gae_discrete_monte-carlo.ipynb

{
 "cells": [
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Generalized Advantage Estimate (GAE) Monte-Carlo\n",
    "\n",
    "This is the GAE algorithm with a discrete action space (only works on environments with discrete actions) and Monte-Carlo parameter updates. Generalized Advantage Estimate (GAE) is an actor-critic method which proposes a better adventage estimation function, the generalized advantage estimation function. Although this algorithm performs Monte Carlo updates, it still uses bootstrapping in the GAE function, so it is still an actor-critic method. We will implement GAE by adapting A2C, which means we will synchronously run multiple environments in parallel. The advantage function in A2C will be swapped for the GAE advantage function. \n",
    "\n",
    "The advantage function, $A^{\\pi}$, following policy $\\pi$ is defined by:\n",
    "$$A^{\\pi} = Q^{\\pi}(s_t, a_t) - V^{\\pi}(s_t),$$\n",
    "where $Q^{\\pi}(s_t, a_t)$ is the action-value function following policy $\\pi$ and $V^{\\pi}(s_t)$ is the state-value function following policy $\\pi$. The advantage function essentially measures the advantage gained (or lost) by taking the action $a_t$ in state $s_t$ over the average return of all immediate actions, and then following policy $\\pi$ after. We don't know the absolute value of the advantage, $A^{\\pi}$, hence we need to find an advantage estimate, $\\hat{A}$, as a substitute. GAE is a good alternative to the advantage function in A2C, because it has lower bias at the cost of a minor increase in variance. \n",
    "\n",
    "GAE is defined by:\n",
    "$$\\hat{A}_{t}^{GAE} = \\sum_{l=0}^{T}(\\gamma \\lambda)^{l}\\delta_{t+l},$$ \n",
    "where $\\delta_{t+l}$ is the TD error, $\\gamma$ is the discount rate, $\\lambda$ is the decay rate and $T$ is the final timestep. The bias-variance trade-off can be finetuned by altering the discount rate $\\gamma$ and/or the decay rate $\\lambda$. Larger discount/decay rates lead to higher variance and lower bias and lower discount/decay rates lead to lower variance and higher bias. \n",
    "\n",
    "We will be using autograd to perform gradient descent on the policy and value functions. We only need to construct the objective functions (loss functions) and their derivatives will be calculated automatically. We will be using the Adam optimizer for both policy and value gradient updates. The loss functions will be combined into a single loss function. Autograd traces the paths of all the parameters for the gradient calculation, so combining the losses won't impact the gradient calculations. \n",
    "\n",
    "The policy objective function is given by:\n",
    "$$L_{t}^{PF} = -\\hat{E_t}[\\log(\\pi_\\theta(a_t | s_t))\\hat{A}_{t}].$$\n",
    "\n",
    "The value objective function is given by:\n",
    "$$L_{t}^{VF} = \\hat{E_t}[\\hat{A}_{t}^2].$$\n",
    "\n",
    "The final objective function is given by:\n",
    "$$L_{t} = \\hat{E_t}[L_{t}^{PF} + c_{1}L_{t}^{VF} - c_{2}S_{\\pi_{\\theta}}(s_t)],$$\n",
    "where $S_{\\pi_{\\theta}}(s_t)$ is the entropy of the policy function, and $c_{1}$, $c_{2}$ are the objective coefficients. Increasing $c_{1}$ will increase the update size on the value function (the same effect as increasing the learning rate) and increasing $c_{2}$ will make the policy function less deterministic. Decreasing the coefficients will have the opposite effect. \n",
    "\n",
    "\n",
    "\n",
    "For more information on the Generalized Advantage Estimate algorithm, see the original paper [1] [High-Dimensional Continuous Control Using Generalized Advantage Estimation](https://arxiv.org/abs/1506.02438), the blog post [2] [Generalized Advantage Estimate: Maths and Code](https://towardsdatascience.com/generalized-advantage-estimate-maths-and-code-b5d5bd3ce737), and the notebook [3] [RL Adventure2 GAE](https://github.com/higgsfield/RL-Adventure-2/blob/master/2.gae.ipynb).\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "import numpy as np\n",
    "import gymnasium as gym\n",
    "from IPython.display import Video\n",
    "import jax\n",
    "import flax.linen as nn\n",
    "from jax import value_and_grad\n",
    "import jax.numpy as jnp\n",
    "import optax\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "from vec_env.subproc_vec_env import SubprocVecEnv\n",
    "# Print numpy array floats with 2 decimal places\n",
    "np.set_printoptions(precision=2, suppress=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Type of environment, options include:\n",
    "# CartPole-v1, MountainCar-v0, LunarLander-v2\n",
