From 01e08be96435c106323d46d3049858f316370d2c Mon Sep 17 00:00:00 2001
From: wangxuguang
Date: Thu, 29 Dec 2016 17:28:27 +0800
Subject: [PATCH] complete
---
understand_sentiment/README.md | 58 ++++++++++++++++++++--------
understand_sentiment/image/lstm.png | Bin 0 -> 131072 bytes
2 files changed, 42 insertions(+), 16 deletions(-)
create mode 100644 understand_sentiment/image/lstm.png
diff --git a/understand_sentiment/README.md b/understand_sentiment/README.md
index 40a1096..15c2828 100644
--- a/understand_sentiment/README.md
+++ b/understand_sentiment/README.md
@@ -12,7 +12,7 @@
|剧情四星。但是圆镜视角加上婺源的风景整个非常有中国写意山水画的感觉,看得实在太舒服了。。难怪作为今年TIFF special presentation的开幕电影。范爷美爆,再往上加一星。|正面|
表格 1 电影评论情感分析
-
实际上,在自然语言处理中,情感分析属于典型的**文本分类**问题,即,把需要进行情感分析的文本划分为其所属类别。文本分类问题可以分解为两个子问题:文本表示和分类。在深度学习的方法出现之前,主流的文本表示方法为BOW(bag of words),分类方法有SVM,LR,Boosting等等。BOW忽略了词的顺序信息,而且是高维度的稀疏向量表示,这种表示浮于表面,并未充分表示文本的语义信息。例如,句子`这部电影糟糕透了`和`一个乏味,空洞,没有内涵的作品`在情感分析中具有很高的语义相似度,但是它们的BOW表示的相似度为0。又如,句子`小明很喜欢小芳,但是小芳不喜欢小明`和`小芳很喜欢小明,但是小明不喜欢小芳`的BOW相似度为1,但实际上它们的意思很不一样。本章我们所要介绍的深度学习模型克服了BOW表示的上述缺陷,它在考虑词的顺序的基础上把文本映射到低维度的语义空间,并且以端对端(end to end)的方式进行文本表示及分类,其性能相对于传统方法有显著的提升。
+
实际上,在自然语言处理中,情感分析属于典型的**文本分类**问题,即,把需要进行情感分析的文本划分为其所属类别。文本分类问题可以分解为两个子问题:文本表示和分类。在深度学习的方法出现之前,主流的文本表示方法为BOW(bag of words),分类方法有SVM,LR,Boosting等等。BOW忽略了词的顺序信息,而且是高维度的稀疏向量表示,这种表示浮于表面,并未充分表示文本的语义信息。例如,句子`这部电影糟糕透了`和`一个乏味,空洞,没有内涵的作品`在情感分析中具有很高的语义相似度,但是它们的BOW表示的相似度为0。又如,句子`一个空洞,没有内涵的作品`和`一个不空洞而且有内涵的作品`的BOW相似度很高,但实际上它们的意思很不一样。本章我们所要介绍的深度学习模型克服了BOW表示的上述缺陷,它在考虑词的顺序的基础上把文本映射到低维度的语义空间,并且以端对端(end to end)的方式进行文本表示及分类,其性能相对于传统方法有显著的提升。
## 模型概览
本章所使用的文本表示模型为卷积神经网络(Convolutional Neural Networks)和循环神经网络(Recurrent Neural Networks)及其扩展。我们首先介绍处理文本的卷积神经网络。
### 文本卷积神经网络
@@ -22,44 +22,61 @@
图 1 卷积神经网络文本分类模型
假设一个句子的长度为$n$,其中第$i$个词的word embedding为$x_i\in\mathbb{R}^k$,其维度大小为$k$,我们可以将整个句子表示为$x_{1:n}=x_1\oplus x_2\oplus \ldots \oplus x_n$,其中,$\oplus$表示拼接(concatenation)操作。一般地,我们用$x_{i:i+j}$表示词序列$x_{i},x_{i+1},\ldots,x_{i+j}$的拼接。卷积操作把filter(也称为kernel)$w\in\mathbb{R}^{hk}$应用于包含$h$个词的窗口$x_{i:i+h-1}$,得到特征$c_i$:
+
$$c_i=f(w\cdot x_{i:i+h-1}+b)$$
+
其中$b\in\mathbb{R}$为偏置项(bias),$f$为非线性激活函数,如sigmoid。将filter应用于句子中所有的词窗口${x_{1:h},x_{2:h+1},\ldots,x_{n-h+1:n}}$序列,产生一个feature map:
+
$$c=[c_1,c_2,\ldots,c_{n-h+1}]$$
-
其中$c \in \mathbb{R}^{n-h+1}$。接下来我们对feature map采用max pooling over time操作得到此filter对应的特征:
+
+
其中$c \in \mathbb{R}^{n-h+1}$。接下来我们对feature map采用max pooling over time操作得到此filter对应的整句话的特征:
+
$$\hat c=max(c)$$
-
即,$\hat c$是feature map中所有元素的最大值。pooling机制自动处理了句子长度不一的问题。在实际应用中,我们会使用多个filter来处理句子,窗口大小相同的filters堆叠起来形成一个矩阵(上文中的单个filter参数$w$相当于矩阵的某一行),这样可以更高效的完成运算。另外,我们也可使用窗口大小不同的filters来处理句子,最后,将所有filters得到的特征拼接起来即为文本的定长向量表示。对于文本分类问题,将其连接至softmax即构建出完整的模型。
+
即,$\hat c$是feature map中所有元素的最大值。pooling机制自动处理了句子长度不一的问题。在实际应用中,我们会使用多个filter来处理句子,窗口大小相同的filters堆叠起来形成一个矩阵(上文中的单个filter参数$w$相当于矩阵的某一行),这样可以更高效的完成运算。另外,我们也可使用窗口大小不同的filters来处理句子,最后,将所有filters得到的特征拼接起来即为文本的定长向量表示。对于文本分类问题,将其连接至softmax即构建出完整的模型。
+
对于一般的短文本分类问题,上文所述的简单的文本卷积网络即可达到很高的正确率\[[1](#参考文献)\]。若想得到更抽象更高级的文本特征表示,可以参考N. Kalchbrenner, et al.(2014)\[[2](#参考文献)\]或 Yann N. Dauphin, et al.(2016)\[[3](#参考文献)\]的构建深层文本卷积神经网络的方法。
### 循环神经网络
#### 简单的循环神经网络
-
循环神经网络是一种能对序列数据进行精确建模的有力工具。实际上,循环神经网络的理论计算能力是图灵完备的(Siegelmann, H. T. and Sontag, E. D., 1995)。
-
自然语言是一种典型的序列数据(词序列),近年来,循环神经网络及其变体(如lstm等)在自然语言处理的多个领域取得了丰硕的成果,如在语言模型,句法解析,语义角色标注(或一般的序列标注),语义表示,图文生成,对话,机器翻译等任务上均表现优异甚至成为目前效果最好的方法。
+
循环神经网络(rnn)是一种能对序列数据进行精确建模的有力工具。实际上,循环神经网络的理论计算能力是图灵完备的\[[4](#参考文献)\]。
+
自然语言是一种典型的序列数据(词序列),近年来,循环神经网络及其变体(如long short term memory\[[5](#参考文献)\]等)在自然语言处理的多个领域取得了丰硕的成果,如在语言模型,句法解析,语义角色标注(或一般的序列标注),语义表示,图文生成,对话,机器翻译等任务上均表现优异甚至成为目前效果最好的方法。
![](image/rnn.png)
图 2 循环神经网络按时间展开的示意图
循环神经网络按时间展开后如图2所示:在第$t$时刻,网络读入第$t$个输入$x_t$(向量表示)及前一时刻隐藏层的输出$h_{t-1}$(向量表示,$h_0$一般初始化为$0$向量),计算得出本时刻隐藏层的值$h_t$,重复这一步骤直至读完所有输入。如果将循环神经网络所表示的函数记为$f$,则其公式可表示为:
+
$$h_t=f(x_t,h_{t-1})=\sigma(W_{xh}x_t+W_{hh}h_{h-1}+b_h)$$
+
其中$W_{xh}$是输入到隐层的矩阵参数,$W_{hh}$是隐层到隐层的矩阵参数,$b_h$为隐层的偏置向量(bias)参数,$\sigma$为elementwise的sigmoid函数。在处理自然语言时,一般会先将词(one-hot表示)映射为其embedding表示,然后再作为循环神经网络每一时刻的输入$x_t$。可以根据实际需要的不同在循环神经网络的隐层上连接其它层。如,可以把一个循环神经网络的隐层输出连接至下一个循环神经网络的输入构建深层(deep or stacked)循环神经网络,或者提取最后一个时刻的隐层状态作为句子表示进而使用分类模型等等。
-
可以看出,隐状态的输入来源于当前输入和前一时刻隐状态的值,这会导致很久以前的输入容易被覆盖掉。实际上,人们发现当序列很长时,循环神经网络就会表现很差(远距离依赖问题),训练过程中会出现梯度消失或爆炸现象(Bengio Y, Simard P, Frasconi P., 1994)。为了解决这一问题,Hochreiter S, Schmidhuber J. (1997)提出了lstm模型。
-#### 长时短期记忆循环神经网络
-
相比于简单的循环神经网络,lstm增加了记忆单元$c$,输入门$i$,遗忘门$f$及输出门$o$,这些门及记忆单元组合起来大大提升了循环神经网络处理远距离依赖问题的能力,若将基于lstm(无 peep-hole连接的版本)的循环神经网络表示的函数记为F,则其公式为:
+
可以看出,隐状态的输入来源于当前输入和前一时刻隐状态的值,这会导致很久以前的输入容易被覆盖掉。实际上,人们发现当序列很长时,循环神经网络就会表现很差(远距离依赖问题),训练过程中会出现梯度消失或爆炸现象\[[6](#参考文献)\]。为了解决这一问题,Hochreiter S, Schmidhuber J. (1997)\[[5](#参考文献)\]提出了lstm(long short term memory)模型。
+#### lstm-rnn
+
相比于简单的循环神经网络,lstm增加了记忆单元$c$,输入门$i$,遗忘门$f$及输出门$o$,这些门及记忆单元组合起来大大提升了循环神经网络处理远距离依赖问题的能力,若将基于lstm的循环神经网络表示的函数记为F,则其公式为:
+
$$ h_t=F(x_t,h_{t-1})$$
-
$F$由下列公式组合而成:
+
+
$F$由下列公式组合而成\[[7](#参考文献)\]:
\begin{align}
-i_t & = \sigma(W_{xi}x_t+W_{hi}h_{h-1}+b_i)\\\\
-f_t & = \sigma(W_{xf}x_t+W_{hf}h_{h-1}+b_f)\\\\
+i_t & = \sigma(W_{xi}x_t+W_{hi}h_{h-1}+W_{ci}c_{t-1}+b_i)\\\\
+f_t & = \sigma(W_{xf}x_t+W_{hf}h_{h-1}+W_{cf}c_{t-1}+b_f)\\\\
c_t & = f_t\odot c_{t-1}+i_t\odot tanh(W_{xc}x_t+W_{hc}h_{h-1}+b_c)\\\\
-o_t & = \sigma(W_{xo}x_t+W_{ho}h_{h-1}+b_o)\\\\
+o_t & = \sigma(W_{xo}x_t+W_{ho}h_{h-1}+W_{co}c_{t}+b_o)\\\\
