V~)XIRBigvxR+C#n-LAdyWKeTwAhqxl^l|oI`Ac|FZVAVv-5E
zsZ$sId-{u6=Q5M+PxocBdF{R#dvNh+KAj~JF-zZQom(TP@WPT$(o8elClz1SHu#r$
zp+tc#LcCW}G~q{Cw$=f+ZE06{nIeLJPCfH1U}`|cTQ}Dmd6lc){@DF<iTh4bMwT0F
z%M>LOULL!({*m_oPHBb>dB2K(Z}Rp0e%$Iu<^&OmU3br}_H5j_?9_+TEdrC-ZPu|Q
z=&g*2`pT7a#MbE6HaRwlD-(_~nCWwF=v;fNx|rcwOr2qCR=|oS>hbj|drQ?Ky8q9(
z`s~!M6|LSu+c(7fu|>>s7w266^^h8aPInLMrj?$tcg4V-dh2kF)nMPH_^S-u9kojs
zdZu}X?`7iNVSkw`G})eYmHEr%S=%4!c{8o~(rI&Bi#?+G=g}z#tgrJO@{#23n&15=
zY0bPThV`-2Y$iOf5kGBdbs%cO;rA`7JC-pN2dOhjbeHtAq+h%&BH<LA%OEXj$JV~C
zgW>h`o*2i5i5WXe#2(4mzmdJiDKkAV!JvN<mjQp@$tfHg!fuvG-Co5OG5P1wDevuq
zf0c68P7|F|cks{0`ae>lr4#G_vJ1q0uvq(So(S8F?y6tgf3%mfMl6boNeB!MFK_zG
zGl^;TI-vyH!{7b|F}$8y@+Wadq8Deg?%D%K1i!ZIpTrgV+D~@=ROSfwUQcs1rZdV%
z<8xQbC-goOFJ`y?6lMEX^+9Cmwdt9Ri{g)PZP1k07k`}}IOz}5*~+kUKM(pPHav|o
zdVO|x@Ja5N@CMcPMjHc}%q7eFLH2TcFh~mupXOnhJwfEg9+?Abs<QPv%FPyCD0pdW
z@G&*%S+IxTREdPWObo$Ca#)MLtz>u|px?1!ZCr!t?06s1B9p5%Z)I2QwCk?FZGWrk
z>)XZ>c6YY}FCJdywo_f<BE3hGH~sfvb;lT{1fTr4cl_`7RDELb*i_0IVU(m>aycw%
z0#i<oCl?RnnxzagEN!-{91uBKzeMLqiCzC`#vI|m#IWs5!qe_~{J3VP+P&B{EMXsu
zLFg=wj#uSLS_d-jY+akzU^*xM%wF5%ue-v2+^tL4XKCOvedA)w`_19o#2986o8Fzf
z`xt|_%p2a6fXN>Hht4Z#8kA(Xue;`>o#)M?$j8Gldunh(p3judf<^ob%FS0qHgdjm
zRoIi$EYx$kRp4jdo^XbobrpT>MlqfYuP6Imycwo{VAi?yIwi{^6oTj4CA;=>CD@is
z*9}ToyOi;o>Ahdu_c3^XYppt=rgQ9tN8%1ot~KlLOqs%%b}!zDVNJfr;`a}>gi9Tm
zq^}?2!xg9T{CJD+1}+ha=~k~h4S(xBePtUf#D1OEK7_%iRPv|{hXL0CvmDbS^Z7SC
z`>cI>@$uVi%B5m^-M+*$XwJXSx|{R)9(g@(gR4sq`359(-><fGWr_Ur<~*1EVh6ED
ztP*Fq_gXGtcoYBQd~$F=!e*HRZXa^mH$JTC)l78HeW~N}Ld{_H;pHK+x!IRqq;Fz)
z%<xQe+7FK^fyGP=ucuu(v@Mi<L-c|5Iezo9OI8%gFUa1vZI?c`LCuMc+f5GGeD?kH
znrnlL2D?Vvc8MMKq7e$38|HnPV4Dzi`^XU^xrDhtv@1%CALrX@vTrCeJALGP_U|6^
zSyfufZ^Bm{h^U-uRm6Pq_>{SmH=gv*^yKiex0_UY)nVh0@&ioD3}-^4__R-jv23_~
zV7-q2hcg#hIYaZlXS?*+%7)hcI9713IO^>Zo&$4|Sq#>hXLD~*2wT>@E%|Py?Ifn`
z)jewImpr4Lgbu8;`VjhVo!{}+2|HaHDt|ca-Yom>#jiWvrsZs&hkuzHuhzQ3oa}lp
z_NR9Bk5#Py1s|_ZU{$+cYsX~GR=39b)X^`x8cZ9$TusWIb8e-r%r>cpyIggt-5mP)
zkKc;dG8=9ZJ^4>t&)7IBtkH2^U8xk~hJ8gfk&-s$r!+UL(};ez=)tt*LgrBoxgTQv
zjOzDEAFebOI`E`Z;J`u)vm+189^BaVV9Dc9f4+$N^+K#-@452sSXLbP6H>N3r`+&B
zji964olks)OSUN2=x^So!BEYwq|o<SZ^Q4T)r>dZJ_t8kx-tH3*4N0qTnRyg3r2b2
zw$JZQSa-mtb+1^Z)$hZZYzICuKfjb=aOvxwB}}h5HXU`oJ@aKs#GX^OwhNhiHtt=2
zHd<NvTIv0H;=u_wJ@!7>Q+N4UZ1>T5-xGAsERQ|NV$2-X$>i;RV(y|f_ZfER-B#S*
za%-1$|0eTim4^i*4tIyk+D|>y+pW4+R-8G4{f)+t<KL!j5Z!uBS28uc{`l^XOtJq3
z-Fu!tDCTo<4-R~>d9v>Mn{k$`({v{6F`w<{%rNb?RVM$b{?g@NGt7<Dmu2m%sCw_j
zvEfVT^zSnD)&bV0HV4G`e?OctpF4T8RF|FAx%SI%z6UW*`}n8wmDkSxGdm?TH_7K&
z<@XnN&tG}__e-B&1zJ2Ard)4cNgwOmFM6PTP9DE|{4pJu$Q+G#%+iH%srwH6(D#TF
zjjgPX4o;gBRO2pxyzjTlOr?~Dr|#DE=bhFw>T=}H&Rv~7_i|3?oqFc@iDz#AxGBx-
zW1pJ6(`EvH$=;9aZ%gqNg>ODDdSuq+sx8tpFMoLZ^x}=5A6V3<82vK4)HYS{z@rzR
zFYaBd?wGUZ{3P2zK_ht=RtaI@AG!5ArpjkrXVUr3^e=_o=zMSQzPIwcpZS#;d*-yB
zs3~5v=jO9t&+F@IOz!*_@!;!rYFz4US#48xZ(ov=jNq3(rMiyj1^LWY2R?1Lk1nz9
z$*ttOotIF(d+Yy~x1axAILrFFens-XCH#l~9(VtL?e6T-FG422V`^hfMQ1T3f4gX;
zZ9Y}!rAukVAqgju+k9`gZ%XRr*&z1NX2If*OgsL46_F2H$n-n!lugk5i{@<tv22l%
zhdNjc>}Faxy+0>!C7Y19;>9-B8M)i;Zd!Kp`K!!#dla<{;<uTKOA9UCmCA6<t#j35
zCWD(g90pT&UT5x6zq;@wNB9o*vTyfpg!(%+2@4-^m~{2bWQJ$j3~cvoH5d$@%e!tj
ze_OQe?6%uinZDYEM^AS34@uZ`slzO_;bZH;4hH9A-&Y1RsJV*HICbas>f#En%*7Xq
z+plVFFjhC<FtBqL-{-fHoy{+6-U)`sj-r|O{$9~!GMK7uxb5^!v)g&$uUCKHxy!28
zF*u><=#+R_h7E#!i!vG6j>M)E9NhbqVVS{q>tFi38@B{XJlxAAI^#&ErJ>aEp2c2`
zo_FV4GbHHl3tRc=u03C28!N*#(`m;xo;ACj=RA9bxqy|`5&_W}6*2km6dEcmx|Xp@
zSj^d;{cB^e0tXkv8D6g$XTNiCK56|`kRobswBPL0UwfCP*yB7K)`h4pS7z?fK3%&~