    "ENV_TYPE = \"CartPole-v1\"\n",
    "# Choose number of parallel environments, \n",
    "# don't exceed the number of computer threads\n",
    "num_envs = 10\n",
    "\n",
    "# Save videos from training\n",
    "SAVE_VIDEOS = True\n",
    "VIDEOS_ROOT_DIR = './videos/gae_discrete_monte-carlo/'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "def make_env(env_id):\n",
    "    \"\"\"\n",
    "    Returns a single environment object, \n",
    "    which will be used for parallel execution.\n",
    "    \"\"\"\n",
    "    def _init():\n",
    "        env = gym.make(env_id)\n",
    "        return env\n",
    "    return _init"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Create multiple envs for parallel execution\n",
    "envs = SubprocVecEnv([make_env(ENV_TYPE) for _ in range(num_envs)])\n",
    "\n",
    "num_actions = envs.action_space.n\n",
    "obs_dims = envs.observation_space.shape\n",
    "obs_upper_bounds = envs.observation_space.high\n",
    "obs_lower_bounds = envs.observation_space.low"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Create The Policy "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "\u001b[3m                               _policy_fn Summary                               \u001b[0m\n",
      "┏━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━┓\n",
      "┃\u001b[1m \u001b[0m\u001b[1mpath   \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1mmodule    \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1minputs         \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1moutputs        \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1mparams           \u001b[0m\u001b[1m \u001b[0m┃\n",
      "┡━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━┩\n",
      "│         │ _policy_fn │ \u001b[2mfloat32\u001b[0m[10,4]   │ \u001b[2mfloat32\u001b[0m[10,2]   │                   │\n",
      "├─────────┼────────────┼─────────────────┼─────────────────┼───────────────────┤\n",
      "│ Dense_0 │ Dense      │ \u001b[2mfloat32\u001b[0m[10,4]   │ \u001b[2mfloat32\u001b[0m[10,256] │ bias:             │\n",
      "│         │            │                 │                 │ \u001b[2mfloat32\u001b[0m[256]      │\n",
      "│         │            │                 │                 │ kernel:           │\n",
      "│         │            │                 │                 │ \u001b[2mfloat32\u001b[0m[4,256]    │\n",
      "│         │            │                 │                 │                   │\n",
      "│         │            │                 │                 │ \u001b[1m1,280 \u001b[0m\u001b[1;2m(5.1 KB)\u001b[0m    │\n",
      "├─────────┼────────────┼─────────────────┼─────────────────┼───────────────────┤\n",
      "│ Dense_1 │ Dense      │ \u001b[2mfloat32\u001b[0m[10,256] │ \u001b[2mfloat32\u001b[0m[10,2]   │ bias: \u001b[2mfloat32\u001b[0m[2]  │\n",
      "│         │            │                 │                 │ kernel:           │\n",
      "│         │            │                 │                 │ \u001b[2mfloat32\u001b[0m[256,2]    │\n",
      "│         │            │                 │                 │                   │\n",
      "│         │            │                 │                 │ \u001b[1m514 \u001b[0m\u001b[1;2m(2.1 KB)\u001b[0m      │\n",
      "├─────────┼────────────┼─────────────────┼─────────────────┼───────────────────┤\n",
      "│\u001b[1m \u001b[0m\u001b[1m       \u001b[0m\u001b[1m \u001b[0m│\u001b[1m \u001b[0m\u001b[1m          \u001b[0m\u001b[1m \u001b[0m│\u001b[1m \u001b[0m\u001b[1m               \u001b[0m\u001b[1m \u001b[0m│\u001b[1m \u001b[0m\u001b[1m          Total\u001b[0m\u001b[1m \u001b[0m│\u001b[1m \u001b[0m\u001b[1m1,794 \u001b[0m\u001b[1;2m(7.2 KB)\u001b[0m\u001b[1m   \u001b[0m\u001b[1m \u001b[0m│\n",
      "└─────────┴────────────┴─────────────────┴─────────────────┴───────────────────┘\n",
      "\u001b[1m                                                                                \u001b[0m\n",
      "\u001b[1m                        Total Parameters: 1,794 \u001b[0m\u001b[1;2m(7.2 KB)\u001b[0m\u001b[1m                        \u001b[0m\n",
      "\n",
      "\n"
     ]
    }
   ],
   "source": [
    "class _policy_fn(nn.Module):\n",
    "    \"\"\"\n",
    "    Policy function is an MLP used to choose the \n",
    "    best possible discrete action given the observations from the state.\n",
    "    Each discrete action is given a probability from 0-1.\n",
    "    The probability distribution can then be used \n",
    "    to stocastically choose the next action.  \n",
    "    \"\"\"\n",
    "    @nn.compact\n",