h_t & = o_t\odot tanh(c_t)\\\\
\end{align}
-
其中,$i_t, f_t, c_t, o_t$分别表示输入门,遗忘门,记忆单元(记忆单元一般对外不可见,$h_t$对外部可见)及输出门的向量值,带角标的$W$及$b$为模型参数,$tanh$为elementwise的双曲正切函数,$\odot$表示elementwise的乘法操作。输入门控制着新输入进入记忆单元$c$的强度,遗忘门控制着记忆单元维持上一时刻值的强度,输出门控制着输出记忆单元的强度。三种门的计算方式类似,但有着完全不同的参数。这三种门各自以不同的方式控制着记忆单元$c$。实际上,lstm的思想正是通过给简单的循环神经网络增加记忆及控制门的方式增强了其处理远距离依赖问题的能力。类似原理的对于简单循环神经网络的改进还有Gated Recurrent Unit (GRU)( Cho K, Van Merriënboer B, Gulcehre C, et al. 2014),其设计更为简洁一些。**这些改进虽然各有不同,但是对他们的宏观描述却与简单的循环神经网络一样,如图1所示,隐状态依据当前输入及前一时刻的隐状态来改变,不断的循环这一过程直至输入处理完毕:**
+
其中,$i_t, f_t, c_t, o_t$分别表示输入门,遗忘门,记忆单元(记忆单元一般对外不可见,$h_t$对外部可见)及输出门的向量值,带角标的$W$及$b$为模型参数,$tanh$为elementwise的双曲正切函数,$\odot$表示elementwise的乘法操作。输入门控制着新输入进入记忆单元$c$的强度,遗忘门控制着记忆单元维持上一时刻值的强度,输出门控制着输出记忆单元的强度。三种门的计算方式类似,但有着完全不同的参数,这三种门各自以不同的方式控制着记忆单元$c$,如图3所示:
+
+![](image/lstm.png)
+图 3 时刻$t$的lstm\[[7](#参考文献)\]
+
+
实际上,lstm的思想正是通过给简单的循环神经网络增加记忆及控制门的方式增强了其处理远距离依赖问题的能力。类似原理的对于简单循环神经网络的改进还有Gated Recurrent Unit (GRU)\[[8](#参考文献)\],其设计更为简洁一些。**这些改进虽然各有不同,但是对他们的宏观描述却与简单的循环神经网络一样,如图2所示,隐状态依据当前输入及前一时刻的隐状态来改变,不断的循环这一过程直至输入处理完毕:**
+
$$ h_t=Recrurent(x_t,h_{t-1})$$
+
对于正常顺序的循环神经网络而言,$h_t$包含了$t$时刻之前的输入信息,也就是上文信息。同样,为了得到下文信息,我们可以使用反方向(将输入逆序处理)的循环神经网络。结合构建深层循环神经网络的方法,我们可以构建更加强有力的深层双向循环神经网络(deep bi-directional recurrent neural networks)对时序数据进行建模。
#### 使用循环神经网络的组合进行文本分类
-
一个简单的做法是分别使用正向lstm-rnn和反向lstm-rnn处理文本,取最后一个时刻的隐层值拼接起来做为文本的定长向量表示,将其连接至softmax得到文本分类模型。但是这样的文本分类模型是一个浅层模型。考虑到深层神经网络往往能得到更抽象和高级的特征表示,我们构建stacked lstm-rnn。如图3所示(以三层为例),奇数层lstm正向,偶数层lstm反向,高一层的lstm使用低一层lstm及之前所有层的信息作为输入,对最高层lstm序列使用max pooling over time得到文本定长向量表示。**这一表示充分融合了文本的上下文信息,并且对文本进行了深层次抽象。**最后我们将文本表示连接至softmax构建分类模型。
+
一个简单的做法是分别使用正向lstm-rnn和反向lstm-rnn处理文本,取最后一个时刻的隐层值拼接起来做为文本的定长向量表示,将其连接至softmax得到文本分类模型。但是这样的文本分类模型是一个浅层模型。考虑到深层神经网络往往能得到更抽象和高级的特征表示,我们构建stacked lstm-rnn\[[9](#参考文献)\]。如图4所示(以三层为例),奇数层lstm正向,偶数层lstm反向,高一层的lstm使用低一层lstm及之前所有层的信息作为输入,对最高层lstm序列使用max pooling over time得到文本定长向量表示。**这一表示充分融合了文本的上下文信息,并且对文本进行了深层次抽象。**最后我们将文本表示连接至softmax构建分类模型。
![](image/stacked_lstm.jpg)
-图 3 stacked lstm-rnn for text classification
+图 4 stacked lstm-rnn for text classification
## 数据准备
### 数据介绍与下载
@@ -481,5 +498,14 @@ Loading parameters from model_output/pass-00002/
```
## 总结
-
+本章我们以情感分析为例介绍了使用深度学习的方法进行端对端的短文本分类,并且使用PaddlePaddle完成了全部相关实验。我们简要论述了两种文本处理模型:卷积神经网络和循环神经网络。在后续的章节中我们会看到这两种基本的深度学习模型在其它任务上的应用。
## 参考文献
+1. Kim Y. [Convolutional neural networks for sentence classification](http://arxiv.org/pdf/1408.5882)[J]. arXiv preprint arXiv:1408.5882, 2014.
+2. Kalchbrenner N, Grefenstette E, Blunsom P. [A convolutional neural network for modelling sentences](http://arxiv.org/pdf/1404.2188.pdf?utm_medium=App.net&utm_source=PourOver)[J]. arXiv preprint arXiv:1404.2188, 2014.
+3. Yann N. Dauphin, et al. [Language Modeling with Gated Convolutional Networks](https://arxiv.org/pdf/1612.08083v1.pdf)[J] arXiv preprint arXiv:1612.08083, 2016.
+4. Siegelmann H T, Sontag E D. [On the computational power of neural nets](http://research.cs.queensu.ca/home/akl/cisc879/papers/SELECTED_PAPERS_FROM_VARIOUS_SOURCES/05070215382317071.pdf)[C]//Proceedings of the fifth annual workshop on Computational learning theory. ACM, 1992: 440-449.
+5. Hochreiter S, Schmidhuber J. [Long short-term memory](http://web.eecs.utk.edu/~itamar/courses/ECE-692/Bobby_paper1.pdf)[J]. Neural computation, 1997, 9(8): 1735-1780.
+6. Bengio Y, Simard P, Frasconi P. [Learning long-term dependencies with gradient descent is difficult](http://www-dsi.ing.unifi.it/~paolo/ps/tnn-94-gradient.pdf)[J]. IEEE transactions on neural networks, 1994, 5(2): 157-166.
+7. Graves A. [Generating sequences with recurrent neural networks](http://arxiv.org/pdf/1308.0850)[J]. arXiv preprint arXiv:1308.0850, 2013.
+8. Cho K, Van Merriënboer B, Gulcehre C, et al. [Learning phrase representations using RNN encoder-decoder for statistical machine translation](http://arxiv.org/pdf/1406.1078)[J]. arXiv preprint arXiv:1406.1078, 2014.
+9. Zhou J, Xu W. [End-to-end learning of semantic role labeling using recurrent neural networks](http://www.aclweb.org/anthology/P/P15/P15-1109.pdf)[C]//Proceedings of the Annual Meeting of the Association for Computational Linguistics. 2015.
diff --git a/understand_sentiment/image/lstm.png b/understand_sentiment/image/lstm.png
new file mode 100644
index 0000000000000000000000000000000000000000..376afe66c61fe53d4ba383823ac685f4d6941161
GIT binary patch
literal 131072
zcmeFZWms2R_XP?+K}rRrlonJe*AwpJ{~>nz1LcEjydKSbDif3a*}6p$#K!p(9YhK5>rA$!?Z+0JM|s=
zH2jG@*69#5v@;h>L`4nY&Qf357gg`Rm@k
zl|p;`?ZC;!32#KC!jdI$Gymt2YF|134J@?Fr@z|gy$!=#zkz0LAs3f{hIWUdFZ#>1
z*08VW0k_41Y0<=VUi`MrVQrqbeyy-FmGg
z*q0`73%MA=+WmIWn;zQP842gyc{X(Opo)9HuK6BK;3vbMl7;+a?}ATZeyyh5S~@DC
zJkqwcyG(C(;e$v!v0sa{-7|;LXaV{=FERy^lkPA0T0ee$lqGThTwKce%w2l9LgK~hKjqVu5b`$+{FWEWCv9nGz>kl(u
z*M|r19)|99tpd6#`wK-@O{cm1b7F10#_%@$$84_V;hC?mUnde8!}|XHOv&V}@K?5b
zYM*Ha!`>eX()h($$~vhPNYNZTMoHv)zB<3A8(6+ZwVr&Xl#=g-ZhePAWWm)O>Vkq^
zjZ*{9V#B|7%Q@36J`L%A@ObR~eaeP_K#yFH_=7o`0YjF7$N?Fh`e;?!g9|9bZ|$Tw
zJMS>B36)d))VP1=^|;2H-*(gR?_urQXX1WJ7r8yy+9DIeczS@t31`>t(HJw#3<;(NEX&
zU4Kf8jgxrkj2PApu`dJ^O@ub0)giQ(&p#Je3g&nkgiVNZlT6b$obGWe=_3ZEC;KmA
zgga;>gXO-l^#uO9lpi9`_)JOhtP0I-$&ZPa88qZREy>D?f>A-Fi89@)Kdu(zjt31S
z8g;S!kS)gZ@HzJYO}XXmTSCK^75C9KINrQ?Y(Qr&sOB5-P<2Qohw!jtu#hc>W-oxq
z=5$A+#eJ)1UvapxIzP_Mu&lH0=vAMyYR;b--$6T!5T^e@XpT+&^!lBfKJstxGF}ZQ
z+9G&O)X}h-7PhjI!q+;qjx@iPO0d@f_&qo9N1KOM99^l&qi_y&on@j1X7@v<;