z>2^{?7Xy!&l}2Iww%lzux7}O2v~T|pt6s;Dgr?1Z4`0n=7BpO>A+>SF`W=%$9r}@M
zd!UPnA^G#s&Dv!=tNP6!y-{FpvNhkh%0ukY%!>8QCLeAaaqP2TY2f6!SuAkCB4Doe
z{k^Y6SS1u9zO=`l*><z+c9i?(^QKo*%1cc{<M>0}{{MQB#qD7pc01?QSK-g?%#Q;d
zH_8@oHI{6MwPFc6A*J&x>7Sqf(rZ>s2N&J%P2w(Fy_iKp{9x#d;K1gy_fD`_x#Zi3
z_IdBlkdoDq5uX1sTkhGrlM7#PzZICLa(3c|vu3xYZir1&uyR~xp2EXp@>J>8y6+(^
z0ij{6N_Cbe>_Sa^ZgwZQi@jZ^vU%G~;Vtsqf1>=}d<jh5%c}cy-;e%?S9jDS&-!F_
zx>nqAm*GC%BDcU^FX4kI<C%|49`4&}6Fg0X+dfKd668B4%Tb!iG{vp>K(f~V2XCV7
zpDE4J<uA3(nzhMw|Au`>HY@L*kh}S9m|6Lc9~&PP%-(M*!zXHXME>3X`FlVA%vCzc
zRA9C3_U@bA%oZ87rySZm>{y%3gZz9>@J#YQm*Tr*j?4`fP4%_|5wC9FOEaB*^Y%Gj
zACD#b&gL%t`}|tr|AO+Xt6%X+Y`A^Dyy**nd8qpP&#UV~@2=uhzkP+TpqcZzb^bB=
z9gF6_PY}^s>t%88{i?>SdflFXmHQjmA6BVLo_qaUH&H7w{Ljm&rA;Td?=HXh<<qwh
z9wE$$|Gg|Y%fI`rR*!M~zVgfAzLK?bSpyTe&%Ivle`D*n>{{I&U30(MJ~VqjyN+3G
zUjMHm^C<ZPYqzuXvvgOldCgq2-Tm(l;nN$}{jrPvs}))Oqpae6`8&I(>m?ZMq}5MX
z-@JIw*6L;LnVqjsU$=gjvh;iBS<6-N-$f2@liaSe-e<@5(=SWi`aga7`^o+6XM4+4
z^J{9_cURVF8`Z6!R{f`xkwK)b=$O+@1_lPuQd@>bMg|5Z(8AURjHSD95d{MV1_tf_
z!atO6d|4jPl67#~kN^LF**?2}aGCvoJps9Ga~_2SCHNJ2_ewJ}Fv#m}*za*WE4Vy@
z@3dOV?8BRsC6|XvOJ0q<=D3Z4fn&{!_P^%ssr5hA_fKJ9;K&i<|J@w##lXPQb^x?=
zSj{Pcfq|pvFlY@f*j*evOF{Ot9828EFMalf8%Ua`k&!{cz^4VQqaC!?7p&XCC2D7x
z<sGsA|JOg>T2Rkz*7^ELTf67$+wIx^mtNcVU$l{t!QtY`#G5TS#S!0L&HsNg@utT%
zP%u2xT<#lHreY9xo{52B^D&Qv5@rSlkhS17%;10k1r%Haqz4>yAQ6WI5EtZN2GFW!
z2dwL$!DhhKfYu_zML=et*mFSq;g0{S>gVTyj52mh6={mEn^r7ey8YMMy`mtklhX@D
zn)KD`lfEx_%Wm@$WWWs3#-qP>J<?-hZkTx}b{YS((1df@_D7Rh*cgnL=;!~K{+IW!
z!E~l#oA`&-AhE*_U-(xy_swu`%>4K1?|T6WhBI7mJ6VsNy7g|t*Ed1$o2?hiT9^C_
z7rU`YKj6&mrUMLI9-19d@9qEk<QYHLxYAp&pK<U1sjSaDnX2PYymL-qI4XMMsmlIy
z>u(8ft=K=k<o|Wih-jgRV{e@k7z}hn)Ti;9KH{jDS-gLH$^UTC8QQ<Hh2ql$Bp7No
z?szf3{@2WHe`{7={gM5j^{%}rlk}XeyJoNN`YXZ%I$?{ulU-|@fwi3D7yAo;<3kf1
zL*!dR&8N1z{!3~+z`%9lNt?jlno@>tj+|92-<F8||DM>mI$J=(fpKPxpu1DTwwiYZ
zdMsttx#i#U@^7(T3-fZ@`>0P*$)JHLXyy#>JvGIhvpy?qSP))6qpY#=)TO|skFJ4a
zm4rB+UQA(4>uWf9YWpLuHC8{r)$%pkR_iQ34h}D-#hFK+&ppX?MuxFj|7R|X884IZ
zow-V4my`_}oQgJ-*-Z?cyC){N^VH(E?Hq4*P1*0!m{1{H&B4R)tW)7zY+6Vm%d04-
zgv**6E-wlzR4`yTA#&q(4~Iq7ld{eO%R&;aF4FtddVoP;{eduEYs-VmMl5DanYiDb
zSIMdZMc14Id}|xcI^L}2$njzo6I&{lugk~6#&DA<dege2f;n#$He`y<2)|(6e1JhA
zI-&cTR$s!=`>ZTxYniNdURWO#kYGsRKC?Bfki}GnBPWZstm^<nLLTe3gw>A<4r%lp
zSQeV#%{upF&jAL7_5<5=Cz`E!=#UVsy}^iyx#6eijI~{lO=RL+5-w|PING!?fVGj)
zfv<6@Zm3KmFFVuhsK!u12?i0?=V|rD>&s;uEvISqm08%OirxsUE&ef6eOiCf>)Fxi
z)uyMv{assUa}Bl@;Ox4eEbofSU$5DE=XcHCX}TL$yKP(3EAw{C)hN#Yds9SaMBjgv
zUly)5v(d1HgTeS<=nwCm4;I|tFMssd1J%9H?PHx2rfKzgiL~f4?UvkYpi^~q@431i
zr+>|RbBFz!@D6)7Z@%@{zpQ()J!JjP--qS1Hrd>JHUIAPr#Ii0HTUnh!~Jrm^iT8R
zb~$!yUb8h!-Tq%yYAp}%uJ{(RYhR+Yd9@_dVlF|3w8p?AQYuQf)^4oR`gmJiK7G0N
z#sfCV>o!QNJgZ^wJTb}lw|bJ;bG2!E|A*bEyRdEEgUUy}v+8~=Jt_Jz)ZliI%X*3F
zJT_lTr?gqEH<q@{dDdJp-8`-FuFp2L%E?m8Om1xE=4xa-@J7I87q4nUKU1ovpv?aB
z`Ub_K?jeU0MIx5Y<@$3e`_!B#+0lQF-})wf@@#Wo>1Ly_Fwyq&HJab1_Ik5bH61v&
z{`QF%W-WJ&InLCs*s^SSOj})J-m%+Er`8-{WH3IwP|HA{YsyQ9YW?j8&V<SY^0s>>
z_yu?V+0_u#7%Q+^X@#Ti-k+_nrY%YLp0rx)k@DWQ!Wmn1Ids0(t@Y}2)6=o>e!EGL
zYt5sX(VKGx0yP(VJl?WkeTrKGL$Tt9w;D~F4>wK@PY`|lFOaw0F(EIw`_HO|5XY7B
z!J+2yTn}%XKPl*Y+M}=9yYCN2qh(x~@>XuycMG}CT-=>q<I3B-VpEq1U*s8ahBO=I
z>gCTEZr(qY+$g!);*k_z^MP%}Hg=cFB1H8LO|i6^6l*p`V|m6(oqY*QGD~(n(cfRT
z){X1!4+Rb%tKdHdn{z6*7KODv{b(i1FykkW5C6BVdzg&(PEua?Z~|+XP;z~g(Dp5F
zS4RBy;dmJpebzE6xZ~dPN&8~&O9<R9YkZoyWP-;p=~)NFjvB9BX73*SrmStjtvwfn
z7}D&RFNc0~j%%DL)*>5ztdr^Uk&Zc5y{lh8%nfV%8g!CD*II^aot225vdv|M?FXu^
zXD*o$6Pv|ky=C4>(<KLezm|D>7CkF4WMXd6Ja<6rTF6m7gJ98i@&6G*5uw+-wQd}X