    "    def __call__(self, x):\n",
    "        x = nn.Dense(256)(x)\n",
    "        x = jax.nn.relu(x)\n",
    "        x = nn.Dense(num_actions)(x)\n",
    "        return jax.nn.softmax(x)\n",
    "\n",
    "# Create the model object\n",
    "policy_fn = _policy_fn()\n",
    "# Display the model details\n",
    "dummy_x = envs.reset()\n",
    "dummy_x = jnp.array(dummy_x)\n",
    "rng_key = jax.random.PRNGKey(64)\n",
    "print(policy_fn.tabulate(rng_key, dummy_x))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "\u001b[3m                               _value_fn Summary                                \u001b[0m\n",
      "┏━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━┓\n",
      "┃\u001b[1m \u001b[0m\u001b[1mpath   \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1mmodule   \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1minputs         \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1moutputs        \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1mparams            \u001b[0m\u001b[1m \u001b[0m┃\n",
      "┡━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━┩\n",
      "│         │ _value_fn │ \u001b[2mfloat32\u001b[0m[10,4]   │ \u001b[2mfloat32\u001b[0m[10,1]   │                    │\n",
      "├─────────┼───────────┼─────────────────┼─────────────────┼────────────────────┤\n",
      "│ Dense_0 │ Dense     │ \u001b[2mfloat32\u001b[0m[10,4]   │ \u001b[2mfloat32\u001b[0m[10,256] │ bias: \u001b[2mfloat32\u001b[0m[256] │\n",
      "│         │           │                 │                 │ kernel:            │\n",
      "│         │           │                 │                 │ \u001b[2mfloat32\u001b[0m[4,256]     │\n",
      "│         │           │                 │                 │                    │\n",
      "│         │           │                 │                 │ \u001b[1m1,280 \u001b[0m\u001b[1;2m(5.1 KB)\u001b[0m     │\n",
      "├─────────┼───────────┼─────────────────┼─────────────────┼────────────────────┤\n",
      "│ Dense_1 │ Dense     │ \u001b[2mfloat32\u001b[0m[10,256] │ \u001b[2mfloat32\u001b[0m[10,1]   │ bias: \u001b[2mfloat32\u001b[0m[1]   │\n",
      "│         │           │                 │                 │ kernel:            │\n",
      "│         │           │                 │                 │ \u001b[2mfloat32\u001b[0m[256,1]     │\n",
      "│         │           │                 │                 │                    │\n",
      "│         │           │                 │                 │ \u001b[1m257 \u001b[0m\u001b[1;2m(1.0 KB)\u001b[0m       │\n",
      "├─────────┼───────────┼─────────────────┼─────────────────┼────────────────────┤\n",
      "│\u001b[1m \u001b[0m\u001b[1m       \u001b[0m\u001b[1m \u001b[0m│\u001b[1m \u001b[0m\u001b[1m         \u001b[0m\u001b[1m \u001b[0m│\u001b[1m \u001b[0m\u001b[1m               \u001b[0m\u001b[1m \u001b[0m│\u001b[1m \u001b[0m\u001b[1m          Total\u001b[0m\u001b[1m \u001b[0m│\u001b[1m \u001b[0m\u001b[1m1,537 \u001b[0m\u001b[1;2m(6.1 KB)\u001b[0m\u001b[1m    \u001b[0m\u001b[1m \u001b[0m│\n",
      "└─────────┴───────────┴─────────────────┴─────────────────┴────────────────────┘\n",
      "\u001b[1m                                                                                \u001b[0m\n",
      "\u001b[1m                        Total Parameters: 1,537 \u001b[0m\u001b[1;2m(6.1 KB)\u001b[0m\u001b[1m                        \u001b[0m\n",
      "\n",
      "\n"
     ]
    }
   ],
   "source": [
    "class _value_fn(nn.Module):\n",
    "    \"\"\"\n",
    "    Value function is an MLP used to predict the expected return (value) \n",
    "    given the observations from the state.\n",
    "    The value can then be used for bootstrapping, and/or \n",
    "    calculating the advantage from action chosen in the state. \n",
    "    \"\"\"\n",
    "    @nn.compact\n",
    "    def __call__(self, x):\n",
    "        x = nn.Dense(256)(x)\n",
    "        x = jax.nn.relu(x)\n",
    "        x = nn.Dense(1)(x)\n",
    "        return x\n",
    "\n",
    "# Create the model object\n",
    "value_fn = _value_fn()\n",
    "# Display the model details\n",
    "print(value_fn.tabulate(rng_key, dummy_x))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "def test_model(policy_variables, env, rng_key):\n",
    "    \"\"\"\n",
    "    Runs a single episode following the policy function, \n",
    "    then returns the episodes rewards and steps.\n",
    "    Args:\n",
    "        policy_variables: Policy functions model parameters.\n",
    "        env: Environment used to interact with.\n",
    "        rng_key: Random number generator key.\n",