z5oW$>adnv?_Np%ZW~g9r#HVMEByF$rL|nVfel;>|J48MtKO{cHKU^i^OE`acQbf6Q
zQG^=HvYM$%WmaD{hnQlrhC(&tdib?Ay0)urB5gcvr`jkUKYsl1vD56`Hq@+NTXZ-+
zi@wT=+M$$cx|Z^(4CRdT3(TgsOWwIg?mm-9e%I@6l39!zVb(gkd~KO<`Bbb(S+ak3
z#V^cXxUN(O77k_<|px1
z;*CUR3woeqAiAtW`+Cs3Hy)4tZ0@8#mH;ZoPWFK!H)|jrou(yFMa45RWK-@
zSSy~UxhDIDhQ+_+d4C)E^`2X}w?i*1bC`0G-z>c@cU{liy>%xMr@5*9@`pz?cVB+;
z&P<#a>#ayDN!$Js{$=dT&34Wx-Ro{vmiOZx)qbg-*st&_6@M!pBA(lPN+v2b=yC1i
zh}p9&NfFVirTv`Q=gj&>hIxA{8+WtLn$A3+FBmrL<#}^`z$3@nynkr4R7G28^v{)i?Faw7T7SVyA80Ddu7aQEVhn8Ml{>jRRo?rTx6!Z3kJq@dejW*$=EO7OA7+
zPQF!jd#LTJ+dptwP*pI1ncPy^n~qbNiUq&I6p+4LLE*}
zeC3B?fz5THa!zO3CZ0!fQg)1&IOyyxek@-lx*9$hi
zGh59yFSmpAsl0?5Zi<9xgnSiCQU+F=0s0zGoya6eS@gBRv|q^^U4N?*V?Af2x%%
z%@l?;r>&l~N5x`=2S2;s#Z}E`>JpkQcC}<)RCVmV3`P7!`UJE?bj}eZQ6&P49**{o
z4qMi8N*#$hww78eIEj_lwTE~=7_{VEjCJ93X?D^5P5ql<151oBN#KDPzv`Gm0(uaS
z#d(=e@}VntRvTY5;@=`NqK-}Fl;O^{DBWBlDT*h-Bs;i?760K~@6AEum86RCpZnjJ
zV}icO$4|bmf7H-}S%W7^x}qzl8D8^ayLs#7XVN;|(Bc%E%)=f(88=Vd-az?9VmIW!F4%{P62+WZTC`t4XP0#^Iq7CR^nXhP~}(oL&>5
z6|;tZUd@TQuG;G69(93L%UY>cvkl9;qcUY)BbSD3jEq`4Dw%&~ZRj@X`c=iP$c=eS
zla=Y4Zv9wc8vmK9r7>Q@z_rN{Oq=GC{DYHv3X);!m-i=WirBTW0Ep
z57fpV^{Ln>{rt9|9Ih-Qe>R?tS70b+aLntOS6cdG*Utwie!c6a77%>)NK
zbnvS^4<&{IS-6rd9VevA9p<*eM|GXbo#H0jEA|%jrfU-S@Ms9A!}(P_G!CU!75jy9
z4s9|%r!GEnGH@E5DgU_b#x1}e7ZC5hqPrWuRlNUhx+rSyb!YUh?;h{r+kMH^m(^sV
z2hCnt2Nj#Cv!5S&{XXpH4Ce3Jlm2D8v)RVQYn3(m+H3UYg75p{&~dDZBq4@0D>KH7Ow6ah{rV}$ivk$Zkt^WA(CWC?O|$)cCOo?(!5
zxyD@GKp=Pqzpfk0yAQjbtb1*(t8{vrz3S3%6&lCf=TON~Dd&Y3Y)@`@N;+7UU#
zX>l|Aap~8GFVHIBSW9Wzp`nppLH_lkKUw;;!$A49LYTzN)(}n-~v+z6&%irhz*WXbS(ONGSQrvO)@1KT-
zjw*A<`d@z+ghNoiIdnFW_rHI(H=2)j?WzCqcitaiQLanR6sZ2ku6Wn8PX3q8I^Mwg
zhp;FCN70OP|6|95r{1jomyJB$+^HL|DC(+9Dp>#H@wC>To&MW7A%C{^78VsWPLTEF
ze>|R`e3tLOogwmP&iB0Qb3a;Q_lx|G$HOsG{ogVFb1eUN%>N~A|1X>WBj^7woBss?iXSS*@YPnhw?oN92{{XO$m3(|AfL;t3S5b|?FjFex?3@yrsNjaQcN$sO=+o^{(?aL^FC
z-i+gn!2Kg-?|KW)Y3?%!K-CMOCcTbo`tx*0TV7L{vDfY-9UN>=vi+zu>AgxJ=w5#C
zvs1`%d)!GoT`u+xW8L9yHlIYpF#LCooaRFYb^9AHnb6!j1-6I6nUr(%YFtZQetz$|
zQT;Fc2gG^r-iJMSf3!a*G*Duu-YdHG`{!L%jO~vw@t#H>}ztV-PIsNh^`_
zNVu(YuoaW>@d2Qr{P7r#wd*`R$R1JMc<_|Urisvh`PbS9{Lx#T{4=F>>n#+H;|_zG
z@J%`Q-reMU*m7ntanOI4Vf%ZuRcWHAf2GG}Jei(z!%E%JVV9fUzu+aNXV4iZ;P~lE
zA$aPex=XB{(8&%8K%cc2$g)Aek{nAkP8+LumBBM^ZTgKTk
zi-}c+nwiF1eamH&D?QBoNBcjSd`qJ|97Zk5{CrDC^hGA8$gy4Jkupwqss8Ju5QT4)
z$2f#q9ET_4u;+50oWY_hq!WjbprF#Sd
zJ~tk=yp?E`i{@-+R?mx>_G^n^A=}-Lb(*lDw+c)D(BuD-jJEkLQ53}a{bHALWmSzV5FuhS!%bZwB?wOma;ap;f|zvdglwql3v9lWz`xe
zvznaJ_N0T{A|#UX+EHj?Yj}Qw6ahl$sr(9YTDdo4G`o`?etJn-%>DiC!%!>R{1Sp^
z8)K%0FN-<_M-R>`78`ee*m`*C9|RYEgb^O^xI9wAmT$prwEuf9UU|u}UsXNTIiE{9
zbFE)p|6NCJZA{Pg82g^P{>vXvBPX5?VW82V0|7i9r!t3*W-z
z&(+8u-kEdzJsTNj_4PJsLg;fI_E@-vrT)HgiA8o%>kx|qQsex(tQ{mfi}`Es)B
zSqjcCr_Wpneh^Y*9{cJEtn5DK1)J!CvNz8zAwT#?JwY~_lTG}N4+^u9*DtLNQwBjr
zxmezYGQxw*bDlM>o96R9sg^I|Nbh`(jx;JD6#xF{&zS}*pzz4p&QuB5egAT|(CWjp
zv&z*$-=dA;Tw16-eknRhZ_-#*%)XI|i?h4rE$oZEHmc`4CKYuYM+BNa)=x^wfA0?J
zd-<{Hm09X4OFuOZmlyoqglXYm$?7f9#Yx^4L-dEJ?K7`6lj)}zyK*SMBcy`gTByw$FjAp+0!DJQ12rid&Hls<-fXYr|Cc_CTF
z2;Q#;H%3
z*ke}*r;rnet=A-$qApj3gpdx97%qJ1F;CYrkOpdg-k0Q?frB?CmX&5!z|46qbCk
zTz0}MMSd=f{X)AKZ8St@5zfc^C;!?0&P>qOT)fvaTH@mVujjiN^*w*5QgK_?$5y)k
z-cW-RG}d`$cN|;Di))FcsLBHkg>V19mSwE>F_J4(f`E7BuGlTCW&B_Vr
ze(?9malXkHY;_$!6dmKi)nHz0ij{!p-N}-pfW*1#`Fj5S4p7OrUs>Jcuqma8@C
zv&Y-WL%cir_Xbv#SHvWV@4cI;NDVbY%~i#`sDY
zgM6tBdkx2!=rBGyhC5?@fHuv9eH0XTC80L0~Xa4aE7|3b;
z%J}9NB)3u%+cFprd^WmDEk*~}%BOr^Im)hUZY&PwX!gG5{u)3W|MN7&f)uOp6lV$c
zDj~wS+*uMrE&{og?>+?bInpgj6Z;^VcSl;?x`Su#X(`L9xrF#03u6u@BXH7)?mQ;J
zUiQ=?JuB-7+!6fI}Z9u68EYW9ux62c8!tT(6C^*tgUG*5M*`;d_rj;7ot;)9A;#1Shux3lq?eHqBd
zihY}vc8RI#cGS^7P=TGo9_-4oI2)-S_rSx)V{fYhYa#0PZHPWINTVN*qOL@!DaJ3s
zwmR}yq~Ck@VERkBv}fEmA8qOuhRA8a$rJ|IKY@#-#|XS5CyYMwlC83}WG^L5?)|%4
zMp_8%J%`(&=f2*YmT7UYw_TE60a#ujN&~?BbZg&(&S=I
zAzWo^WzzEJr;MbV&Yd7At#|!>l=Q;{|Bc<`+o!|BIJvXbTnwXcn;FVPvfXASaQqe`
z870vPtZqCuUAxQ@ig{fs_Z-5KOUwoxMYLHZ-nf5PeL>ahlB`qdwZHDa7rtLQPg{5q
z9pfBPE767T@%BS%oF_XorR)P$xtx#;s350odhgdqet2Z1@LBIlnfiT{u3W~iKS#;k
z)z_e|1(lAxT>cqS6GBPi#EypVp_suI;EXu~;3|N|r{=+|{
zu3Wbq<6i7ZeQ7cLzS*=tYt~RG(<;q>-L>FwrD`?pZF@?y$!ht`ISL`q>cyi>{dB=e
zr%4Z^Z1w!ofn4n>w0uL6$0cW|`MEi0g&pp)M2M<&emaDV-$Ba65q0mK<1%56ZJd%V
z(aJoRE71|>zVXSi{r2Ec#avwU_hI6U{vTSbhG}wj5<{jcEkLp?K2ka9M16czs_0`7
zh=!9Ut)MFv5&Vuk{+~|j2-o=R6O-Pw;cAzk<^btxrRJK-o|^4RxUcIEoK}x&v`Wq2
zEn?+qm9$wX2o*g3{Dx=2-z4|R^auZ{pAM@N0vu*6n`B|n`LE^IN9ucCbPbzm`+!s`EE@Og;c}Pk(b)fiY7G8M!PWBV;_cGm
zj5ZoI#T~u_H?dDUR7oZLzj`nS&guHuZ}rEPU!>tGWYD6aUf1oF<>eLyzQ44m{v%3@m{miLZ@GBtb62Ol
z&_UmX`%hU4KF4eLGos80YT;b}`eqgVWTU8w8Fed*q4st3U$dI1P80g{8IQf%00RBy
z>>U>6w0j#t&Vhf+0`1f#^i)Y~()KKAz3&~dX3%pL?yc5M9^5;;m=Boh6JCe;^_Wcp
zT*nc#hA{*oR6LKZ%4xL}6ojflfZH;&{F;F{*9E5N>-T+A%Z%SKkc(XTcc+(L^Q{y>
zhVVSU|KZs%)XaMxmDB>R8p#rYIgtDV>GCuw2yA}9*tm?@c3*q=FA<6(aNb)k)1@B0
zZ8o6Mog{7nJZ7NM!OU%I*l4+lNdA`?RVK#2(b0m6=CpVZQovBE=uo9YHVLCby@0p+
z8fT5T@hw66csKwWK<3?PwqA+j(qSs^KqyR02%TUwwo}Lnk~ynQEoA4HLR(qBn>mxu
zU-ECmi@u5%o~vb6mj-
z|A4~Zk2;PC;jiEZ$X2^azx?kHpzQ#P-SYcv%I<&s*gx?U?2O>oYFrQ9-|+p<&j_%C
z)UxqBl_~gloBy8n$@4D9q4|t9RizhW|3mct;{jiFK)s2X6wdmG+5B4!k>5Q>>e>1V
z6YHJ-__4o>`TukIQ}RAfu<%EghKmk{owWjmi7^LqG#|34WpYHGoGyLzDYV$}KE=uA@thy3b}3r?kVzxC#wi(2`AQhqXSMEVf~|Jv
zf!R<&;sHo7OoB6KQR~-WSK?Nw9VF4v^>OdT@ej8I(OCJPqk6xXUpFDRzb2~B+hvSF
z#{`g1YSxqToK?Se43X&Y1%4@9mhY@M**|L(LI3~=2TH9bEupI|q4C^7zlcH5npoc{
zbYvb#!R452v3b-7d^Y^UEl>m}2*H?&brX?=2ly0VK))^Rw!i*yymoJUm8<4xsi?D&
z_YDx}DQh1spA2C=iHmOvQ$Xh&Zl&9k%g(D6%mcoEn%3l@