zJpS|7YSY++{}cM^voAJ1Qarh5p8knbMu9iuT4yYuSTt+d(yh7ot)euSAM8s^34HlL
zj)}S9=Gg;_-hVHTd;3i1u>|*;6p134*N#FFF_R1;@=LeB>)`J=w7SY@*23y1E1A0e
zHx_TZ-~X?8VRqv4HKzCGc^q8Q%AoAOq%I@!`pXNm5}szJe3G&LyY<RUy(#zh%<8MH
zUs_)D$3fP4O8lFtp)EHwUU&s@@FesZZTOyesr2cZ*}3<>b&13#BpJGFYI7Ef@SGVV
z<8NX2eq~$N;!pjVjjqc%#Q8Ya|Bkh?;5WZ7FF$u*-=ul<7K`?t4gJDBb?w>Bbq2we
z7vwi?y=rK-N%x#cwtBjnym>Cu?w+vQ6$?$$Jbju^OMaemn|19X_XLLGDUD0jj!on7
z&+1{C+`j&psm!DUN_%UBS1-=2>M$)k*&AZ(<Sg0GofcCYuxb9sBM+aiQfi%EIPHsi
z>cl5!{U;q-R%f7MvG4!nr4t{YubGq6rn6x;_sM-8Y5t*xym>cPKk=^Iw^G4?Va6*S
z9obj;KP}y+2{C1N?DJler^2yjrLWJrjKkT#f+Wo+O|-W;TK?F0i*Wn&FUNhihg82X
zt>{kLnr!^obK*<MnfFdiGH9z<Rk1Zi<i_@2yd6`XMm$@;_SXAE#%v1#i35956YP&y
zHqC#0v(72u(}$}&+MER<JZBy+Icoehxs`R!u3wr$(eWzB4of-N?pc1rS>Gsr(z1g)
zBkjyn5_U>SX1|-Kc5BLkS+9a7bZ0tlU)uC@TWCP2eeyz07B+^>6C3x2mv5>!_UG`C
z*;gd2ewFpw1hF+)T9(0wt3*>1|Ga+xc*cDZy;Z9$)-T+-N-=a-_$IdyUdagO9fvZ~
z4z$gxRQqEQx9Evd<@_nF3R6#2W*TZ9-_H6e<Lw3}J5b}Scv@rW7Ki<_f)kQN-8BUp
zOB*Yr4j!}kTrs8S_0GMjFN1lU|7Uq|cZG<oerd{g`%=q`tgt!eDG9zSCoX>_kQ#d3
zBUB*bWp53;+|qN^njKdS<dxQmftoWj-txR@{&)6oU|gf=R?Vo1GGVEWJNJdeKJZYk
zIW}?kvL#b5<>|f@F}0X}dDYr1=M3LW+rvd+`6nc2-0?WO#{1Bf4Y~Kv$1O34UAQqr
z=fC)w+t#b>PgvgPZ)7}hB|Sm*uAg`%tD3KqSKt3kChea|vGb?<)!9VXl)i~vWOMkW
zXZ6BGJH2Awt(>^^-lRE=PqR~QR9-ssvd4Ntov7E*h=1#6e*gbw?&<v79zw@lgOnKO
zw#<CPa=~-9Z_$I3Gp;k)RGNKes3_W`@1V{;^Dbk_r88+l4fVab7sU?jFA|>JzwXtl
zchiz|u3h>*S13|xx5nbSf4f$%o*eXEkF&v1;L8J<8C$k~vl2-N5_Qkplfvked2iz)
zhm)}k-FIcBwQtsJn45a4O5}0q!`YdWWg{Xcyq)mGF50glHAsj1w^sH~MVU>E%m)-+
ztdP0UIicEoA%pNLi!)Mu(h1(_H}Bor(`webqiEUvkb5zAt0v7__K7vs>1?g{JMZ>Q
z%)w1(ZZM122FNgMXt85FXgXc#v9f!(HUm%70`}(zwgpYP+s34Gwe0yl4TjTCPilun
zgzxOKxT9tmXvVbRS-tnR3Uj$zlVUan=I)Jq5oKbvPho1^uATA;g0j>1O6yNL{eN=W
z^2OU34LF+q?U%jbbgHma<G?NZulhRt+b1wi%c}o#IZGg&ndiG_*v3tz-4BzFvX%Y*
zKYPxkghzAzMP(xrCd`a^^>4F$LQeIqh?Vp1XowjsxXs1D=6Hd*AR+yd{P%|p+G}!l
zn53__IG}dREwk#t$xBC9+=+}d483H$n9F{{k8gd;*Zea4yvd~`A!`CFXOLsAOhJgz
z-zB+L%RiYKt*(>+8&;N(9yE`i{l=A3hthc4Sq-*&TG>8GdHSm1;l0C4%oufAixsk^
zu2kunvP8alb-*b1OYuVP_s=dfrdjV>YV$!Z;p=<$-`RW(ju$>!Z7@5}&U{9TQ^iz9
z?7$YzebYpvKbc)y*IeOJ+L9&gVD)*%ro#JewPwruQ{L&834r`qyfem#B|&xR`qBd>
zs>XJ&`DV_PeYf6{fz9#w>jP!1Gp>3~^V{uvbnou#Qk+?j`g`g=wb`sLdByU0Qqw|}
z{>OSfpL^<dd@yShk-ZUe+tp|7Oe24(HwUuBZmF8DJoPu9@7Hx1hGqqe4FL_2p)$dT
z82EHboAebe&EMX#I%|>^V|ljI<e<g$TR-*A+>YP6I(#`x+qr9h7BO@4Y?`$9ksQP9
zO{{x6gKR<Rw1kE4jg3UYo~b%3A8ud>%eX4J=EK4ZoX_s9oVIHBoSTy-U+BxcyllUB
z(b}`;7`S<YK6X_z=m`I>U)%GIiJ@8H#RS<1ivty+>|ouQ+95nPe_zI!9XxjK);AAZ
z{<g;a9PyBtZHw;<+>3d~V_@oGwIe7l@iudU;Ogv|S}eW{1{_U%HyfW?GxdZPx`+rH
z9k>?t=%xN-zV_TYZ7WS)ZQ;_^9?^5@WsXN#Un|Y|DJ^Zy@QgR;-L!{HH=f7u{r~pl
z<mLYRpF3{)Y&rcD1K6}%jjc>?rU*1|t=xZ+VcP1_+MKspmN}7CW%dlowr($^yRR8f
zwEDGj{nYhL2DTnd8`8`h{=W9U)H$6cGQjBRDHAOQHb()O8JRv_iEf2|Pcj(KdTpkE
zA|vdardjFYnOm*9KPcz7yS{qy{M@pGzZrU1UcRY3`nY>v&W@~qKI=R)zH%@$D{%0g
z;oQ)ra=uBeme1htlInd*Um_Q8>AfMI(77ez`rQrErD=NcqW5g?8Xc%{y!Th-E^9*m
zsbj1LQ8QQ0Q|4`O6kuUbW9kVjoD$Sn&T*jV?H=E_Q-<YfiF+jrmF68QwOaCL*ZIHV
z$0lt1&S_wLBIM>8RmL~Imn|6GPD>}3Oa9k{cwwU6ZT}5YeB24Y0^`IEs4cxyRrb@9
zVYRFJ^sn+pVmE^RU(7H%aHy&05aToxxrU36&n}A+eZm59zDM(tuIBXn{tObfc0Vc=
z4m?U=7n4qr`M%gB=DJG4p||zS8}4u$1nt~^<=O*Kus=B9A(hkPvOlS{+UUT!D|h_r
zj0}ap-~4N_LHpzX)$=5e?FxUB`}8%-4E+-t6})c(AG1o_oVK>El|h2Vu^>T(cUGv(
znzrQUN_@X2pHR3FcIK)zbIy;>mkrZuT~;gpS-JW3Ugm^EM>gghCvFS7JeJK2Y40xi
zUA3C#$uJ327WJKIoS8m<gA^ZYf_hfo{EOn3*S8<36=E@nm;4*H#PjO!V;>9F@BHrh
zhI7NrgEJdy6YR_VA1t5#JZ;*#wHvaTgReu}Vb>rTDig-|eckT^YfApDV!vVf@afZe
z47Wd6T>E?Wd|SEhule!4W$ZKZl|TQ?V_1EQ{e$-IK2C$rNgJ(>bAhAHjbR#3s7%<q