    "    Returns:\n",
    "        steps: Number of environment steps.\n",
    "        episode_rewards: List of rewards from the episode.\n",
    "    \"\"\"\n",
    "\n",
    "    terminated, truncated = False, False\n",
    "    episode_rewards = []\n",
    "    steps = 0\n",
    "\n",
    "    state, _ = env.reset()\n",
    "    while not (terminated or truncated):\n",
    "        steps += 1\n",
    "        # Select the action following the parametrized policy\n",
    "        action_dist = policy_fn.apply(policy_variables, state)\n",
    "        rng_key, subkey = jax.random.split(rng_key)\n",
    "        action = jax.random.choice(subkey, jnp.arange(num_actions), p=action_dist)\n",
    "        action = np.array(action)\n",
    "        next_state, reward, terminated, truncated, info = env.step(action)\n",
    "\n",
    "        episode_rewards.append(reward)\n",
    "\n",
    "        state = next_state\n",
    "    episode_rewards = np.array(episode_rewards)\n",
    "    return steps, episode_rewards\n",
    "\n",
    "    "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "def get_gae(\n",
    "    gamma, lamda, values, next_value, rewards, done\n",
    "):\n",
    "    \"\"\"\n",
    "    Get generalized advantage estimate's from rewards for each environment.\n",
    "    Args:\n",
    "        gamma: Trajectory discount (scalar).\n",
    "        lamda: Trace decay rate.\n",
    "        values: Value functions expected returns at each step.\n",
    "        next_value: Value functions bootstrapped return for next state.\n",
    "        rewards: Agent rewards of taking actions in the environment.\n",
    "        done: Flag list for end of episode, \n",
    "            False indicates terminated, and True not terminated.\n",
    "    Returns:\n",
    "        Array of gae's for each environment at each step.\n",
    "    \"\"\"\n",
    "    batch_size = len(done)\n",
    "    values = jnp.array(values + [next_value])\n",
    "    rewards = jnp.array(rewards)\n",
    "    done = jnp.array(done)\n",
    "    advantage = 0\n",
    "    advantages = []\n",
    "\n",
    "    for t in reversed(range(batch_size)):\n",
    "        delta = rewards[t] + (gamma * values[t + 1] * done[t]) - values[t]\n",
    "        advantage = delta + (gamma * lamda * advantage * done[t])\n",
    "        advantages.append(advantage)\n",
    "\n",
    "    return jnp.array(list(reversed(advantages)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "def loss_fn(value_variables, policy_variables, rng_key, gamma, trace_decay_rate, max_steps, env_returns, env_steps):\n",
    "    \"\"\"\n",
    "    Runs a single episode in a Monte-Carlo way, \n",
    "    then calculates the actor and critic losses, \n",
    "    which are then combined and returned.\n",
    "    The gradients on this loss function are then used \n",
    "    to update the policy and actor functions.\n",
    "    Args:\n",
    "        value_variables: Value functions model parameters.\n",
    "        policy_variables: Policy functions model parameters.\n",
    "        rng_key: Random number generator key.\n",
    "        gamma: Discount rate for discounting returns.\n",
    "        trace_decay_rate: Trace decay rate for the lambda-returns.\n",
    "        max_steps: Limit on the number of steps in an episode.\n",
    "        env_returns: List containing the returns from all environments each episode.\n",
    "        env_steps: List containing the steps from all environments each episode.\n",
    "    Returns:\n",
    "        Combined loss of the value and actor functions.\n",
    "    \"\"\"\n",
    "    terminated, truncated, term_or_trunc = jnp.repeat(False, num_envs), jnp.repeat(False, num_envs), jnp.repeat(False, num_envs)\n",
    "    episode_log_probs = []\n",
    "    episode_values = []\n",
    "    episode_rewards = []\n",
    "    done_flag = []\n",
    "    steps = 0\n",
    "\n",
    "    states = envs.reset()\n",
    "    states = jnp.array(states)\n",
    "    # Generate an episode following the policy\n",
    "    while not term_or_trunc.all() and steps < max_steps:\n",
    "        steps += 1\n",
    "        # Select the action following the parametrized policy\n",
    "        action_dists = policy_fn.apply(policy_variables, states)\n",
    "        # Stocastically choose the next action based on the probability distribution\n",
    "        actions = []\n",
    "        for dist in action_dists:\n",
    "            rng_key, subkey = jax.random.split(rng_key)\n",
    "            action = jax.random.choice(subkey, jnp.arange(num_actions), p=dist)\n",