zgcf7}aBtPJC7ilCTgUVF?AEVd`8~3yZx~`=bDvhFazG&M({>dZ$yWin)m7eJ8h#k5
z>-ercnyV9EHa!qqOhx$UsMxf>t@ihK3a5!I)$EZk?`|!((g(fz;lO?IWT(%!MswxY
z?5$Ke&ZkE562K=!#R(nmoT7!bCkd||(O_DshBJ#~e%qMqjBmbG$cLIBft8QDiMAsv
zKZB{~JdIhkTsk@?NPc4|NAuQ(ts!HJKXwXnSHLF3$OsUfuYwsWnV}xAJ!*M672$jv
z=Z;gr2OG3kh5D923Zp>s+qOo!&_?4;Fc0#A6WK`A72uFsp%jG?@w0=kQ*dYM6pKoR
zf-{2wJH}TrBvZy1vz|k8rd(OuUCbXX8Pr=8#Qpx`jH6!9NW29$?V1q!Y(VYwGYa#z
za3)L9-y5^KU8PXd2u?UUfrC$@A0E)rupH1URg!N<
z@7}d6-hF%H;aE%ZYu+|NP&)bMQ^F3%pR1o7yO?GI=)Fj5rflgzGOMM2UHVog^7hht
z=m(Wko@eXTRLasJFIB?A&n&UqTqKf*h>(5exB1~*Pid-SMR@Z22wLm|K^$k&wwh_!$s;=7A!v4XeC?>GQ&6l
zx6-`?+u1hDmOiAnu$7#+6fBJ3-4DYaK<2f4Z(AOw^j-k`Rf2Ff5uW()z2{Imj&S)w
zwgu%)zbsKIgpN)14oGE3<34}Z*rbsFN((x+-H~3;&G(N={NDw(?e3Hqn4(PKQgQ$0H6?sXNOKr`0F-yc4`)mmu>an~I
z@3$x1Iu$|CZV6NoT5lpckQ3eWpaq77#r5tgR8cRC@KKZRU*4tZ$TR<(>vZBt2*-Qj
z>m84Y(2~W8nkxEM5GxV!T3~3c2*(0B3xc&b-nO|c7I#4D(rpfZc~ClPc?&a_5qn2-
zYQk-`1axO!Ur=%yq20fq9_N}Bf7iGw2r1V;40~1J$Njx+`AwQya2R5y$a5$a)b(6a
zZ;W@)Y{#onvt57CguOUcQM5-ZAIHyDmwOnW@TU>DjXtZRzTz-zpspzAIBS?rJEiUe
z++!#^zjh%?zt~uji~ImUXp04Fc`Fe-;Xj@CbM^!sdE5!+?XVZNBD!wNC23;{us@d*
z#yYvi?UjwQGIXB4x*L(IV{I6T*PpP!Z_em7IT{3+dCd3eD5xDYSQ8ri7ORz)Ix+^YVo)L
zdmpvREG^u&R~%4B2diG9v0m|^UFOGRiJPd4uBh!(-q;dzTYp~6GBL(^-6`l^3#(Qt
zKi^h1$$NI@;@4e==M$DA{=!-ioV`lNWeO6c4gI#mf{L;nW5nUXUbL;aB7u~O4|HT&
zpWm=rK)aRqd|?MGefWxLvCU|gEm62c*x>gqAcbf#?YK$qQmTS6oUT2jtK3JE*{sVhj&t<{?*
zYq2qgKT3deTMrtlVgC_x>n4d=G&FJtA`j3lyDgGA=IsONNR@O2|sB6S`B
z8?}{{7LaCA9J$MPoOb&R36#sQdr4!+}hfilS88l
z_zH8mahPodOL~=Ul`=tqx=*T1WE6F|)o5w^ZIRy=pp?!!Tb$dJ0QQ4nNiWxigGe>2
zc#P`dwi20?{LWpop-OVdYOW^el(v~%C;l~;sgDeac7#5JSLwP~
zsVW`uT^PU_!a)zh&e5sp5pbUb5|pi2<+yA%2ijx1swW%LU4Dxb
z2K8$hVGtw=Y?&;1<&ELm$kMQpUQVmn`2M8Fc5L)+9yx
zp0m{aL*buw#Vf`mW^^kiU{7DxgNL*9^(A|$(9wRcs%G*>e|#Eh>3LH2<%BqFF>Q#=
zVm}@s*r}=eK2YF$!bPPYBPv{3xa~bfg0)KF>t&N5{LE_S|Kt5Z8LmyTF*Q9=m%|
ze1jkQKuah^g^&A%b
zg})%hHzo!T=|d*YE1YA3{Z+JNntiL32fM6C(q%dlCB41=k*11SEk~oLD!_}nw++N2+O|%aU@fa2^y8enVOLAtf7_}z^ExaHA;iFX
zx=}Cp=H?w*b$Ic2W+5^J?rrs<>W9l@L3fpRv+YJIAq7~aGArt@$!UY7r@9hF+4AFg
zcbI3}BJ9w|AUOs5CL{KAE3BBXon;L^zNTVp@BiBTQnw$G
zRUw&+;9uNwv-+^mJF!7#{*s$Ohk(Zl&bRjH!3mFT-Ljj7rii3dF|`bSWr$nE*EZ~w
zOV)F-PL-L7yFY;86(bTCbYHzgVX`NLV@Fi@&jG83@9FYPMc~ExdaGk>A*@l~v%=$N
zJ6E3cLY>F2Zpl2Mxd>h*e2i*FK-QN|e!R%BRnR8HD7CSsQYtgI@uY^7wTXy})Z?K*
z68Emz9MH~VCV!VmNKT$9aw3k@ucF%4^Jw4N)@CzexdS@M@=P;fe=DsAq|60>d
zlvCtD>l)Ysqkd$1Zz)u#qfYkJSxcB_l7v|KoHL0)@M>4$6DDT~5!wNS{W*dcJB;wZ
zROO&EK@J|})TnjXPEP7oatGv7lddIg1W{X*
zO?p;Usl@TSAUI>$W^^sD{q_l+3NS*S^&a$n9CbG}iE_P-L|%ke4HYD?^>%Z4bWk_%
zg!F;o{=mT{?}>0DgsbdZ?Yp7~s
zYWvX8tb*t{RdaD}Nj6L!NCyFXa|e%+q1yHXi)8f++m^atJu*1shUAe|D687vX&?@c
ztKKom3(rah7r#QADetu~CAfxxGtrX$h>)a#gV-OImAaoouZ&+oMEAv!60;n(P6zkP
zYtW}^XRGC!!w^AUKfd5j!sd_e50WSjgc~7ygbS)+qr|HV_i>Xl$GnaX*bWpPq5
zcj_btC^ig8e~pZl{@J(1h9;pw!&%IK`d?(iP}
z9vy+7Uv3RRZH7_%xq5Siu24yXzTHOXTpA@Y5<=2!28EC
zqa|jMe82LlR^Ct4xEXD7ECHXn2>H^E<{xytM?GX+DqDb#RaBv&p*nKe+ml{G2d8#0
zd{mC*Kd4rt`E`d2f|=_1ar1n{_nJdYohD8+Id9>o7;zq3ns{tj_4g{^)lXRp
z!0_t5R}KDSrD*u1nwbHZIJEB(vaU;gVvL(aU7}6`?t`Mv
z5eT$^^&l3N{bg#0(#ITG;%RdK$~OJI)uf=ZXAe7+&Pgg~Kd|4x%P8iR
z%~$4GF77o43{!jm=#z@u2I+B{K>S<@W{{_j+?8f3Pzg8FHrtT=WU>Md#H?__jlsu5;`jm5ZcygTSR@Wk3<1CD0xlAS6Pac-oUO~*%wL5!E1q@t2bWj;f9yi
z%e_)Qvjo(j6g==04>aw+(6@kUV->pJh>@*dS6ijFV6vu|k46@N=hHXO(IK#{6V31P
z9!9_}vn|M9Aq-GQ3YY8LFHqfxHYnUAVAdi+tDWCeUMkYf(1
zpq3LGoyJ)t@5k+{K+wKyUuB~69i`Q3xwAHH
z`B5oVx(p~R5=5{%+%@yJ
z-uT2i1@qJbB@Z3oxu=->!P@+N|^_Fh`mwd5T^F
z=d}i`v&`kCa{Ye;N@Q$L>nB9YU9Zt`^gY%ADZ3r2M*GvsG7Y{(srrC_C<#6
zwt=94&5Xaw1V!XG{u5IAU>QgnJDj-BNV+}0rciEuKk{c52uFGE5SQSpY}8xZhl=}G
zx}XPkJN!MTo4i@GGxY>N2kQ9YorIDTw$K#fZQIF$T6fbb;d(NZu=j>
z))v1nwk`uUcDmbl~*YJG6Ai&=Q08erLe|_i|zdqNE7^0?Dv3FxAILBz%2St7?!T28)W328bbdTm4qptI@k$tX23z;_-y*Vy>l%VN&U^)v~lelvZ7z!}3VK)tTZy4dGKFyjQU{wW@i%tg>yhtC$;sYZ_In(tH0
zwgKd*+MX=11Ynq(FYGsuS2>kQ(s*#7YfFbA#|x@cx-ga}k$0XMmK%CizTuO!@;GxgHjgC{+v6BhCI%MZf*>tOs;>_iVno&KMS(4r)Ebs(h
zWDihsThr!5M1-Y3QP7(N2z)%!0iC&FWl*G}21Koi7)bwoIwb&p=iES=sa6qR_zSp%
z1jd7Tv3uLKP-`p)j>j4|CiOfv`w6+-mgEuz-`TDksfDf*$R2W-OVgtzs%#{mDwI~pMB6Kz#Hyt}o$Xqt!B$2DgCM7y4uMYdaX
za*VP&lk;hB5VSe(%m$!{lMftLT&%BMsa!ON1k$WH%ND2!^B1zG&Z{7jTGe7+)qDGM
zjuL1-KY-o_=CKqz-UPeHWiR+kP;sPyngm}N-*C=Wvf?_tx
zEbm5OR)BrCi(Rt$WbHVnj>Ewv(PvLp{r(|M0G6Y5K!a9>rij}Lu~$=4IQv=X4ae*NcC)s;HNkoE_3w-3Bmd-%k&>l~I#1|=%9?#Eg|^~{sc@f-%f
z=mW1|M+{F|t+q8m6EX?XO9Rrhvka8tvK6cnq%(BC(v>DlLG}ahv~ZRE{8!ub$4?6$
zG~<#LA3&ClS+dvsQj(~WTrPt8IMD&>LqNPx8Sj4E<5Gu3H7?neH`mN+7>6CZ%JAj5~DHQm=-&h-2hdiuf
zdj;nZl4FDRCyd@80^~
zDaP~K3G(rRj!2I$vGPaTL8eKIy0vD6eQehB3zobBFgZR}
zGL?EBxI&>8`%$SSdWYU5EPWBz$7)I%qhmGPLqKU;+JPUv#UY%@4{B)S3PF##h+o4^%S|(v;tYvUn@~wo~h@NSF
z-zvksw+7N~!Ik{GKa#h)MDbIHsx3x22Zsm4`AIONVesa7kfxBv>Lhc7>uFGJ7Wr7~
z%xy0D-8n)Bzm#9Jn5n%s)LI@cVrbs#!w*t+zc9-c;)zBx|I=^O{Wn%G6-H3^SF;86
zb5y(><)02z^7I0oCxEiMS7wzr4GI00$9ZH;rCIz{f<5L$MY;D93)kJ#K@ds
zBqs-EF!J8vV#vh{j;D(6fK^!@>QU
zBJ^<&!g?XOaOb3tcW1Ni7ALESF(TK^4nb)Z-$5S(FxI+`rS2-$U~+%ngF1J4;TQO~
zdpv?QO2Wp8K^kPN_qt)G=L~z5K?79!JHmyJig_hNzkXiCg9gZn$T(Jt|D+56Qul0KD=!~EJ^=H&6l-FqJW
zW35)%-S6j9#jw@ffoA6E=lXs0H5217AL;^9^$0*2ccn!h%2fs$y`_Nb!wgcS)~TaA
zER}E%K*Kjy%z8kVph)>H0!mHg{2`){n-2zf=IP%fu2z}iW|(c~n&b$Sc<#_$+-!ZS
z6E}&)+Y|cU3Plc>BYExMP(l5tr*YO0K6na}VTGr!gzbWeJGXeU#O*+#P^OH=ioA{W
zIEMkv>7}S?L01Cm#rmS%FJO0Yv5ry5De7yAY#@?_;Ni~0UMV94C-l83%O>h1`w$gt
zEy7zu=nOeWD<~i{sIqAc3(S<)#?~Q48orsmZs56o{}H;|e6-71Er1;UQs{sNN>gYoMkDPfB~b8;jlrQU*SI01oAHxU$@*?#zCM
z7Y7z$bUg<@?T`d=#nx+5{OcGA-8&R@Lhg
zX-p(Ne4GrUc9j`#H(ID2`V@IG@07j<*WT()V9AeGU8nKcG(p3~B+949k_n7suvq`-