zDU92`Jt+(B|2gUWtI1KGk_@xI7rHoK>u%%l%GZupKd{G9%R5yjA(r#ec4hVhXR6m-
zxy_s)z;dwhVo<}f`h%zD?l$`@`tJ9Pvx<Imuf7%jy2E=l=Z78YiAfb}|HZD{KlS~*
zvo$A^tv@opc{kJSe(xLx<Fc)yS%N7~-uJRLI0}^b@SR!lXhp@lDU4;ix{Ia$UYb$b
z_a{PRrl8ZWZB-h4#a$Z<kM8TUTaj{}(I9_9>WxCJ^7zYFcS<L?8Ly0~odgOF_O}O4
z-DNa;e|_#6R*j!F8@PWj@?5-qZ^R_m4I4i@oL*5VpHX?Lu*g)3FZgAC*ms4<_lyRn
z9#QU-+GR5;zqkIeC}PgpvE{YM=KlEDT0U^-m>Es{^w79W*zTuGT4J1D*^;|kOoZ$2
zSVW7MMX?^e^vKmd+tQrr)$13_|Ap=2IxsD2Zi>&nwGyUm1`|(wQ<hlfCw%-$`mgWn
z_sx|31};wB-yCpa-1gwz%fn0G{xat?DE*iJCaQ4OY)<RUIhI=2#TICF=ltFF=+F<3
zJ9~5A<v(4`kRvL#<d7xX1|Oq2OoxBp`ukBZB#R+IfQ7xxKe(e-^!S}wrH>hcpZ}ku
zTP3{v?xL(-uCxDcl>Yj=vw!DDg<n>`PTfBh&NxHr$t0VdJ7aneF{niaGbXs2zndC$
zgdgn3WB-qwza5fuwIG7w^p>Q<r(Q~XRsX+oOLAH45{)V9>#XL@NPhMDg?GJnJZnOu
z<GLrQ_G|_hXK^vS{$4K5_<KDlHS=9;e9OEcHQO@h`N8b#>bHbKUw+|j=~=qjreNs}
z`9v+fyqgA!78wVpvdx{V%A|9V%QIs_a;4$=3Fo%#`@poOJS+9Uzid#2?_84bRE{Bg
zwZ**~H|H>~nfrh9oUlt4`Q8!HmgY^pQ?+~FuH3QilcBQTin&%kP17a`Wj5&>obY)P
z)BQu5E5-i6u34uhfChdznjRd;Nc(r|+}3OF9$wn|VezlFot9S){k-QhGf^b#_Tp*G
z$wo`I_1W=dJYCJ8V{oGL#7;}*8#V$7cl|yjTa-$IE30beh^#X2cx%S69na3LyPG_J
z=DZ_P6<%-n*QfWWtlOGbtQs-zrb&jTXWrfw3e#9;nmO(D+?{&h?SY_|irYe`JyyQ4
zhNr<%K<36?U&Tqn;^}@3C%<^!eI2#MKyh;Vi&a}Kr>%Wa7x6Y}_UGwOiZ*^+RIZt+
zw8)EdQ=+4n^4qBM#tpID`Ky+`|6U^ZL<SVf4-TBk=016cDMC$Z$(q9Jb>U@m73OcR
zzVh_X(hA*+B6n4PySm<+FQK1zh~b!gxQtjzqO<8)%L6TQi*?^g9=KI>^wic`-M`O&
z|Np}bstFxrZ^SW3n_pY0&UGN_*M>Sb(Z>gTPI+zdo6u{MxOUkM3$@;TiLV!_NObmy
z=hyI?mWn2I6xZ~oUJ+wCER%3(*{1nNPj5fF#OwOEt+&_zkp;D460J7uT_&)ZF+n`9
zq4Z0n!7qW?6E~X5T+`UNf%lyI^GyqrXL5=aDn9+gd%E)Jhew$a`?f4IUaDYx+Tz%>
zbYCg%gl~7!xBDDp5R<&T!^>URe<csdR6(l^9xERTFkD;oB<omjW8uln;Q2Gc6J<rL
z*^L&5-?;SWUvq9}MMQ3S#k8xd{~c|p6WbY4tHc@^8PF|dw=&`;gZG+#U1`CDXHH9)
z?o|q3x-SUwexucfy$oq8=V!7f%yl%Hb3*ywMgGohX2(P~3iWrEc7`u=el7IJ`qB9#
zs%sUW=I1!DEdAv*OK@t_pV<d4Obg$;_0O7(jMFxK=gFJSyhg&V_R*rZiPJTh5(G+)
z@a3%6Jfiq=^G{9bgu7p&KCjt*JX!m8eF;}-JyT)Shc$N|eqx^-|M&TBp8N3&cQ@?a
zYB62yZdh*{$Jdo+=ie(;8ZOc+*w9!up;<X=-rc|7r@Jvkm>&~nORg?CykqNn{d^{M
zH&AluJKpH~REm!?!S?mG?c4kAnTf?sON)B{A((sRuUAihzPLZ%xc=(p#i>tb$DDob
zoq0+3UHqM2^RvSDeR`eo<=0ySH?b$#iXNAJ_WpQuxH@jDVW>#L-I6DF#Z6C!?n(1f
zUGa7MVOCJ^J6dh{%P>uIwLa5<o~Hg^C-a|9C_TT^c;)^5e;)DIpUdr^Vlru6u8wJa
z<u4cQ2}_ox>c-kGP`oVQtGT5!tNBkH*Q5p8R{pNnlQ>{CFRr%OuIlggO@WMNAN!<1
z(Xc@FMqO~n#F)d1jAvGaC{)=!VptsYTd(lpv)Nk~na)^itT=6gkG$sg6HgT{ELUUQ
zC2QCqS-#@8d8hE4rMCYyRK7QyWn_?Gaddxj;8%<7XJ#GY;N6yQO~3w$*>~im$A*Nq
zeVd&<kAyA`TDVceYn5^Ngy^d#l5^4+jJu6@rBCc&SY7kDjTzhkVvA5*zt4j~*vQCx
z*T(C+ImB`vKWU29^tj`z8}~=&!NzqWlb`bj)m`6K`}UaY>#aN+cIMv!*<6rdb|CE2
ztJ60M<EFLAXUCn3vzf%xCwTM11*iFY7pG-s+pjzjte9!={Gnsl`(Hb%8<@{5`kJP~
ztn*FOe-W2plY)i8+Lf(SrCl3l{+?^qaK=r^scP@%$m#Y~GPjJ~b#hvQ{SD_QtGrWa
zUH<FDwhXZY5^L9|aUBqf|Cn-*MYf0yRC|W-=}dZlBBf!g^!A=PXWS-+-TS)w)JgT%
zO!`kEo(1i6TO*}5pY^s?^PHUmRyop}8Ls6wy<7Xtg(c!up2}7CYLFRUHl)nm9N4k6
za;G`7h|;Q*54}@+yn>B#WlCpv-=Adu=cJtLl-K`G|0?`E+386{)SOu+*1lUM7yK$*
zuJAtZTl}|oH<owW^*yfeTcEOoA&O01$j8Tulht?5mc*n<6DGHCn!J6m?9ik5#T}<}
z^VI$oUy}3K8&@65^1l3gRc=4WrSG4ge6Gz}uw|3~MLxIvK_4tW$C#X|Z~C$E!RLL(
zZ?~sgCpfG;lm1`qb?Li!^Skpu)TF-n>$h1v(en4N=9jWJB|nE2es3$Q)_uS04S#L_
zm8yHo1#@F;w^thddUnh$=mTThyp^e1IemMsJ#snyc-}KnyAuZ=-}v+OU3>WdMRLz;
z-NN~|PW`=R%Zufgmh@j+eUbTK=cBbhzW@K(zrD=8KHB_yxUsy(`@i2l)os0ar@MIh
z?v3;7s^5QGY-qn=)yDq@bKguZynNqQPG;x1lQaB{|LVQmeD22Wi`RQ@FP>j+p0~Z6
zf4SamSsUk9Yy2xF=&SFk{^el(Uhn@;jXw9h=d%yWZ>;?pAa!=njbhjMY9ZU&|A~8V
z{hw-`aA55V({Q!;&c)Ay);}=cGwX|>l)Ja~_Add)>?R#zx7^{ryLEp0`xQ5}f9RI}