    "            actions.append(action)\n",
    "        actions = np.array(actions)\n",
    "        \n",
    "        next_states, rewards, terminated, truncated, info = envs.step(actions)\n",
    "        # Check if env was terminated or truncated in this episode or one of the previous episodes \n",
    "        # We only want one episode per environment\n",
    "        term_or_trunc = jnp.array([term or trunc or t_or_t for term, trunc, t_or_t in zip(terminated, truncated, term_or_trunc)])\n",
    "        done_flag.append(jnp.invert(term_or_trunc))\n",
    "        # If the episode previously ended, then set the reward to 0, \n",
    "        # we're ignoring rewards after the environment terminated\n",
    "        rewards = jnp.array([reward if not t_or_t else 0 for reward, t_or_t in zip(rewards, term_or_trunc)])\n",
    "        episode_rewards.append(rewards)\n",
    "        # Log probabilities are needed for the actor loss\n",
    "        log_probs = jnp.array([jnp.log(dist[action_index]) for dist, action_index in zip(action_dists, actions)])\n",
    "        episode_log_probs.append(log_probs)\n",
    "        # Entropy is used in the loss function to prevent the model from becoming too deterministic\n",
    "        entropy = jnp.array([-jnp.sum(dist * jnp.log(dist)) for dist in action_dists])\n",
    "        # Estimate the values given the states, then set all terminated states value to 0\n",
    "        values = value_fn.apply(value_variables, states)\n",
    "        values = jnp.squeeze(values)\n",
    "        # We're ignoring values after the environment terminated\n",
    "        values = [value if not t_or_t else 0.0 for value, t_or_t in zip(jnp.array(values), term_or_trunc)]\n",
    "        values = jnp.array(values, dtype=jnp.float32)\n",
    "        episode_values.append(values)\n",
    "\n",
    "        states = next_states\n",
    "\n",
    "    env_returns.append(list(np.array(episode_rewards).sum(axis=0)))\n",
    "    env_steps.append(list(np.array(done_flag).sum(axis=0)))\n",
    "\n",
    "    # Calculate the advantages\n",
    "    next_values = value_fn.apply(value_variables, next_states)\n",
    "    next_values = jnp.squeeze(next_values)\n",
    "    advantages = get_gae(gamma, trace_decay_rate, episode_values, next_values, episode_rewards, done_flag)\n",
    "    # Calculate actor and critic losses, then add entropy \n",
    "    episode_log_probs = jnp.array(episode_log_probs)\n",
    "    actor_loss = -(advantages * episode_log_probs).mean(axis=0)\n",
    "    critic_loss = (advantages**2).mean(axis=0)\n",
    "    loss = (actor_loss + 0.5 * critic_loss - 0.001 * entropy).mean()\n",
    "\n",
    "    return loss"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Learn The Optimal Action-Value Function"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Episode: 0\n",
      "\tReturns: [12. 15. 24. 11. 20. 10. 27. 37. 14.  8.]\n",
      "\tAverage Return: 17.80\n",
      "\tLoss: 19.19\n",
      "Episode: 1\n",
      "\tReturns: [ 16.  35.  41.  20. 101.  11.  73.  34.  38.  14.]\n",
      "\tAverage Return: 38.30\n",
      "\tLoss: 30.20\n",
      "Episode: 2\n",
      "\tReturns: [ 53.  66.  47.  22.  41.  29.  48. 109.  40.  20.]\n",
      "\tAverage Return: 47.50\n",
      "\tLoss: 36.52\n",
      "Episode: 3\n",
      "\tReturns: [ 37.  48. 116. 158.  75.  21.  46.  78.  35.  21.]\n",
      "\tAverage Return: 63.50\n",
      "\tLoss: 39.41\n",
      "Episode: 4\n",
      "\tReturns: [ 35. 102.  54.  71. 151.  21.  52.  31.  95.  32.]\n",
      "\tAverage Return: 64.40\n",
      "\tLoss: 41.37\n",
      "Episode: 5\n",
      "\tReturns: [125.  41.  85.  92. 117.  34.  82. 122.  42.  40.]\n",
      "\tAverage Return: 78.00\n",
      "\tLoss: 63.86\n",
      "Episode: 6\n",
      "\tReturns: [109.  76.  57.  73. 109.  23.  86.  95.  40.  54.]\n",
      "\tAverage Return: 72.20\n",
      "\tLoss: 64.78\n",
      "Episode: 7\n",
      "\tReturns: [128.  91.  88.  51.  61.  42.  85.  92. 107.  33.]\n",
      "\tAverage Return: 77.80\n",
      "\tLoss: 61.09\n",
      "Episode: 8\n",
      "\tReturns: [ 74. 138. 121.  58.  50.  21. 128.  88. 121. 127.]\n",
      "\tAverage Return: 92.60\n",
      "\tLoss: 71.76\n",
      "Episode: 9\n",
      "\tReturns: [ 63. 200. 169.  62. 106.  38. 163.  94. 108.  87.]\n",
      "\tAverage Return: 109.00\n",
      "\tLoss: 61.05\n",
      "Episode: 10\n",
      "\tReturns: [126. 200. 171.  50. 106.  76. 127. 141.  94. 200.]\n",
      "\tAverage Return: 129.10\n",
      "\tLoss: 75.18\n",
      "Episode: 11\n",
      "\tReturns: [ 82. 200.  84.  55. 200.  70. 108. 124. 200. 169.]\n",