zy5t6ATX*6xhx29`X>_}Z5L^03qkg9HBjAI??iaV@XQ#mtG}9;=ZWLtVgF#CbkfbCc
zBEycwh)1eZndx!w-BxS3KIjy(9?)s|M>Nq!X1qsHyNw=g$B>9m<=
ze6MW8DwpQoG|&1pB-YOM%%8JsDuXFPtR+w^I8wdZ?DDyHyM8cZ3mE+yy`
z4l@B^FM>bj=kx#m<3ga^M1`GEX=kGTh2n%L(#=1H95Ob#wK8so6iq^UiEr3+e92OF
z_6_w7NTUJQaQL*-dBBiees!y(b)b-khcP`l41jWKH~t;#VRRU;?KKp(m;@;6k0^QY
zkcA1K!zeDlgWxZyET0=R51lg9KKC*hj*
z__VG2O*!aNjtjaCQ^M5b<49dYA_loHHU~vfWw$?)?H>_6a@taM-1FkebV}Jk?|ggX
z1MB57`0yAN|A!#Jv_T_~eCEsiabJ3qwCn3bXdDVaZWfrKDq1sXFPK=xFSn8DyWfX@Zp`P?`F!Fb2nZD(I5K>pB+h*@oqL4A
zv7y|%5N^g2D$b)(Qm%RJCGQ>t+2@`sCA2_S+TYvXyX^S-7e^EHqUFUnP$OT6ZE|D%
zjTN8zG}&mKjy)c`?_$NJFj0LgmhVmK9Y0)05E%{ULxB~Q5Xo_0lKJ?}eh%V)6k^NH
zHH!EnyGB7hIj8xBLjPSRxef~OHVYx8D2Z%Q$7N*+*LeMf)IGSJvm0_wF3lmZZ}dHH
z`$`*2elg6DrwGcbOu}&IM`X?=K>s1!0C9v-WO6)Rv6weRCo0a=-D3wAN;bV-8Blgf
z`>Sf@BwrSO4Cid^HRei6?41Q1NeUS;0LC={-1QY(yAezb$e>(M%7me|^CLQc^b&~l
z)iK{kz@`H;Bx>D%n}Z#p2FM){$5j-j3Ytf_AZ-ld2^qTBOLEV(8^L`|^c4_+F8Bf!FIvi&4V6b(W)I0FU^IP1*
zz}FsDT5{tj_IK)3Ip&TEJyDkUMW0HhETqg9at+R-7%q|xr$*|Lf9kbmgC;i^V4WW0
zlgTj=c+X%xplLJ=_WTkMTSuPK_fBs<1S!p_@zg)iZKTSzSt|Tk`3qH;6EC4GcGEc9
z#19$T0X{wqj*VCs=NqmtvlIF^g^oU3rs|pnf@?hpoP>qsx%zbzAhQw{{`9j;0!hLz
zwo%%Qg0Yq2^9FcK;vB|qa8Yjr=>J%TiEn9Pb(jvGPYx)8?NMZI$5{AN?L177(Ua59fRUYV@#xpULGH*H&w`^;keuDDC{<^EX?|QBj*I
z-S6f3*TshBlY4sKa$0B&B^&heNvf5y)LFt+MGNq(=Bd>{22rldf`Fc1A)3c8Ub8&d
z-L$jUQqrCc2zo-;dK+D0>^r#t*|9UzDwP1^xorATr0kzR^;N;0$6>(io0Td}LZ8iv
zB|v9*NM5XZ3JdNVEd#a{x9OLGE8BbLGKK~HVZ`*)laIUn%6p@3!g(fe3%cERq)EAO
zajftNPD;q>@-JtbuVf7wM@K0Gdb%X3-5UljtWU7+1jGRK<>
z^mOUQ*2Xk1G4_)fQ~E<-(R{2iKnye}8+VMxyI?x*14@FwoyH4~a0xSL37AU7_hb$6
zryh<6^=hG@iEOhiiWrf;)mf~u3WZhNH2WH!EqIzB?=J@K39}wQ^(KJ=feO$4l5)5c
z`eErQkzA|6>IWeu!2u0YS#dr%<=&b{hIIrfZU&q}!NZWh9|W~{%=#PDCH1L19rl_2
zoTj(1BL?Vto>-ROA-}DHlzBGX6f;c)rr->~$$DcR~
zSPwpN67U{;0{#njQ$E5SYMlBg5$+HxY`riJ3mosOnt9E~cI^oaJRq}zA4a&8_)@vZ
zXMjs{!p$Vr?j&kZcx^Qzk;^A=-{e!^tL|`K+1%ul5f&&?Aq0$C#Tf!1m9Zu4!0omX
zLP<)(0TZOfMcA*|q|{Q%VKC{kbat`j_(=E+r`qf>MG*CW=1}o1GdKL}i$1mcg@r_@
zI2it1{H-m9d0e!V@G#OR4&Zu0TBZ{~pG?i?`>S@L6iWqReuask0HwY5S_st>=GycO
zaTNlxsttEno~3_=NjP0-?Agf10rO?)pZ&Xs92hW7eLqNWHv?}9h6+D`zg&2Ry$EFN
z#mPFaXM1A1AAaqc6~^7(M26AwbgOQfQ)lMh{5CC$@l}sj-r%yF(pZ`A>Eng(XbDqk
zIsAwR5yV~@#W?Ou`YB-6co4YfFwA}5*_J@>fua5J*8*EZU8Tl##vnF(fTn1oCvlZr
zaRa&S1f(hIyP)|$cLtM~3T$R1kl|rn!9$q4PI1Yk5904BwiGcGnc%?hXp!){i|BxP
z7x}AUAj|SwzM*$>5~D#b?>_ibCVFzPlcE|8R)%*|?>0$boWBDa3XTnh2{M8_W#wa?
zHu3n%W8oAI<)F3<{W?!bEg1Zi;xcv{ejNEuN$M39-8KXdUu*YGI8M^1aE
z*p@p9-X;fG_sILlFV9}QdHM5P#|nDGCP^sGWT}PrOVuv5K`I{GM^Pf$O4?WISgIw!
zC5JiTo$a@zNKvdMQkc_wX$tqWIvrQWE}GV*g_<;(x#Gz^b$r$J5~_0*$2l0-b~JK_
zTewTW`+~ZLZwUpWKDOaSR>OGEP_cnH%zcGgegAWF?zklc?jCJBWopw8Q?AbzR8pl$
zA`kE0diy}Jc5j8RcLby1rcp=C8M64ZD`c7RAiEYo&CU4~fVkq)@L
zCuQz4v>zBb^sM-Jpb-SWD7kHz(nNe3&ndV;G32?wE=iVg=CtW;I;XEq&u?U}
zXd1;nPLHvDQKbiQjUc8M>xJU8X{9}Ta0eqY!Qx9Tx(+tF6*4x^;*mgl3<8P-3T*GE
z`l{e!S}!p_)pY~!)JP##T-5@QnL1+Cep?3dE@T+%(6!r{7Ow*+=fH(mw$8kbF+{3u
z_ZWuR>k}|>mYsqmEd_cz1hE1SzE?6akTt5-_L@(gKR4q?-~G
zM7lc^DJcO_TDn8J5kyM5q+7bX?_79%zkBb$^W(_gd#&f0G3FR!9>B}ij(H78$gOm}
zoD+)kV?QTqs2-;mTt^FZZa-862Kp`+awwZl~Lz-?M
zQ|3#oV^J;}piUpH<~r*jP?7~d1pX!8%RZk6Z;OGY{5Mvuo?zxK*KaKk587|elH57b
zFH|sKh8%7Tt%2`jhhk!w{egLkQR^W;m75nnC?K1;1wiTHL?Xn+{Qu|Gi{9w~z4CJe
zki7(12nRDk#_MgwJ@O-l*yjgWMn@J^2CV-+__UHIe&_~OX+O|3E;m0k=YY!74XHfG
z%vdmkDB8uv|6knszb7b;sd^=<&R`o*LwWnS5q)@9q}=
zgyg|(1bC-M*xNO+k%Vsd@lq}-<~;d+KgxdtX??JPTtykdZSWBR^(%KL2)Zzz1hJrb
zY~i~sw;^#d`R;UxGI$5PM24#fCw#iCO{D)C_5aEG9~1AwAu$5}FT#e65OHxXbk8@?
zUIxs#(0p2EXXY)DuxZ)sPO(KRsveQ}?3)2Cz!>fM(=p;40WuOf#@gk$dAxj?=d^_J
z3aZ}UT3AA;fxu1bM=sp5YC&=BFV3b`5zw_}|agQO1OL7mo
z?rZz4BYbwynRYJeqg_R8qWz8Q>g
zEsprNNP0ZG?hk|9sYFS08Pq5MbhQ-Gq#fl=}$w
z&N`4L?+ncts64HE<0J|AFnQ|;m2Bf(bKQ$4>jR1qNc9-bKACz4iX~`$Z0YbqYa9^FDtRWEouq7~5D~+bUGsypc(*+;(m1s`T_0R0MB8Ly<+qXEyPp
zy>@r3E`)Li5W9R$tdR!RS=dz
zG8|U&WF2VkZxFgsKl*zd839$kGRQU6i^i-^?G{o2OG%;_7dkyQCRp7;aKqjb1YRfl
z4eZZug!*38xhk|X^bAJ1Dp}zyLT@S9x@HGqhSGED#PPZgh%ZG~kV|zTOy`<9dsE$i
zBk?nCm+)?YV8buAVt+ne0`+Fg^`LYiD-SBp(hzK9#4>IDnl;nL6vTtY&I6E?v_|jn
zH>nwrsYnZD=$2XSt4#7mcdMm|+-+5Wwah|Ft
zHcqmh)_GSCnPxScEQH#>Or)i969nSI0Sr7cWkT7O+c`7~i|-o4i1=!WNuGOS<$sZE
z?6Dh|Cv;(>)H5{3=QYw)Sg$}eq8Sql6-RWYjTCuT!xIiELDMnOE(tsG#JbWkcY+}S
zoXa%LfOCuyL3lbYZp%MKFbfdV3T;=!m8KAi-&bGZs}*!WOwwv|qVSH0FFb-T1FkfE
z?Q_UTPT;sH28PZ5y|@XbCopc~lwPE<8PWsC2QxXvD(Sg%NX+VF{PC>J-KrIMv6>ok
zOE3Tr9n9$X0dXMPKyLu3H0Nxlfjzsc2(us}69G9SGXNKhuu`C+?e4}0CMKqmBp9o9`y4N(G@CK=%CKt(!4PvhOKHt=r%>dln{BO1f@ST{dfp#dH>
z-+(RX)DX!rgl;*OGk(RYSLK~hE%?fICr3NO61(ujuB&Joz}bGs1v%7_efzW<@VbLB
zozvs~40=p+9xkQ0cMH3SElv{Mt2he3-d%YA;(PXd;O_(D8$ErDfO|%Kr?f7J&TcF=tKOD$6^+Dp$N!jKqfs85D1
z(1@siS6tLLum|8jE61KO1bYT0#WSeUcR-r1n)bYD8gi0mpN`y}H&E{R(FpcZE+q&3
zcQS~pFzwG|-py{>Nqdy{Qm!a4470?UrKw-gWaO?O^sd?#5F>109uW=p5q(50Aa4TV
z3RGOT&+D9`XN>PfP0>;2m1!!O=pJmNHvY*EKlGn`&TQuvl;co2<|CheSoSB}P`v9x
z?cEDye_=Cdco>C*R#a5RTvZFCmN4y~J#4%g=ar8`V5UU2fN~QtgMX)!S(=RrM3ft9O(wM-(N@)
zS#Myl?$fp@32=6NH=3w+yE)=0BYRK0EDB*5h#nKP%hmO
zz9pifuK*~^=(`odoWHcLnfuX~e|HUc>XRs_P2K=6Z+RX3W|HK9%YjeH8L8Q`Hf6=L
zZViGW&I_N}QVOVzE4I`WdP8Nvfo2L36SRK3b9!>7K2}2bhw0d}
za+6-0WH>r7K*UgxjFNhMRCPev)2{#GB(2^2zZdOQQ2p&ivuwPPY6ykc90^%@Vggre
z!Ika+1e_V1gU&Y(Yy=Wr&cs0Zt%Xo>76zrKPcssrtIuIHSoT(burL#xnl6Xj!~*3r
zKVPXNs!Qdg;CnOvvj62C6lwd@RJim-P*9a@=wIsriS|wDna$?!Fu9Hvf(b}giHsdG
z4;id*DAEmA5F}X*g|n#Z@4aN&-B>5RD5T%ncDC10doy>Kzjl
z5QB(f>*v;L%^w!KBA47IY_FojuYGk1`yI1*dX#vYW%79F6SUZy4eE6
zc7@2sTi`TiP+yN#!iM0Ypnm{_86&2LpUzjzNZMi|Opz$_CRXDagf(|qJxmoIt-?NR&k{2PmNqlOy5JQe1b#$w%gz7mHALmHkhFs&kYxU?