zDAeSWaA4K_bMX2+iv)%#$2Py(A2fxTEx^_K>VGqRcAf=Ii==91PiJNeaQxi)wd?=`
z(-wob{trtIFlbCppB(RTijjH2oT#hyYb(KOU)FOiXJ&J#JoEgYPYydz!=&gBPC<MU
z4jp`ZCQM;w6Ie7&d{WT?2BwmapJ#;28nE*yXc)(V4B@!A1}x)L@}PBzj6s9I()hz4
zuQC=NU}!4(Xex6atJ*JR8}6iUegB|=k)`rLoJ9hkghRx$sORVRUn?m%z;LKoqmh{{
z<?o4)y0w?@?%A#z_Wtg^>k4NUF|%1D@XcU;*1!lREfX4;pEbZpK8XhnmLO?}BDR9L
zPi1;KnZJqGr!AWU*Mq7Psz(CVoCKdK&XxyeXe^g$FbwnqvD%BkEQaJobBY=a9W9@N
zSw;-UTtLjGqEn2=Tr8(B_jDGWVw4bm3TA1rgINY@r!?6mgr@}afmq9B5`4f+4KNGM
zn&lwV1N|()ro-%!0UQ35Az7&zi#?oR)1OXZ2AK}GXCcfQhyxeF%m6#^7z@OKFsDL|
z2mm{f9qPVQP$R&>kl+&l4hDFDEC4%{4-yQ}P@tM6V6#E?5pN4Ltb4F~MH3v>!F(uw
zfkx?p8Q@sz=`4bpffA0Iu;9QNT~H%XVrxzjG&o2|6m+o!8U_YxSiPd&j3tG`QcW^;
zzYKKBi}`V@)`;Qy)~o8xMZ_H@hE3H|v90@bRxupt^_-OQRn6{Qv&zpkdJG3;H+=4&
zz2fyvZHEkYomcW3cC9JU_Oz?~#`U1#>c57{Z(JYd^GC!={d>$Dw)!!Dz&7S<|9336
zziC}<=31%2BmH8<jg5<owk$5%^7y(|*`HHW*FSmkgvH?8E|>YcE7q4w>`R$e{CUC*
zhXl}My^Hz=|E%Lb#hDK8_hDeDSbE@6esJsQU&&5K>@%2Gf7BOdV2IP$aJju~p0mCS
z14Bo2LMU@%<&AUkkNJJqGcYhTu8l}oHEaJd^%gzQ;b_;?Hn3jbaVYxmcc-}pybKKo
z6buC9qe~7mGB7nCaA-++;m*at!q&(LT2IjEm;hQ%z~n60c7TC}4J7Uea?^v%#+7gP
zN?mWi6RB@|x4!U4BKNELOI<5=@0Wl1!TpzSrM&_R8%T3QI?FUOX7_EkG9%V4=eovy
zqwTRpU72gx(+|7zpR29^mD?vE@z2mHfuSLxQIPj-MXl_P7qe^&VigRIhMSvLI+}NN
z>rH;vcVOGv+ux6SHqL!t`DL5M-n>hH3sj!7zvp9QVLRZ?=_AM;DI6rqD;#k&e2!Se
zO^&oTd7q9XpZRL+_4Bycj8$*e%$(R?mwKN0^UJOO_fPrfx^GEG&U*oe1csV92UetK
z82U)58JsS;9@!{bQflj$TRwUEw=ls#sYc;O&AcqngsYFf-1<9z(SGkG>^-N>Zhx(+
z<G0Iq-QljUmhoTq6rMlYK68(hT*-U&^tsY3YzJ1eUb}wmSB+Mr%B%W&+spMuZ~VI>
ze$KM-Yxec_;!M$--_-nfGrv>rNnriC!!6-S?8|_IERD4@RgbGOt?m$-`a$p7&4*r<
zYJT(X9AD*Y758R-@8qAi=j(lmuW+3=aSk)f{K)+eKZMKQZm)fB$iVPFAYrK<oBC=e
zH7kpir#Rj$$j+~xy!!9Hs{woS6Q3?w{q^pvXsNAHjiP@UFGpNzmhkF#ODHL<kBx2A
ze79P%ap{-otpU<=W!FU}Bo+o`Ih*oxf8BFn--)9umxS+1Our_-d5gGn0>hrT1kK7v
ze-@d?d&<YTH@-d7CH^|EtX@IaNwYI*?SW(QDjWAYuW94C6*pU3{Kf>+{7nbC?E2?@
z5&qY6psVa{QN-+=U7HUn+<iEURZK)(+mrLp+s{i5oTxg)*mQo)?F|Xf*%}!?R5glz
zDbqb$<eKlY^|g5I<mLZwZ#kf2Ik7p##Bj>t)h9dxq5^VPY+?PjxI1=xi_7Kwy)P4u
zOTO>$TYml9|6f~wPnh*3d=0n!HR-7Ac5ShUfH@_zW8QAC*57bsjk>JDk2AM;MeWTw
zcp9#AuUYavI_b!i9jSqlYZ<<7Il#4iv9jQezK*vN=R>u9MEjHE85qRWJY5_^1b;8O
zujlbab<GL(O;SJ3OjW&psQm4%)z&gAZm*O2=5Ic=d`D0t>rv0`Crfz}et*uKctC5h
z>`R7&>DTgRywtG!16olLr)ThVi`4`*<}!J$&&6iZE#9x+CM^6u?V_|z=W22HllL}A
zHOii3-{rxiJX@?xbqZ^mZ9>HU6U)osMn`<Sk^UgEv9{b+=gQ6Uh6l#0cZr|4yZyud
zFY{l`TYQ^$w*zk@;|HO})UuD-FSP%5z87E0%PZ?7=;x`r-DhfG@}{j}ey_e1O0s_Q
z)qfKzFeAzIqmlN8S!xX*qte4v#mjd1HMSo0TzPV;&Bqmyjk0H^n;!wCv+WN*v};80
zM_&<Rh*RHisP69)ZWpV$y=kn|ZX}s}HQT%@H-u^PXC=Xitk4Uy6*jayi~QG{8ki7X
zw6rwRq+awq-*wFm8!R+UcTNhwemHc#KtyK!^^2AcA7+O3&K1rVkYM=lnjp#@l^p7~
zEZ%#iWn=1^SU36OeKn~*e_v?7HWpn}<H@_2X>sfJyPwRoH^|numlkfCmfo(tH_Ndw
zD62?!uD_L{!RPCf_yuaxMBlIFZe;ur&vI;cR!{8fw0Ewi*ZYDuzU9{1P$TSG_Ta`<
z<7$_LYdbn6YsH@m&zSn9&oAlgn}VfG++CBk%Iz0eH+-6MVBKX#iyvpS?j1I3Kfv&V
zpQ(73|FI<@zV=tnbM@G3y%tIlzpb@w;-5F!s@w0YZaOgM;N;JejaNfv_wMdXZ`T&R
zvANJlHTJKZz>PvH2KM=pN7=s1@Ut{Bekf{`Rc-V=C%Wlv+uT{gH;$@rs4|XRdyOkw
zEX1zSblsMq*GAg6W*BK3{5xEl1vMxj;b#d0pPliqRL)<^8xAm3tUNF${rg9`FW=S$
z9F}M^W_n!f@TKvQ*o}+QC)8M{-EggJJdju9A$$Gf+k&}F)mDC5N*i7mHRdj3F8DAr
z@>%dV4dWHM3~QVctXFotpW9e!=a%3j<>V21cS*vg16dDW`E$I{;?CI<w@-FmY(isU
zkbd<F;p4f+vJK7W&*VL5J;3n8pXsq{WsUEnO3kmX?b0)ZW^CT8$NG`0XS%xD2K(oQ
zGwiNw&HQU9FM8wZPLL_V3I4}DSMiCRS7Rx=>%;!wf$`?(JYOMD$nIzIUdEEvyX)VK
z16O8q=Xj@u2(lilh)Z|`N-^q>RWgfo=e)n}&+%r{OuhpL)5AVq%vLa92-7z(u6z0K
zPb1Ug=U%LC?0(`bVSlzA=&E4-7BMHqcU`A3^Le4hN<M)a(2}l?oNFYOt@tzZz1#mp
zrp>jRToR_=7rUXF)mWMH#$=~fHs49f#<ynR2v;y*cqAOLeEs7#PA(?p^<ldMPb+RX