      "\tAverage Return: 129.20\n",
      "\tLoss: 75.10\n",
      "Episode: 12\n",
      "\tReturns: [131. 156.  75.  63. 112.  73. 200.  91. 109. 200.]\n",
      "\tAverage Return: 121.00\n",
      "\tLoss: 70.22\n",
      "Episode: 13\n",
      "\tReturns: [ 86. 195.  85.  72. 200.  92. 200.  90. 199. 164.]\n",
      "\tAverage Return: 138.30\n",
      "\tLoss: 78.75\n",
      "Episode: 14\n",
      "\tReturns: [138. 200. 134. 200. 128.  78. 200.  76. 123. 155.]\n",
      "\tAverage Return: 143.20\n",
      "\tLoss: 81.83\n",
      "Episode: 15\n",
      "\tReturns: [200. 200. 147. 179. 125. 136. 170.  82. 168. 200.]\n",
      "\tAverage Return: 160.70\n",
      "\tLoss: 93.14\n"
     ]
    }
   ],
   "source": [
    "# Number of episodes to train on\n",
    "episodes = 16\n",
    "# Sliding window for episode evaluation (needs to be less than episodes)\n",
    "episode_window = 3\n",
    "# Episode will terminate before max_steps\n",
    "max_steps = 200\n",
    "# Discount ratio\n",
    "gamma = 0.99\n",
    "trace_decay_rate = 0.95\n",
    "# Step size\n",
    "policy_alpha = 0.01\n",
    "value_alpha = 0.01\n",
    "\n",
    "policy_variables = policy_fn.init(rng_key, dummy_x)\n",
    "policy_optimizer = optax.adam(learning_rate=policy_alpha)\n",
    "policy_opt_state = policy_optimizer.init(policy_variables)\n",
    "\n",
    "value_variables = value_fn.init(rng_key, dummy_x)\n",
    "value_optimizer = optax.adam(learning_rate=value_alpha)\n",
    "value_opt_state = value_optimizer.init(value_variables)\n",
    "\n",
    "# Create multiple envs for parallel execution\n",
    "envs = SubprocVecEnv([make_env(ENV_TYPE) for _ in range(num_envs)])\n",
    "if SAVE_VIDEOS:\n",
    "    # Create a single environment for testing\n",
    "    os.makedirs(VIDEOS_ROOT_DIR, exist_ok=True)\n",
    "    test_returns = []\n",
    "    test_steps = []\n",
    "    env = gym.make(ENV_TYPE, render_mode=\"rgb_array\")\n",
    "    env = gym.wrappers.RecordVideo(\n",
    "        env=env, \n",
    "        video_folder=VIDEOS_ROOT_DIR,\n",
    "        episode_trigger=lambda episode: episode % 1 == 0,\n",
    "        video_length=0,\n",
    "        name_prefix=f\"{ENV_TYPE}\"\n",
    "        ) \n",
    "env_returns = []\n",
    "env_steps = []\n",
    "losses = []\n",
    "\n",
    "for episode in range(episodes):\n",
    "    print(f\"Episode: {episode}\")\n",
    "    # Perform 6 tests and record them at even intervals\n",
    "    if SAVE_VIDEOS and episode % (episodes // 5) == 0:\n",
    "        test_step, test_rewards = test_model(policy_variables, env, rng_key)\n",
    "        test_returns.append(np.sum(test_rewards))\n",
    "        test_steps.append(test_step)\n",
    "        test_episode = episode // (episodes // 5)\n",
    "        print(f\"\\tTest Episode {test_episode} Return: {test_returns[test_episode] :.2f} Steps: {test_steps[test_episode] :.2f}\")\n",
    "\n",
    "    # Run a forward pass on the parallel envs (will last an entire episode) and \n",
    "    # calculate the gradients for value and policy functions \n",
    "    loss, (value_grads, policy_grads) = value_and_grad(loss_fn, argnums=(0, 1))(\n",
    "        value_variables, \n",
    "        policy_variables, \n",
    "        rng_key, \n",
    "        gamma, \n",
    "        trace_decay_rate,\n",
    "        max_steps,\n",
    "        env_returns,\n",
    "        env_steps\n",
    "    )\n",
    "    # Update value function weights\n",
    "    value_updates, value_opt_state = value_optimizer.update(value_grads, value_opt_state, value_variables)\n",
    "    value_variables = optax.apply_updates(value_variables, value_updates)        \n",
    "    # Update policy function weights\n",
    "    policy_updates, policy_opt_state = policy_optimizer.update(policy_grads, policy_opt_state, policy_variables)\n",
    "    policy_variables = optax.apply_updates(policy_variables, policy_updates)\n",
    "\n",
    "\n",
    "    losses.append(loss)\n",
    "    print(f'\\tReturns: {np.array(env_returns[-1])}')\n",
    "    print(f\"\\tAverage Return: {np.array(env_returns[-1]).mean() :.2f}\")\n",
    "    print(f\"\\tLoss: {loss :.2f}\")"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Plot The Results\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 600x500 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig, axs = plt.subplots(ncols=1, figsize=(6, 5))\n",
    "axs.set_title(\"Loss\")\n",
    "loss_moving_average = (\n",
    "    np.convolve(\n",
    "        losses, np.ones(episode_window), mode=\"valid\"\n",
    "    )\n",
    "    / episode_window\n",