zZskny+d!eBRsxXAxCE|5=7d+BHcz=pm3$a5O(zAuzeD
zFUH%yhHu69i-ar^OaT#I$K}Y1fRjI6>pM^MeU#-AYIOuit=p{u>6jGh1xRB5+CQma#VJ6iS3<
zz^e;p3MFt-6@WtBT5O0zejI~ve-QyMB3n3hwS#8-+LKR>VG{9Tpct6gE(Q8P2G{_6
zq5fV!j@zB8Tt84~v|z1W=~&`-=@xxt?QgJioe+hnNj=1v22m4~c%q}gr!fMyD+rL%}Sbjk6RL}oy;$N=D^t0gN{vnXrM?Hh&9Bm6i7t%ue(Ue%9m
zM%Cf1QP|tChc{`0r96c}o~M)bdZ=d!QXDSFDp(%Gq#U~t+wiMga^t6f%T~`9kj6l4
z;0C(H0hnFe?nuYu9HPl?4g+udzyjn3DYqQK3n6PVTiyle4~s(FcId9%g$*!P!Pf0aOx~&bYuOtzLw_hEVX)0r-zFbLOPG>;OU5
z`nPb$kSSOJem*`{*JX@umLL+E8u(hsqU(_0yDtszy@XX%@Xha=noy^Nhbf+M1PM)&
z=hxe+RdmI;>ft!|p2QENTQ3Aqm!r9F+d-DevgbhOn~Ira_wvv#s8-lp8FnT4rI#Ml
zeu2TtUEa3>&;zhRR!bGN^#!fzX7dkZ5OIvP;kn;2iFC>kxJyLJ
zCey#lJ;;C@+ph{hOGJ8(*ij@{`v>>>t$x8VrF~d3{hQ3
zHxgYlJUm*;pSM=k+{if?yCDd+zrhjW&j!M>Wc{``K;8K|_KEoeu!s
zHc%)_`X2#goC7o0-@9v*!$;ljnr!px6~HNFeZ*2i`m@8wUT6aNt%eXt>3!4vJ)rOCt>(X9bM0ltO
z1y8PufOa7{2A{4HXs%g}V4r0ZA6hYV4C-&Y68pK{Oc>J{Gj0HVM2l7mn;a^<@a5yb
z5h$<_{PPe>kWJJFhu`YT-=To*p9p9`AvA+!cX#kjEDF6U^2sUbWFG`*8Vn5>lWNtga%U6E;2>
z?lQ%Gsz(d=ZLJ-JQNE~$M{>gw7pU81bMJ>d5s@Da05qSIH2K3Uijem1ECmoO!(Jp-
z2FzbEkvvO{NTI{hB$lbp19k<08TU|94!Yq0Tr4RyM(kq&GncH2?SapIY&k4yv-`jd
z5?tJO8;X12)}uQdBTC4%V;377!>!*Qn*hWfZq}Rl*I3(G*LX?(1qCFlWsw1aoFJJUT2(-p#cIsw~gx4M)5Osy^Rs3-XbxM9pv
z&Gm=`3NWw|j|nZ3&Ic#j$j&dhNlFt59wWZN0)YD5zAH_)8$G>ZI!}Q8$ipLxZcA9O9B
zBe0aIEcp*V+~{1Y}Gw{e>@XmHOn*jBwA+*IBN0KJVDgi1ef?m06j0?1+UU7sDDjgQ%2>nJBq_DQr{#6US&koR
z9oh2-#?Y|)Rrb!l+}ioL<7U~$&7reH&mX6{_pFO*Jj#A5Wxs+-E>k$y$KPP6IQO9P
zTz2Z0r!qUS>6<*;k_+iwwFeeGUPxS=Kq#M~Ogrvu4wlatl-GVuQ@g(g8bTDFj~tr?yT62R-7E@%#1iJSIi7KykFUOG=3FR5p#6c&?zS3V6mVt?^9$UdbhbT
z)94-g-Rtxy{-&TxFf}R)Rou9alTLA^^qRU@|?_woCAhDpK#Kici&Nq
z%8ydBNHL0=$UZTtA@a@LmJq}hJ%jm0itdm!4~&%dcy;9!b*Hf8R7$Cg5T)1)t4Fk6&-I-<9kEHiQIsnkNnb4u)#*)I>;FekE~dgEfz1;$~O
ztPWuQ5vs;oD(4&hel&iT%crYg&D_mHL0V|rJh-F~Wz<7;=cVk;XmvdP*H`z1
z=1^j9g)6XiolV;EDkxu&uEjlOKqWG)MwS};#_}&n57xwaKhoVbx{G6xfBkW5YQ;>0
z^^R>&4++;d-e#*4-<@`fHQH7!!M3&e82>%xIogDXm8(COl$Ek%LWw!h5^q_wJ<>GOcRd+($X}960_l^g
zFPYvrpXXM1im3OA;MbyK88?!27W+Xi{s|7z7KbMvwtC+cDu?71vxlRp$vnP3z2al5
zByiryMb-*0c~fzX5)KnQ>h?i0XCxzK40XM_9!FJ9
z4-h(D#(>(^-za!Z+?KZBA#3)Qr}YjU!c66BPiS(5C@aaja_LLS$JsmVtBUW{^_AbP6MKz&w9F?u2R2c
zMtWg?=Hph^%uk<9`8cC3b1zou^fU%`PoSuWOn1A)wXp}B%8$+m_en$+!#Mhu{#D@=biq8~
zAW3bXg`k@89U}$hExqX*U6^;14{J-dE>`{_PqNT`p=_&k#dz{+vst?1Q&OY7;I|?a
zO0Ti!*w%ca>&ze446=;oJ>qR3$?B#{yL)2}mKDs6%gSp`%OO%G(yWK{IJg_VwQ=gd
znSyFN_^{~CVBj7GxKbKk%)zaiC6vGZ-L2PsJfE;b<>O>;n@L*!SsZFp(fpTo(Y_vH
zpC3ol{K*r$;lwWFrZX>fF_BD}b6{u+h7pzvcQUOiXoQls;<*HvJNDaspR}w=R*VW4
zDOBk92HcywvHp}Dr<@BNytUZhbwbsZ9JyWju8;hjk#Wg1g?iPs-UHwCDBFN5e61I*Mte?ldYWB3MuT$#cH$|I{gwX5!+xes=xd#hYp
z`W}xi^kuKaaP}%pwIfbidi&vNLuv(Ahc|6Z}JNr=@{oZ#-GIfc|V_
zOFPjcas)X`QsPX#QHSYEGTa0j4057oxf*%xlO=2UrHaE?F}8P+1H0rpR>J`T4XH-o
zqWWqS3EqslWqRf=ViVU0q<@y;r+G*5eq7SJ*l1Uyj&!Qq&+1B4q**^r(VpFG$0zBz
z>m9uI
z)8dEiHl~f_-f!rtJ(so$t=gogPf-tSFU0b){f^LdDCppf0>=`5uc_`xAHFRKqRTa#JS%4l|z53w%NW*Ta2gewt_CPt7r!?oKfz`%h;F(($%{6f7a$4
zavn~b%JKwR4A=~%t*>B|ydjSSCJe=fr7r(GLv!7UQ-`&aYQq*ys(Hg}!w6bhRgT{=
zTt>m`G-YdAyM`DGBERo-aUGN#-Z(t+)ZzF{j)5MUAaAfB>KHH&(=Nv}`<^;s%a3<1
zVA4(9u@z;k+B3v1^K@K)pPx`d9SNzO?_E1_hMabWTcZKPo*K+r6@qaNy9)M!Y5wy{
z^2W*cwoR{~=#}L{nsrL{n8L%#sdM!X)}LnUYh?$1wiswqN~cjruss_RYIdP7
zE5#S&DOTyIwZ_Q@e9+nS$yv2rOy;kHj}AeKJ0Z`CTdN!J(j+)#CA@5PK$a`!D6(Sj
zva-cxYug2#iiQ9leMG0%YM@Q*Y3987p;t|CgsdfVD?b{K7*~{T3wNwot^y_T3^|lW
zLd_3r{3~lX4>!>FXA&QZWdSs|w%k&_+*32kbz3R0^4n;9(QKHOxwpjt+wG~Zp0wYf
zuYYCor;-SWR;EKzMu7pr<4f)X?nz$nHJC&+56Sx3Q=y6u;@Taq{(92Tzoe>L2++C-
zj`ma_k;t2L4K0C&s^hKf>hqFU#)~hQnnJ7;rDMaGZtM2P&g7fQ33@B-U!w5>hO1&-
z!SVvE{M$q|tz0h5zAhXGulVbow+%gxvLZ9=9@vI;BXb}Li_uTzi(b=-PV~JtO>zr@
z#;T4g{;jr+VB$;iX7s9vv!lPfre9c0%J1?`0%_9~(82YBMxxx9b=4|!&1txH!5h=&
zbK8}I(Y6(`zed1$aKyN;yxnA$KI2g%EWw0U
z-}gL&6?d`P>`D~DD+r1w
z18A-&4G
zYRcVK{cR1#29`?6aIFWvvZH=EDc$qL4^4dgE?(lWU`i{fIl#xw@_7i}K94r39*AO?hm}hS~r4x<0eicK*Y0
zvyj*IVWvUsCz00LjjR!2Gywdqn>4O<=@O)gIrrW{Sgt7eqf(DMC5Eb&tM&^S~KIVT&MJ
z+pH>W%WD;2Z{E?Uy60>CsY{TngSy#(y`wepFoCf-eMKYNvt(u~lFTqB|LqwppHJ2D
zl7iInVrMXke+RjaJbBed@TnryJY}CWQ#AC1I35$L#G69Ljsty&b7?EMZ-TEVx26vm
zV3(#+9ANnL7QcC%7h7ms8E^KuJZ+0pv+zcrUJZeZ%h#x`3-&iv@o~NBeZEQ4Pg&|Z
zw}Mb`Wm)gR&OP7mMg`q9eRi6xs$=^w7tWy=4FlhJF7s|m?%S$LnQ1kdo&+NGUq?;M
z9N(ABF!Pz2%{F~T!oo)nSM$~bc`8Ze_kFieb1B1;KrItfRLKY0i;^eqn5S@`GZVcc+$6j
zvd3h(V{?|Tiu2|FwghW$+sDyi!gkt3+0xV9%?~zfM6$PxoZs~Mvu565G{3>N&Zk7G
z!*`FM{C;=wJm0}e(y_}g18KC5eaftfb%%UT`E!{2hy??o;ri|gFaKjK9KGeeqmwe+
z?q|4gGtXi28$WZscM<+`4GXT#r%5_6bnfp<^?`s%vgTX%vV>I?8`iclGL%
zO(7iu7mkm}c-jRQ#3brj_UA4<
*Dg#f+Z;z|PX4e_^EvLW(g=K#a0IsTcRL|lN)|64iyxKDNw&4%t20mO`L1NLn8W|_6Iooc
z=rjIM^(Co^SDNLMmm5t>gd}6=d0iE
z@Gi#b5X7A4`T^a!X@om<^R=0o!qB%RFzQ1Ld1v}nby~|L2vLl)5w#3j8hWX
zn?4fGbN8YR2O!MOCY@s7xYkl^-=|MVZ+U#rOZ{Z?ERqPc@er7teOr&
zp}7vHETi~LqP6Yk5|c2+oh_r|ltVeM71bm(8Ljy0#EqCYb0pCSncslnz$B+rP;1?{