z&ho5xveV@3gS^{Z5?ogt-~!#S#bS1`Q#rBN)K+D~m$|Ihw%1>MZoe<@i$KJQ_KCA%
zrW(9XZ8)IkKUb24?LZc5ST&0qdn1=n#J^=s-kusCxYo?y5wy8FD&Y~|eb0oV$=O<G
z`*<1|KdfXr-Kw>=C695-{A)suZ!2nU?^$_(%Pe_=#)jL>M{5I9d!O!N&34Kwy82$U
zn1iQ*UtB^y*6IGPL*AOVUi<0(Q2P8dn8j`Xb*@HMPM@yHMU%6wr>5TL`!?ag5@we9
zAh&3-JX2Zv@m|UCHygd2b|vVzWLMYbyX$V472DsfoXw%*k#MN}qsdOG=C^W!H{Ob$
z=V4+_kYy^KBvk2rC@*$(hYI(y;GG+*|B3B9@IzzsmbK3;bPev;sQcGW{CSr3S!U7O
zx;0W(M^#za4zzRi@aippF@ar}>GU&}upexEHPVeQFTPm&QN@_!&2*E8nMJB>r`vlK
z4H)b~629pi4oi2u#v8ct-090qr|&GhT|SY!r?kTJ<&&#lm@AhBpR5l~SUr#NO`z`y
zvu`FlXR-dhs%vBPFh1qqQRC(V3=y3NW@ME7EnS|yrHf-G{~CeQ47cy?DXN+MGDIja
zVXB6~{g&8+OwnI?6810V^q;J~|JS$u3LHEQ{z4I#KFt2O%scNb#~UXuF2NbrJ?8iC
zu}rRDT_*K!o0q^0n<-PSs6W$e{8=Wk^uQ5|m)S*6>w69v8`*r1|CgQ4*2uU)tnsGt
zb+u&-A$#hT*4{t5=D?e`Q(u0QS%1KcxBf;s>#~jmUPX*rKa0f<GOb>;v!Jo?()A~&
z)~~pFW7a9A&rhyCSis%LxIw5f^h@ME6SLE6IHrETePxNy<jO0pD|6ny<-Wh9K9F<G
zqjly?)|0$$TePm-&#+TtL-dRTT^UjbnI11$e{yPm&fm}8DUFq%vI;J=f&)Y7(IWkC
zSEXxbPqWqz4)x9D&Ut?C*4=OSpB<dhVHnnUQPBLCz$(^l%Ki+K4jAngQuF6Nv-z+E
ztJ+b|_{rM$Egsf7s5XY~tLsa2yso~1g^BsWR;Jh4dEv6#c#AJ4EZFj9n%Lqtzwf`V
z%gtH0PygPg=-mr$&ps;1o%3gjT1(uqqv4-SE^vefN9Jc;PBWevCms@Zp6{{93=#EO
z-*tC9!}1qiQZ}$$&&Gb2v(a>gF2kOngqsFWV$9#_PV_eN6WAml=GQ*|_Sb(ix)%MK
zc;mb2?ybSQ_G&T<a_2ZZW!}89tKen-@>6=VCaE{fnRe~t<a@3AX8Nq>)>@t3UdRPW
z_wO>A*G)H*Ydl$LYVpD;fnkkn!r>H!cSrV##^;2snQAMyBP{-C{HER*^LzyZVXfQe
zS8|72*8KYRdTWZ^hMa&i{dGPs+}EhDuDJACsPUwAPjEug9Ff^EYdv;vmY08#9c#CK
z^NjH3=5_uFnKtox3=d*i+?K~aT3i&tWbAfJ^IoB&fW+N>=k7ZuoHTeJF78xpwZlU^
z@%H~o?H`uR=UlR9Qujvt_nF2s)%WTN&e#JA_+y?MPlm2ftvWX&&f|OPr|$=BKrZ*=
zeAD%a>3}w?o7z6pc^C6jUVI8VuBm14R8ZpPJe%pqbPS}z)-Nj%432*pt)g9X%eXj4
z?egnMr(WGmoWnHv(UKcatkzAPRdX%(_vPP9*V!-$f6gkn@Rc*i_)oBZafk#%oa%;m
z*HVnPmgl(F%j$;Sjga4XMcE+z_X}3HDRZoijmo`+ulpS~-|emWBJCm5<Z_AS2bNf9
zn(ds#l2clqo1F9e*9CP0!}V<JvFi_<QtSe?4>uWXSAF>5u}=8&O?xZPo?rfrtu*ZR
zwrFp2-TQV`C2xK%YxSOKzBVMG^Xeb>C*hwDNhujrf}*Wwa?oV$SncxU!gs<mDjFUb
zf9BFjUe3nA@LKD5ticBE@I`CWN?m?$`n5jeq0Af~aTSQRuE{}@S6|W9Q0(5rG#OOj
z$2Be$kYGqSx+Sx?zC`KGzdS|rgWC>%JM-zq!yP(%^&IXbPGJ&$yyU`@D6fT^jdZuh
zHF~mh*cq=saO`vQ0S1N#wno_rb3|svgsQfjOH90d>6Q450+4ICb53M~`d~EzPIIQr
zirJpquPoQNTjKZmPp_1vw*~$wI}$I9t=mYM*uPEb?Pd%I6hv5Sm#41t^W5-2@JU|F
z^eOsW*vvifrJILA!r&=GLEF?|z60PfDexfFanSIS3~0Pb1~kA_03Nr3NEd>JnGgaD
zV5P@FBR^pE1rTY_u#RCM-&2S*SVl(pX#gKsniz{98jphpzreb|8VkU~MId)TrJ>fM
znu~7jlwh#6ewL{APGLqi7&I`4(_omr=>CD4j23cGgJpyvK|@6^V7CYwG}wJGg&8VM
z>llGXH7q5dMTs99R304W7>Nor@ClCVvOdn7&tDjr+ZdU7BqR(DG&F)3Y&;99`!v~j
z3=$F!Ffbnre0Y<Qjfa^{!oVPb;rq{%j0YMR8yJ}xnAspC0~-$mGaHEHkzkN8V1SSY
z2_UvX0z<+91_%k!1|jLE8QF1cJW$8MorvtXgaZr^$3aM_pWr?l44P*!G9PFl<5d5Y
z>hjOGw#go6yrI95*G_fKTgBx$waoLKBzcNXdBrQJmzPLCd9pT}+p_i2+6ViW`Oang
z60SGt@r$E<Yb`#W&aam9+9(>y^z*|*?OGk;tv~Sp%I)wzMuvw!jl1Sqb1}3ms}yj(
zxb887LqgW`pQ5#eG7J%&Z!=<c`pX?%J<qa<fkiB#{o^Y2J_UwG)*EqqUL6o*Il#ik
z(9Lgf{Qv`p(ScVVSN-1U>U&(cVJrUzFZ(XG**979`OIVk7_QkJkmC0%GtOUa?>3Qv
zIiZu$?8oD~d*AHeX%^C-e@5{DLquA`-D7Q06B(Et8W<ZuWb8_{0}KX@P|jIR7Y3gM
zh6E;NhlZJ8@dJmL*X-PSRX73UuB-eX=T^m+EnjQ=ROHd^{`2!}r{yx}XdW<g>kq3e
zF8DZSYW=4Ayd&Oalcx*MKOWQ@|8~y)iupR$>q5R=@?Uc6<Dqr>ObyZe8>(jc>^{n1
zJx?~FH$x=h?Z<SBaIN<16-<&0Q}{PzsrxQ-)c;$&_X*3j{X83<7~R}{DDKhP`meSO
z1+N*))TMp(*V#61U38kkdz#3hRhF7p7#<vBymrDhIE_WSl6TE~wunt}iRoW2PBmiK
zah2irx%A5)P1QHJeQW;vaoM9bK6Zv~jtyx+3;ut-cUS$ejMdrmev{QV#7QN*y|}63
zl+`Ks2O0mZWEk#LH5@(K{eJTH_uS6~E_1H=`S9{%e)-4S`yYqTmrGaMu*2&AhwZ!f
zt7l(3mfcY+%f}!ln9v^h<<FzPuWKgnoA~}}|J_by&JFI1<NL3cyUG8Ybi0#}oq?Nk
z!<jQvRf4P>Og}wmR%gHQ_l@WA#F)9M-!EQx$dKU3_)TMbo|VJZ(>@9OMfS4N+blNy
zUh%Sk!Jxh&_vsU69bt#HcNjvmn5<9CH?+}u78>2o+^~~b=e>%2!c60avv=O_@QP*D
zp2BC|<^8S1f+51EA@#9ZIopOSEYtFI|DJh%rl}7U0Gu1*3ZDjgOx?v6_Ch91?QDtl
zK30Z2+XF>&5BBDyO`Uz9@AK~0&)YZcf+)Vj5_;M#XlHi9UC!OwDvR_P8*cM&`1|J6
z*NqIbO&dj}f3-%-=e;gq*f4{^`rNuXzi$?<%uYyey(^k$cmMiBh6F>#YiHNZc^xC7
zw&B_l25a4S`t`3hWce8K1QIU4JF?Nz_CT6%LVTC>S=)x43lB3MXkuDpZf5^BKAq{e
zS!TWZ&I_Mq88%#D&@Sc8sek3x5As+*Luj5^V=rrj-Z!vEeH+fsI#70O+pN1s8u{26
zURyTy9uHPuTNYAk!4Of`AiA#6mv6)4jcetWUEdpJCBv}hB}<u2;p*luOZnItRT
z?Vm97D_^Mv14kND_OfnkD;WkI`vcGRuuVI1?cdXDyY?|LG|f0LNw|Iabj}yj3<;u)
zX2)5s$=zMDZ`ol+2M6IBsg?gc+L;?fS#IpPeqh#O@AztvEw5S3G#9u%J<Kf2#}Fo(
zQ0$Yicz%5-$eJvsZ2s(ywShC48Ky}le2!!GzGfr&Yz7}Y!$Y5hs}GHJ`9Us^Y!Fr3
zkT;XTJ4$Bv$^WtO3=G=y4sfN5EN6|lG(Z1U0mFt#4B9GTdn-=9e#PLuuI?&GQ&vOm
zJex+-uFBHVhNJ1y{@E`l&ty!v$+&Iy`PKPLnRNc0js&Hf(isOP?Vmf9Yr_*cncJWg
zZ*pK8mltcq#;q4MuLvw>iMaFV#ft)lg2N2jRx`7iUb`<ozIe|((S)}T>p(FO+K{TY
zVeymA`tf&bh0bzqxK>~f^6^2IFmH1M?E|Y0UoDfVW0)qNa5OvN>L<o&TP#4jx;b@r
z?o2$%aJ&0XJ99%QTSOX@bYIF-rZ=F#3uV1=WA9?UXRDrseKuxVv+OYAfkRAdYR_`6
zx%4L8-d^{B6v&2&&-|b7{PHI8{iC}(PfcNXZI!^Vrnh{rQTEjm3kHsh+xXZS6kokA
zU=Uca48&a3FDk*Pz#wfU!_e{~pq;rvpcJGoAQZ&h3UY<OQTt>$3m%3RwSSVI!`m4f
z4ya5&@J*JTf$8&f?*s;x6*oQQE!h|j?s%MehJoeT`?#f_yq~yrFt98$wy^N{bXwAd
zp^0n5i$mPUnHm_Fb1oF?FmM<gs4LR|QRjFP7+53{nq%FW8a^=j)G-RodS6%Z>&w4o
ikFhn6X()D}{@(t{^Q*#l{|Wj7vc}WZ&t;ucLK6UX(;+zk
literal 0
HcmV?d00001
diff --git a/lessons/5-NLP/19-NER/lab/README.md b/lessons/5-NLP/19-NER/lab/README.md
new file mode 100644
index 0000000..44e54da
--- /dev/null
+++ b/lessons/5-NLP/19-NER/lab/README.md
@@ -0,0 +1,47 @@
+# NER
+
+Lab Assignment from [AI for Beginners Curriculum](https://github.com/microsoft/ai-for-beginners).
+
+## Task
+
+In this lab, you need to train named entity recognition model for medical terms.
+
+## The Dataset
+
+To train NER model, we need properly labeled dataset with medical entities. [BC5CDR dataset](https://biocreative.bioinformatics.udel.edu/tasks/biocreative-v/track-3-cdr/) contains labeled diseases and chemicals entities from more than 1500 papers. You may download the dataset after registering at their web site.
+
+BC5CDR Dataset looks like this:
+
+```
+6794356|t|Tricuspid valve regurgitation and lithium carbonate toxicity in a newborn infant.
+6794356|a|A newborn with massive tricuspid regurgitation, atrial flutter, congestive heart failure, and a high serum lithium level is described. This is the first patient to initially manifest tricuspid regurgitation and atrial flutter, and the 11th described patient with cardiac disease among infants exposed to lithium compounds in the first trimester of pregnancy. Sixty-three percent of these infants had tricuspid valve involvement. Lithium carbonate may be a factor in the increasing incidence of congenital heart disease when taken during early pregnancy. It also causes neurologic depression, cyanosis, and cardiac arrhythmia when consumed prior to delivery.
+6794356 0 29 Tricuspid valve regurgitation Disease D014262
+6794356 34 51 lithium carbonate Chemical D016651
+6794356 52 60 toxicity Disease D064420
+...
+```
+
+In this dataset, there are paper title and abstract in the first two lines, and then there are individual entities, with beginning and end positions within title+abstract block. In addition to entity type, you get the ontology ID of this entity within some medical ontology.
+
+You will need to write some Python code to convert this into BIO encoding.
+