zOc81iAgyR7w}Eg^mhj;bi;$?s;Z}tCtEQm7VzXyOm7V@1M^1g5)Tp$hD!cld}(C*rvyC|)?`_rqkFEv)NTAQ-r{MUzi?Xd&
zi97AS`icCC;TI%j(?9niZDTL*-mo^g4&S@84Br-LS5~OhOA>2Ioz-w-+7>j1xQp~U
zR?6LJ)<3RV+%rQN*A5+>*lcu0EI_9CNV0l7BMFy=@!A+i4a0~C$uhFPK{6|}C>HnU
zj?ayyo0T&^2(laocdUzPg~vHY-WHBL6>@}x97m~rWd(z2f9JQ$nyr5fvQZ3O6<(!?B3Z#;)<^Uk(iEIpm)2-`=|`^Ycokjpv-REc
z+9?#8;HcIX>WjtHENrn{=8SxcX2GIU^4E1ALnF8eCH+Qir&o$)Sf%<{_vXlSIP4G
z0bWI0Hz$OOCCgJD*bHm#;w+VWmMvhJv$$>^Di8IV4AZ`kMRO
zYO0+T+Bz{4tMi@3-Cv{qK`$h%O&bE8a#1b5H
zY@H-YA8b;KldQb3lGVuX_3rg_kzb%(NwjMJIe0nvS=ET|T5Oa{)rlSKWO-hqu)w3e
z5g8l=gUDBplXrJ^W6+a!N>z`S+m5GUTF_?pW((wzQ4LTrlIE{PR`i$QRyr84(xZKM
zCp|?fhHr|Mxl`K;2GQ#r?`thx2uOamGX3hu!(6hCf!pEW{=zElWL5R1Lm&%G8QkP_
z$PDwd=tunXyk?$U7pd652L{4nK;#mK{0Q6jEgCVtMQJ+C6UO&m8_
zPza18_!lFV&!E|YD*D3lXK9Nedz0+93wCgNTcgXVi#%!V^?(n)++HxCI~O!PJSw
zD6TD3mde7j-g9=s7rKu&E2Jm(EHi-z!h?(OPOTuA+Wx
zPHRXQ=S}fK=ZDNqk^YrF8-QDYP&+FfpJltV~g*5
zKijgD$V$D4qrH$k^!e#-|GExVXT%JG!Wo};mEMcNoJwZu$<|W400(nNv!maFfh*=rNB`&W;#-{I#uWvk3ixO(M4+#0@3y%644%FfnkSQ(%xE
zt@|oV!ZIwwfqf=mYZzylh8#4STxqsm0p4|S4Gr62f)V;y?S|3HTAcB)8aEgEPT#Ep
zRYlE@v{T$N66)r?t(&Y-xUhG0
z@1XmWsOo36R?S2&MM>RhndWG|3}721Wxg3OP7zn_6k=YZC8ieXPtmfe%o@0zvUwh{
z+Vx;jN%#>%S(Wo(TbO63Q+n_ypGQ_TUX{D{;LcR|>*gocLC*6#z$7yr6s$;@i>(-7
zb8@wP7zSgF@Itwt^y5bQ;~dVpLz*f^HFk3$FU1F`2-tK&>`c>laxPO-&bXbFG1Am*
zC`Ag~;zf!Z6^2;%=mGKUtJ4-eyvAJkUF4iSqxah|KAQxGEU3ARZpD+yHpyj+-y4to
zgwC5%{oH+3!+?4-Sa(h5U4(Of;YV7ugMQ1Wn5xSlGP!fa!r}VcmaAV|xbtS1#?q?w
z?Jv$CWBC%xO-6d;r$Vcqc@fB>GrfYHLLw6zU5DN=y^ovf=985ab!FmXW6JLtN+$mF
z0jRnB6d5ZYpH=UXgF&q?Tr`_jHnWQ;x@
zp&ET7d8fIz`-~f~nS>oZQQn!&Rt{l+~bc$ha29B${7sZ!}8jMfSH(7x?4V#1`qfSQNT*xxr-SYO0
z9MwybC6r9|{dCD^;pig6iie{9l9_p%HN6iXhL+If+*QCdx3}+@%8Cl({a&te(Qf2o
zz&Uf*71+usrjAyxL+f5EZWIh`MdUfAGX8HeNoThbaTY~IuQjS4v$XL-iq6)cCy6hxUg8dUFcm1!ZS>9YnV
zSDu#h&RL
zmI)1CM~;EoI8O|hhlz3NTk*5Ry7|9%+7^_9==-mV9~OEapO;Q~H^a~|*(q}Rhv3bm
z%F#s@GL%Q(&sR17$d(A`Cgt>TNx5+7v1ms@E6G)ooZeogYd%
zzIDfc@ClmVRaAzgVW08KqqO)Nr+*M~xa2b1yxL$3EWG=y@~3jWiJkK#J3C~ZL
zpE6_S`nHE8pkC}X$-mgepXy*gwJ`>!yxpImR&xu?s&^&qy4-({4AL_`y+*9Vw`*hDeD}uKecX;6%)JD6HyEcDco1=Xdn?F0Zt19{1`Zq-fbm7tjJE#Z}p_adzY`Z!LKxUL=3Gomfm$G;XD3)?KlR
zTE&AG8VQ|>d)}lx-0wiFkR8O?gY)FowX%tA?(4BKL$belM*evLkg6`I&bvJi0$m?g
zTBPlScNV^-X^o&QboAPi;~GApnOY^cZAJAv6COMwW7;r!T;*KhqJOM~#mtYS-!sTn
zaExX+gzSs>c`OXvsR;NzjtrCMAVle4
zf!Fq*k9%A?<@`F@oeET>t3?H@$r&*+beDgfr_|lB+T41CwFEZv6Ko{FH0?N^qZX;v
zoR=4vOQ0SkulIn4xl9j`o<;f$9|@eFa3PR8AWFxBw_EbEFH=)}ifc^OWej=~Uh-vC
z|6$nb#$(#_-T;yDz-Hr{2*E>^>rX-LERZe>fSH2rKWcg~x8I?-y99**xEt>Z##
z-W6pCXTzKnl5gaOyBc!*V5bNs@G5MpC%~DHz~A>E7TOrG=ZGWR<&+KJ3k=%LM7>Iz
zNXK@$@ljTDzGfdBka$c#!I$E5^nQi}zSBPw$c*7a!^|9aa+|*`V&b>xLIL6}2O>1N
zg$5uSc7qJ%4bUII0qpgioLX7N6WXb8ohJ6MC7@|dMV4bLF)S8yEOX-lsFcfEmHi{e
z)BX1kA^5s$@O3rI4=>aF9cIy)0>2!K1Nd7c)RaKJ`4TT3VEy{;f1Lb&z?j|u6l-BX
zO7=hp=)!k}Hdj8I?i5o02*}4nB;ISW+kfi+l!5-der>t0tY>OZJUtIrqQBhAhkxntFe>I%Bv4=CzUhiOvjuVK+-OexD
zqK!Z|)VgNh9f5D!*(dFW-M?^)ELPAMa*~fXIC=Qs-K`ZrjYW%Chz^wsl*7X)(`Qw8
z_B8i>UmH+yP!1rLg|PAM!XpZ@P{}5fGLY!-MN%9zoxJ~dtj@R{lCJid0pyYhR8Y36
zqGVYIa-D6afpGc0{AFeTVK^=wCgFi?AfOxsQzq?EctPfmmS$zVzFXqEabI_x^g0}2;4x4iJf#8BG`Acw5tsyM*v_5+yT`bm*8Sft@X`0z=`r<{aicfy$0lN+wUsc
z<3IR(LWpmU`S+B=rJNEU>sD5{kd@RWzF!JA4yX(jw+ZjP~~tJp6CC
zlEYgB%>^PkeKicERf8DE^#^FQeY+SAwl!Fjaq{=N(LccfFwcBMYEUmU(d=3k3zLWL
zXW5`C;0)MYbC|M*?7)vVR?H#4|KY;v*SmFb7T#;SHK?w?o3UNhA72)H1KzC>)gi!$
zYedAp2vrpHhlZZ_hhp{I2{%pGfCF4QA_!YdraCI|Zu=Tv`}<3YS*~M_TOnf`by+l{
z^=}X#W1Za5SFdV8eC@7TXr+2h6ltd2@^dTIm#yFI5%iv
zf>lt$oOE1WPGop0=RoB)Y$hnak1JUWpk^Ou=`6hXE;98ZTk=N{kE|RAH&ZFkAgo+dd_+{f9ptz|)4h3<&XZRs^jA3pdi0;@=z0
zbb{-S*7rVOtTSmD^(d^KrEUu23Yg>_|8W`Fp$2Ds?Zahc5^x6gNR1)>!9uKl)d2HN
zbQ`W&Tfp{%va~AnbKu+b6QZVwH9_mfe!md+IRnOABu_
zPzP)3-d#8pQA$_%j4ipcx|(ZF&|TmL9PAXqXHp1*hUR?V*I*W4U4zF=<_(@0t%wr8
zIJeYt?kQ}~(hU00Oa}pc0=WRxn)dIJW8%htQIYZOdcp%ZJI_Hlq8Ta861_~OJs2-d
zR0gU(?c+RX!pH;YM^Yep5ES+y3x4a8aL%6AHGjVrnb|z$Q)E9wS95V7&wx_m>;(9V
z)K2Aa$|wn()7{?zV#PX?90@lusEb(d+NBU8xwH!xF+;Ql5W9~H{*6PIqQS{Y_*I_+n&C3>aJW_Y5i)zQTl
z{`ZrncF~R*I
z#oATB#gv4{bQ*r>1l^$1=4Xzt>XDahhRH!h#pO|pYq2aZP}MHG<2Zfua60plQrsLT
z?DXlMCeLHM4Di84U7F4beY*z!6FG2V`>hhT5L?Yfuv=(zslV)44^INhGAn{d+K2=-
zpZx-4;Z9#D2R6eMGT^`I0+p}`MGj42|8`*B9(Yw0v_=Uc3&wBzi0OSmuF?`{fULE<
zGad(vn|Qc2R{VS1SXe+8y@*(kM1fmmAw;G!%383*%0sN?S_}|VSTxwWx*wn#B3vJN
zAbk5QJuroNU=_-D8!bx51i@Uc(LloBv?y7{8v;{s&esy71=??_T``WCgy9iH76KM)
z#-ZZ12>Q+f@GJTMJ_1Y3f49^cDBozP?&P@Mg_>A__WVQI7@dbc2{(Mh**olAf+5b+3+Xu!|iz3LnORz%4
zGzRRQn)z_1^P+A&d%uJXoMAC^5oZU1ZRQ-5VV_ryZ%@BRq9&r<{?#L(aXuaNm8^fB
zF=O5Khj;ZM!u&iEBLkhW$Wp(iyoH&2#Cn3Z>33{n42LGxBM>W44L1T*s=DRi}RLO;i_gZ4P0F*zl
z1^I>b>{m~Uq=Wo&b;9*7dBGMKqY%9sMXC;Xs0Ie?pe)BcVG>Qla=!RC7;xY=Tt)$z
z_f<1uiHF!H0{0=o!)WdH1l+)H$X-OSuvjPs>TD&bM|pIHyNLc424^r0#6`L6VcDXI
zx|W&8U}${EIZ&!udgx7#=Nt8Ad}*AUZ3cU)C2-lYhH-Ll^A6O^omIb*rA3umn_es=
zN4Gnkrx76Ig>({rk#2&wh56D3@~xhex*;kF$t6hn#%cNKxI<78k>-vck+N{Ovk@U{
zPnOjuV&4*egZ*N?*27kPvis!U)lLs1M5{PrSmad(N~Q+MZnh)0!b#YJ#UayN6Z0td
zw$Tbu5p_ditiY4wdh!ya;6)g$nYLL~osss-xV{^@2Aa3PT4+j_QWSow99#Qi{!)msWbPr{8G=K$l*DwaM++hhU^7Ebr2=!`%
zXv>cWsD>yr8~sQzW|fFB*d}BtY5Stk{7f4t%T=@}ka!S-#Dkb;H<0cU;sIIiSD7-I
zD7v5xH6=V~5get=R{e&-(WUt+;`dah3@@|H_c;ngr$^Tv|5XKNFpyXPX5y@08DUOU
zs%7ylP4P?b4bbje+p}=g!d@X0^NE*n8QNvN{U?wxP#cE58FSXM0A3h>@j4rEEied@
z<;$u3XWU4#+`Jy^lT1hq&!@-?89&e&LIq-(#?qJBdk0x9AM*M^f421pM7IL<
z?VaRNC25)$h%=3(ubRBfSw(zIVlM*7tY+WuGoL>7z&BV^<*LW|pr;e;QOU^wQ;X#T
zur7;wGyhXN29(u5RUCf>C?bk@hp{yVjR|2ont#tGa2*R#wI)YdqlcD$!*E<%UTx`v
zk0hIV)*D|+E<(!=Li(jxnS^I3iVB~!7DzJFkv#}y`pP$W0^6{f?Bk!bl*2pom
zQ@3=T5@jZ#9gQa4P*OdVYE{O4zBCJyI=6_Hv(Rr(UXP7`!>1pfvvK%*mJ+k@EGzMAvd`2+)aDjIWr
z*)f`vyLGCF>jbnO76v(h1#0|W_T6$$lptP#+7?z?FGO+ED?!3?iUUx-mvPOmw_{k#{jjTp8njW)4%*N$hC0D4_rM}
zQj4Hum6x2BRkp{FH2VMqQnK&ii|dF2vf%?vWw3hQBUTvjjFO=OGUWVN3<)HOo&j!+
z)3=DX&i_gY#>*I)SM%Q{)5RGb2i#r1KcvjVx|6j&EewimHgHYOVzTWn3%Fz8x-1u&q(%Zu$$ow10+`}QT~1=$V~W7mll33I^^
ztOs5I*%YFbG%*4-t{QmolpXIcv}8|3+(5H`oI;kL)%8q;XoI~%EINQzXc0!smUTDC
zyj=FhUNb
zm7P*t{+%0d!GCZu(NP~<0NI~{1J+SjQd2EBAX|+h^(ByU!hG9%BnbZx_lhGyHnz*L
zh9NE&!DrRuYM$klVUE&&nj8$#E*|2FY%F4!`pF2VEyX`z0DSr=)h
zOx964i?PO4^59Sw8zY!JFG7dIhk~6vTrq|sT5z{-2Sd{N6={?4xA`M>1Iwf?RfS%0
z7pVSP=Te)SE~jU{LCpUkueER-hbr@SkWdeZC*vf87W&J238LFp3EG1!fS+$XQeJ}O
z>X6&ll4V*Tp;H0T#u>FGiF&8|<8Jc*q#LnRs^YURB-
zQu-H-l6ACsN(j`kt}s(%W`XaZ)>*#X=g7jum+5yJRp%U8)vM3;mxO$S+|i^QJMWdn
z%TjXYvC?b4aNfr^wn&|e%K#V>i{-guo`F)B%1v+Uw5kwM7NOGV6l6{1>l6b(nz;07
zT2DUN*sGtTg2=RTmKdY~#=`3t?>v5G@1`dLJcS-y0{ylavk$%~=c^miIW1Z6$KD^G
z0;hMEZNBrzajs0PTFmk1)^n|)fvvJ(&zdRo1UeB!ms@mI>`Lt4`d
z{DF+W+YhZTKJqEamZuIADxJVVcJu-3Ai~6twFQdLB{ylm(a_vbnQ%|3Fe0|;-tsZB
zap+AA99W}feEC-*9H!N=`=#ZozHGr_4xlii>P2Erv0n(GPFRUq!fB9pJ|+{d`Q5R
zqZNq2L+7ivS<2ChpH>_%#6|#5_2V6$v(HtYwDnwc+~*qdw2O`g2OMkibF~xI(Z|>rgEL(5J?9n!(2ZVOFc|R>Rf7T>K06JyKhU+`lr2^lB=$x+i8E*DT*S6b
zdxQVgOwXhfEt5y#=|83fZ!oXk*Gerug#6-EMMrB!w0Cu9%c3)H_g(i*o}-=5o%~<~
z%%5^J^Do`E4(jT_3g+CRE)i32&7K6}cnnm+Pw}o~`#oukOJl-
zPxfi*li2q)z{(2zTz6#M
zFgYQ7YYt6nHF8{5mw3XM82sVTiPU%^aC*abTXUHL`W0Y{I`MZ{qgmwF`RB5a=9zO!
zjb46lDM{8j!b}A!+3bq!0=$%lpZ&O4OgEY)}%ODw^&IAfc7=dU4UlMokxX(`UDIYA;g;dQ-vSfnu{!DdN+{fRUsZe5=0rfWnYe
zjI&Ah4OPukXWdIPK~yf&0w+xes2^vV(!#p$*g?FbvCS%KId6>1we}rT#bxZtum!GK
zK`)xcgL>o^C{8?oz9BFuuMScBNfY^Cl8F?%s!gfx#m}oA>X2cc`v4Ft**b|tV!1tE
z$J8SxI#D{aeRlNus%QJ~wVxUzs1m;b-n_nkTHvI`G#WEaF+SYav>^_f9z6-LG|{Cl
z(J1$&OZMIqo=9l1bCWMx(6tXrLMLRx8GRYGfzwI$B!^j`bffa9lF84g!}}Ur9T)C-
z^!^K4J@#m(6+vV|YtxHaQ`ZTUPIO2f#oU)WBo+2=yp^+z_3VsQl-yW|H_v<@{@nDgvZRq2clQbUg7cd(e|3yY6Bvj_sRj
zAdJ%(tzR%;T}GB8C7(8cp3*sunI)W25);MaRe}S*>vx`f({#CTP-SgVA>?d|8+w7k
zz_IK-s0PK>1@}(nDQ0Zsj4$8jai7r#^$BiQ2e!O^Vk`7+f}2+g>u#PfD5e-KWAB%q
z?_;@;yC&kVqO12^PEoL1`s8cyw5cyr%5%3wP`@WZL^YPt_b9Ow^=%l_o*b#;_hQW_
zKWsk-|INTXSUf;d-=KFV3n%{nFehTMq*!uoaV>ndRj)@-NM4-Q^QnR7dJ**UhHK(480?4;oZ
zGy*UeInicqYHx@Rgo$|-pnxKPK>IrVo{O>mr3>*&!pbctCj)5o8q(UfhNhXviK1=8
zT|~mI2eMqFC$}b)LND}E$Weovr2R7QH{a(qrc*^d_p)%bt}bn?A_y9?>Dr(Cu5}<6
zX8UZ-Gteb>KD-Vl^rJ)Z_@oI1Nnb=^X$KH=heyswtm+`kQD;6R?VP%ig2Mih631`&JByY7}m
zN)N#9VwH-Erp`sP5!0N2xJ&eVO60K{{gL>Q=8z=ndIKKDL!-T>rk=JZy-TMAo3EWRF{b1N0SEJd^9ZE2YaTx=DCu^ST!L=sw%#{DtjkORe0G1IqKAX0A>f1!ehgnENSq=QZf!R0#m54P8Rq39SeSQO73hAH3m2DW?
zx%z}JOo%Lwcu?+sN2NmO7|Yet31TOcq_lU^M!GVXSYqV^ner-pWvz86vi#DQx*DZ9
z#P6DPF-26=+l%M9_F+nxYi-YC7||EB&3RnY1?D}Oxab-J)ll^(LI%NN`7~Z;#^jy`~2ZuHFt*k}Bnb!K{mAGDy$ZsQ2*HLs?P)
zT2*xBtkrt+=g&`ntW=C7ir}7Jif)sJ($4Ry1u1w!vS(HQ2FYz9KRyjBwT55}nw@XL
z+A_`XuKFC*tBV0t3x!#~LHHuv%Dgqwbteb!*SW|z5NU-@Nwvjmsb^V^x
zfY*lIl)kWzICBY^wly9Uo-itUVgfb!3D(&C=%|26KZ2TEMuMC3#8100*!6ox^mBuH
zewbIae*A;&Cy&sA#D|{syRqJfyb%rwR9xCl!&6s_p<+F*d+^KlQqqV@y)c>$ceqC;
zyWWlU{tT8;U_1q&8r>li!F>R}%f~Pw6JTD2hKIp2?KbQY7D_#d8d
zRP|3t5hsroDmVG@>L^@+07G!;*0vSDNy{7vx#ETKZf`OebD(6_8p5w5au96wjX
z=yUMDq8eT13z4z7GHmJGlV%$uJaLU8Vw`Wq%p4U*B<2?R1
z`G5vC6x7=lSP;^Yic+*d>F0sxgGYwP1=`wtiRsrHPcBd3q+oQdmQLtd&x2(9!pR3R2Mf`a^
z;Z5_`!w5j+N{uV~4KvNqjx9R{q1APxy2-vHca{jUzV|%1sC%w5rIQ`ML<$weAXl<6
zMI11aX&(m7S^@Q)rljRaOuT|=;pjj0!tDHZ^amaS-0g0G7XC&hvOXm*GUm~QC3%v^
z7cr5^Bfo-!U=bdcoo02a4y(sZ2S{p(!&Yi#CN9qM;dFhDUqJOAU?HqGcGMEum0c-H
zt6nS$=zo%lfO+=3>-h<`%&;-8y!?(*)ufk8O0Fjc3H@EJ3d=uzWc~WL4))2D0@@-}
z9#i*V3>R2z_Nwg(lE%-1A)+IX>z;;+thfJ%h->V%XpePCT_;#YlT8vyA;QUS8DPkkZChskptSJqqlk{}U
z@w+^eekG5Jt1U`LA67hcyrJVd@KR~@{?~rHk?zK^^{aY5>6m4s
zUVc=dA!A+@bDDjE(iFavBoph?KK-M_UFEO>UkCD6?%=ni*L_7#tDFGuPkskXq~I=U
z;c$%&`aO@KB%+;M91?_&)RcUWG~igLQUop~Uv2Aa@by?Grc1UDmw|!#2+Me$7S(~U
z;EOFNo*uF?tWU_ijO?undp#dq-W5!Idq;44zxD{3N2^au)(0zL6GY%qUOxl+3MPjE6tdVdsO0(RZCZ2$eTE=HqaA5e}
zxjM0^OI4xYgCG*I`HNfVzq<{1sxyz2<}bo@s{IBC*VC_h4U1_<&Jr`0aNC~oMNc>|
zbi^euBHwugcUcUP*$$PJC%w58Cd?+)WDrGiDG_kBuk!IEB!kFOLU6by+