+## The Network
+
+First attempt at NER can be done by using LSTM network, as in our example you have seen during the lesson. However, in NLP tasks, [transformer architecture](https://en.wikipedia.org/wiki/Transformer_(machine_learning_model)), and specifically [BERT language models](https://en.wikipedia.org/wiki/BERT_(language_model)) show much better results. Pre-trained BERT models understand the general structure of a language, and can be fine-tuned for specific tasks with relatively small datasets and computational costs.
+
+Since we are planning to apply NER to medical scenario, it makes sense to use BERT model trained on medical texts. Microsoft Research has released a pre-trained a model called [PubMedBERT][PubMedBERT] ([publication][PubMedBERT-Pub]), which was fine-tuned using texts from [PubMed](https://pubmed.ncbi.nlm.nih.gov/) repository.
+
+The *de facto* standard for training transformer models is [Hugging Face Transformers](https://huggingface.co/) library. It also contains a repository of community-maintained pre-trained models, including PubMedBERT. To load and use this model, we just need a couple of lines of code:
+
+```python
+model_name = "microsoft/BiomedNLP-PubMedBERT-base-uncased-abstract"
+classes = ... # number of classes: 2*entities+1
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+model = BertForTokenClassification.from_pretrained(model_name, classes)
+```
+
+This gives us the `model` itself, built for token classification task using `classes` number of classes, as well as `tokenizer` object that can split input text into tokens. You will need to convert the dataset into BIO format, taking PubMedBERT tokenization into account. You can use [this bit of Python code](https://gist.github.com/shwars/580b55684be3328eb39ecf01b9cbbd88) as an inspiration.
+
+## Takeaway
+
+This task is very close to the actual task you are likely go have if you want to gain more insights into large volumes on natural language texts. In our case, we can apply our trained model to the [dataset of COVID-related papers](https://www.kaggle.com/allen-institute-for-ai/CORD-19-research-challenge) and see which insights we will be able to get. [This blog post](https://soshnikov.com/science/analyzing-medical-papers-with-azure-and-text-analytics-for-health/) and [this paper](https://www.mdpi.com/2504-2289/6/1/4) describe the research the can be done on this corpus of papers using NER.
diff --git a/lessons/6-Other/22-DeepRL/README.md b/lessons/6-Other/22-DeepRL/README.md
index 15ecb6c..6f3266f 100644
--- a/lessons/6-Other/22-DeepRL/README.md
+++ b/lessons/6-Other/22-DeepRL/README.md
@@ -94,7 +94,7 @@ Reinforcement Learning nowadays is a fast growing field of research. Some of the
* Teaching computer to play board games, such as Chess and Go. Recent state-of-the-art programs like **Alpha Zero** was trained from scratch by two agents playing against each other, and improving at each step.
* In industry, RL is used to create control systems from simulation. A service called [Bonsai](https://azure.microsoft.com/services/project-bonsai/) is specifically designed for that.
-## Assignment: [Train a Mountain Car](labs/README.md)
+## Assignment: [Train a Mountain Car](lab/README.md)
Your goal during this assignment would be to train a different Gym environment - [Mountain Car](https://www.gymlibrary.ml/environments/classic_control/mountain_car/).
--
GitLab