    ")\n",
    "axs.plot(range(len(loss_moving_average)), loss_moving_average)\n",
    "plt.tight_layout()\n",
    "plt.show()"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "For cartpole rewards are assigned for the time spent in a non-terminal state. The episode length and rewards will be identical in this case."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAABKUAAAHqCAYAAADVi/1VAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjguNCwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy8fJSN1AAAACXBIWXMAAA9hAAAPYQGoP6dpAACCwElEQVR4nOzdd3hUZd7G8XsmvYf0QhIChA4htEi14arYEFARpCv23nB3rWvv4qqorzQVK2Jbxb6C1EDoPZQkJCQhCel95rx/oLNEqZrkTJLv57py7ebM5OSeqPDLfc48j8UwDEMAAAAAAABAE7KaHQAAAAAAAACtD6UUAAAAAAAAmhylFAAAAAAAAJocpRQAAAAAAACaHKUUAAAAAAAAmhylFAAAAAAAAJocpRQAAAAAAACaHKUUAAAAAAAAmhylFAAAAAAAAJocpRSAU/bQQw/JYrE06ffct2+fLBaL5s6d26Tft7mxWCx66KGHzI4BAACOo6XMUs1l7jjjjDPUo0cPs2MAOApKKaCFmzt3riwWyzE/Vq5caXZEAAAAp8Us1TxkZ2froYce0vr1682OAuAUuJodAEDTeOSRRxQfH/+H4x07djzlc/3zn//UjBkzGiIWAABAs8As5dyys7P18MMPq127durdu7fZcQCcJEopoJU4//zz1a9fvwY5l6urq1xdW94fH3V1dbLb7XJ3dzc7yjGVl5fLx8fH7BgAALQ6zFIA0PB4+x4ASf9bZ+DZZ5/VCy+8oLi4OHl5een000/X5s2b6z33aOsgfPfddxoyZIgCAwPl6+urzp076+9//3u95+Tl5WnatGkKDw+Xp6enEhMTNW/evD9kKSoq0uTJkxUQEKDAwEBNmjRJRUVFR829fft2jRkzRkFBQfL09FS/fv30+eefn9LrffHFF9WhQwd5eHho69atJ3XeoqIiubi4aObMmY5j+fn5slqtCg4OlmEYjuPXX3+9IiIiHJ8vXbpUl112mWJjY+Xh4aGYmBjdfvvtqqysrJdx8uTJ8vX11e7duzVixAj5+flp/PjxkqTq6mrdfvvtCg0NlZ+fny6++GLt37//D6+ztLRUt912m9q1aycPDw+FhYXpnHPOUWpq6gl/RgAA4OS1tlnqWLKysjR16lSFh4fLw8ND3bt31+zZs+s957///a8sFos+/PBDPfbYY2rbtq08PT119tlnKy0t7Q/nfOWVV9S+fXt5eXlpwIABWrp0qc444wydccYZjvP1799fkjRlyhTHWyt/v37W1q1bdeaZZ8rb21vR0dF6+umn//C9Xn75ZXXv3l3e3t5q06aN+vXrpwULFvzpnweA46OeB1qJ4uJi5efn1ztmsVgUHBxc79j8+fNVWlqqG2+8UVVVVXrppZd01llnadOmTQoPDz/qubds2aILL7xQvXr10iOPPCIPDw+lpaVp2bJljudUVlbqjDPOUFpamm666SbFx8fro48+0uTJk1VUVKRbb71VkmQYhi655BL98ssvuu6669S1a1ctWrRIkyZNOur3HTx4sKKjozVjxgz5+Pjoww8/1MiRI7Vw4UJdeumlJ/y5zJkzR1VVVZo+fbo8PDwUFBR0UucNDAxUjx49tGTJEt1yyy2SpF9++UUWi0WFhYXaunWrunfvLulwCTV06FDH9/zoo49UUVGh66+/XsHBwVq9erVefvll7d+/Xx999FG9fHV1dTr33HM1ZMgQPfvss/L29pYkXX311XrnnXc0btw4DRo0SD/++KMuuOCCP7y+6667Th9//LFuuukmdevWTQUFBfrll1+0bds29enT54Q/HwAAcBiz1Inl5ubqtNNOk8Vi0U033aTQ0FB9/fXXmjZtmkpKSnTbbbfVe/6TTz4pq9Wqu+66S8XFxXr66ac1fvx4rVq1yvGc1157TTfddJOGDh2q22+/Xfv27dPIkSPVpk0btW3bVpLUtWtXPfLII3rggQc0ffp0x9w1aNAgx3kOHTqk8847T6NGjdLll1+ujz/+WPfee6969uyp888/X5L05ptv6pZbbtGYMWN06623qqqqShs3btSqVas0bty4U/pZADhJBoAWbc6cOYako354eHg4nrd3715DkuHl5WXs37/fcXzVqlWGJOP22293HHvwwQeNI//4eOGFFwxJxsGDB4+Z48UXXzQkGe+8847jWE1NjTFw4EDD19fXKCkpMQzDMD799FNDkvH00087nldXV2cMHTrUkGTMmTPHcfzss882evbsaVRVVTmO2e12Y9CgQUZCQsJxfy6/vV5/f38jLy+v3mMne94bb7zRCA8Pd3x+xx13GMOGDTPCwsKM1157zTAMwygoKDAsFovx0ksvOZ5XUVHxhzxPPPGEYbFYjPT0dMexSZMmGZKMGTNm1Hvu+vXrDUnGDTfcUO/4uHHjDEnGgw8+6DgWEBBg3Hjjjcf9WQAAgGNjljq2388d06ZNMyIjI438/Px6zxs7dqwREBDgmIF++uknQ5LRtWtXo7q62vG8l156yZBkbNq0yTAMw6iurjaCg4ON/v37G7W1tY7nzZ0715BknH766Y5jKSkpf3h9vzn99NMNScb8+fMdx6qrq42IiAhj9OjRjmOXXHKJ0b179xO+bgANh7fvAa3EK6+8ou+++67ex9dff/2H540cOVLR0dGOzwcMGKDk5GR99dVXxzx3YGCgJOmzzz6T3W4/6nO++uorRURE6Morr3Qcc3Nz0y233KKysjL9/PPPjue5urrq+uuvdzzPxcVFN998c73zFRYW6scff9Tll1+u0tJS5efnKz8/XwUFBTr33HO1a9cuZWVlnfDnMnr0aIWGhv6p8w4dOlS5ubnasWOHpMN3RA0bNkxDhw7V0qVLJR2+e8owjHp3Snl5eTn+f3l5ufLz8zVo0CAZhqF169b9IeORP4vffkaSHHdo/eb3Vx+lw/9sVq1apezs7BP+LAAAwLExSx2fYRhauHChLrroIhmG4Thffn6+zj33XBUXF/9h+YApU6bUW8vzt3lpz549kqQ1a9aooKBA11xzTb01uMaPH682bdqcdDZJ8vX11VVXXeX43N3dXQMGDHB8L+nwP4f9+/crJSXllM4N4M+jlAJaiQEDBmj48OH1Ps4888w/PC8hIeEPxzp16qR9+/Yd89xXXHGFBg8erKuvvlrh4eEaO3asPvzww3pDVXp6uhISEmS11v9jp2vXro7Hf/vfyMhI+fr61nte586d632elpYmwzB0//33KzQ0tN7Hgw8+KOnwugsn8vtddE7lvL8NTkuXLlV5ebnWrVunoUOHatiwYY5SaunSpfL391diYqLje2RkZGjy5MkKCgqSr6+vQkNDdfrpp0s6/NaAI7m6ujpuTT/yZ2m1WtWhQ4fj/owk6emnn9bmzZsVExOjAQMG6KGHHqo3fAEAgJPDLHV8Bw8eVFFRkd54440/nG/KlClHPV9sbGy9z38rmg4dOlTvNf1+h0NXV1e1a9fupLNJUtu2bf+wjlebNm0c30uS7r33Xvn6+mrAgAFKSEjQjTfeWO8tlAAaHmtKAfjLvLy8tGTJEv3000/6z3/+o8WLF+uDDz7QWWedpW+//VYuLi4N/j1/G9LuuusunXvuuUd9zsls0XzkXUunet6oqCjFx8dryZIlateunQzD0MCBAxUaGqpbb71V6enpWrp0qQYNGuQYIG02m8455xwVFhbq3nvvVZcuXeTj46OsrCxNnjz5D1dHPTw8/jB8norLL79cQ4cO1aJFi/Ttt9/qmWee0VNPPaVPPvnEsX4CAAAwV3OepX5/vquuuuqo61dJUq9evep9fqzXZRyxYUxDOZnv1bVrV+3YsUNffvmlFi9erIULF+rVV1/VAw88oIcffrjBMwGglALwO7t27frDsZ07d57wapTVatXZZ5+ts88+W88//7wef/xx/eMf/9BPP/2k4cOHKy4uThs3bpTdbq9Xsmzfvl2SFBcX5/jfH374QWVlZfWu8P32FrnftG/fXtLh29aHDx/+p17r0ZzqeYcOHaolS5YoPj5evXv3lp+fnxITExUQEKDFixcrNTW13hCzadMm7dy5U/PmzdPEiRMdx7/77ruTzhgXFye73a7du3fXu+r5+5/RbyIjI3XDDTfohhtuUF5envr06aPHHnuMUgoAgEbQWmep33YEttlsDTab/faa0tLS6t2VVldXp3379tUruX5/F9Sf5ePjoyuuuEJXXHGFampqNGrUKD322GO677775Onp2SDfA8D/8PY9APV8+umn9dYPWL16tVatWnXcAqOwsPAPx3r37i1Jqq6uliSNGDFCOTk5+uCDDxzPqaur08svvyxfX1/H29dGjBihuro6vfbaa47n2Ww2vfzyy/XOHxYWpjPOOEOvv/66Dhw48Ifvf/DgwZN4tX90qucdOnSo9u3bpw8++MDxdj6r1apBgwbp+eefV21tbb31pH67SnfkVTnDMPTSSy+ddMbf/lnMnDmz3vEXX3yx3uc2m+0PbwcMCwtTVFSU458LAABoWK11lnJxcdHo0aO1cOFCbd68+S+fT5L69eun4OBgvfnmm6qrq3Mcf/fdd+u97U46XCZJUlFR0Sl/n98UFBTU+9zd3V3dunWTYRiqra390+cFcGzcKQW0El9//bXjStqRBg0a5LhSJh2+TXvIkCG6/vrrVV1drRdffFHBwcG65557jnnuRx55REuWLNEFF1yguLg45eXl6dVXX1Xbtm01ZMgQSdL06dP1+uuva/LkyVq7dq3atWunjz/+WMuWLdOLL74oPz8/SdJFF12kwYMHa8aMGdq3b5+6deumTz755A/linR4wdEhQ4aoZ8+euuaaa9S+fXvl5uZqxYoV2r9/vzZs2PCnflanct7fCqcdO3bo8ccfdxwfNmyYvv76a3l4eKh///6O4126dFGHDh101113KSsrS/7+/lq4cOEfBqvj6d27t6688kq9+uqrKi4u1qBBg/TDDz8oLS2t3vNKS0vVtm1bjRkzRomJifL19dX333+vlJQUPffcc3/qZwMAQGvFLHViTz75pH766SclJyfrmmuuUbdu3VRYWKjU1FR9//33Ry3fjsfd3V0PPfSQbr75Zp111lm6/PLLtW/fPs2dO1cdOnSod3dUhw4dFBgYqFmzZsnPz08+Pj5KTk7+w/qhx/O3v/1NERERGjx4sMLDw7Vt2zb9+9//1gUXXOD4+QJoYKbs+QegyRxvG2MdsW3ub9sYP/PMM8Zzzz1nxMTEGB4eHsbQoUONDRs21Dvn77cx/uGHH4xLLrnEiIqKMtzd3Y2oqCjjyiuvNHbu3Fnv63Jzc40pU6YYISEhhru7u9GzZ8+jbttbUFBgTJgwwfD39zcCAgKMCRMmGOvWrTvqNr+7d+82Jk6caERERBhubm5GdHS0ceGFFxoff/zxcX8uR77eozmV84aFhRmSjNzcXMexX375xZBkDB069A/P37p1qzF8+HDD19fXCAkJMa655hpjw4YNf3h9kyZNMnx8fI6ar7Ky0rjllluM4OBgw8fHx7jooouMzMzMelszV1dXG3fffbeRmJho+Pn5GT4+PkZiYqLx6quvHvdnAwAA/odZ6tiOnDuOzHjjjTcaMTExhpubmxEREWGcffbZxhtvvOF4zk8//WRIMj766KN6X/vbz/D3GWfOnGnExcUZHh4exoABA4xly5YZffv2Nc4777x6z/vss8+Mbt26Ga6urvXOc/rppxvdu3f/Q/5JkyYZcXFxjs9ff/11Y9iwYUZwcLDh4eFhdOjQwbj77ruN4uLiE/4sAPw5FsNohFXkADQ7+/btU3x8vJ555hndddddZscBAABoVpilmo7dbldoaKhGjRqlN9980+w4AP4C1pQCAAAAADilqqqqP+zGN3/+fBUWFuqMM84wJxSABsOaUgAAAAAAp7Ry5UrdfvvtuuyyyxQcHKzU1FS99dZb6tGjhy677DKz4wH4iyilAAAAAABOqV27doqJidHMmTNVWFiooKAgTZw4UU8++aTc3d3NjgfgL2JNKQAAAAAAADQ51pQCAAAAAABAk6OUAgAAAAAAQJNrlmtK2e12ZWdny8/PTxaLxew4AACgBTMMQ6WlpYqKipLV2ryv5zFDAQCApnCy81OzLKWys7MVExNjdgwAANCKZGZmqm3btmbH+EuYoQAAQFM60fzULEspPz8/SYdfnL+/v8lpAABAS1ZSUqKYmBjH/NGcMUMBAICmcLLzU7MspX673dzf35+BCgAANImW8HY3ZigAANCUTjQ/Ne+FEQAAAAAAANAsUUoBAAAAAACgyVFKAQAAAAAAoMlRSgEAAAAAAKDJUUoBAAAAAACgyVFKAQAAAAAAoMlRSgEAAAAAAKDJUUoBAAAAAACgyVFKAQAAAAAAoMlRSgEAAAAAAKDJUUoBAAAAAACgyVFKAQAAAAAAoMlRSgEAAAAAAKDJUUoBAAAAAACgyVFKAQAAAAAAoMlRSgEAgGbrxe936pdd+WbHAAAAaDYyCip03yebVFljMzuKXM0OAAAA8Gd8kJKhF7/fJVerRT/eeYZig73NjgQAAODUSqtqNW1einbllcluN/TUmF6m5uFOKQAA0Oys3FOgfyzaLEm66ayOFFIAAAAnUGez6+b31mlXXpnC/T10+zmdzI5EKQUAAJqX9IJyXffOWtXZDV3YK1K3np1gdiQAAACn9/hX2/XfHQfl6WbV/03sr4gAT7MjUUoBAIDmo7iyVlPnpqioolaJbQP07GWJslgsZscCAABwagtWZWj2sr2SpOcv762ebQNMTnQYpRQAAGgW6mx23bQgVbsPlisywFNvTuwnTzcXs2MBAAA4teVp+Xrgs8PLHtx5TieN6BlpcqL/oZQCAADNwr++3Kqlu/Ll5eaiNyf2U5i/+becAwAAOLM9B8t0/bupqrMbuqR3lG46q6PZkeqhlAIAAE7v7RX7NG9FuiTphSt6q0e0c9xyDgAA4KyKK2p19bw1Kq6sVVJsoJ4a3cvplj2glAIAAE5t6a6DeuiLrZKke87rrPN6RJicCAAAwLnV2uy6YcFa7ckvV1SAp96Y4JzLHlBKAQAAp5WWV6Yb3k2VzW5oVJ9oXX96B7MjAQAAODXDMPTQ51u0LK1A3u4uemtyf4X6eZgd66gopQAAgFM6VF6jq+elqLSqTv3i2uiJUT2d7pZzAAAAZzNv+T69uypDFov00tgkdY30NzvSMVFKAQAAp1NTZ9f1767VvoIKtW3jpVkT+srD1fluOQcAAHAm/92Rp0e+PLzswYzzuuicbuEmJzo+SikAAOBUDMPQg59v1so9hfJxd9Fbk/orxNc5bzkHAABwFrtyS3XzgnWyG9Jlfdtq+rD2Zkc6IUopAADgVGYv26f3VmfKYpFeHpekzhF+ZkcCAABwaoXlNZo2b41Kq+s0oF2QHr20R7NY9oBSCgAAOI2ftufpsf8cvuX8HyO66qwuzn3LOQAAgNlq6uy67p21yiisUExQ81r2gFIKAAA4hR05pbr5vcO3nI/tH6NpQ+LNjgQAAODUDMPQPz/dpNV7C+Xn4arZk/oryMfd7FgnjVIKAACYrqCsWtPmpaisuk7J8UF65JLmccs5AACAmf5v6V59uGa/rL8ue5AQ3ryWPaCUAgAApqqus+nat9dq/6FKxQV7a9ZVfeXuyogCAABwPN9vzdXjX2+TJN1/YTed0TnM5ESnjokPAACYxjAM3ffJJq1JPyQ/T1e9Nam/2jSjW84BAADMsO1AiW59f50MQxqXHKvJg9qZHelPoZQCAACmmfXzHn2SmiUXq0Wvju+jjmG+ZkcCAABwagdLq3X1vDUqr7FpUIdgPXxx92a77AGlFAAAMMU3W3L09DfbJUkPXtRNQxNCTU4EAADg3Kpqbbr27TXKKqpUfIiPXh3fR24uzbfaab7JAQBAs7Ulu1i3vb9ehiFNHBiniQPbmR0JAADAqRmGoRkLNyo1o0j+nq56a1I/BXo372UPTrmUWrJkiS666CJFRUXJYrHo008/PeZzr7vuOlksFr344ov1jhcWFmr8+PHy9/dXYGCgpk2bprKyslONAgAAmqG80ipdPW+NKmttGpoQogcu7GZ2pEbH/AQAAP6qV/+7W5+uz5aL1aLXruqr9qHNf9mDUy6lysvLlZiYqFdeeeW4z1u0aJFWrlypqKioPzw2fvx4bdmyRd99952+/PJLLVmyRNOnTz/VKAAAoJmpqrXpmvlrdaC4Su1DffTvcX3k2oxvOT9ZzE8AAOCv+HrTAT3zzQ5J0iOXdNfgjiEmJ2oYrqf6Beeff77OP//84z4nKytLN998s7755htdcMEF9R7btm2bFi9erJSUFPXr10+S9PLLL2vEiBF69tlnjzqEAQCA5s8wDN398UZtyCxSoLebZk/qrwAvN7NjNQnmJwAA8GdtzirW7R+ulyRNHtRO45PjzA3UgBr80qTdbteECRN09913q3v37n94fMWKFQoMDHQMVJI0fPhwWa1WrVq1qqHjAAAAJzHzhzR9sSFbrlaLXhvfV+1CfMyO5DSYnwAAwNHkllRp2rwUVdXadXqnUP3zgq5mR2pQp3yn1Ik89dRTcnV11S233HLUx3NychQWFlY/hKurgoKClJOTc9Svqa6uVnV1tePzkpKShgsMAAAa3Zcbs/XC9zslSY+O7KGBHYJNTuRcGmN+kpihAABoziprbLpm/hrlllQrIcxXL49LanHLHjToq1m7dq1eeuklzZ07VxaLpcHO+8QTTyggIMDxERMT02DnBgAAjWtDZpHu/HCDJOnqIfEaOyDW5ETOpbHmJ4kZCgCA5spuN3TXRxu0cX+x2ni76a1J/eXv2fKWPWjQUmrp0qXKy8tTbGysXF1d5erqqvT0dN15551q166dJCkiIkJ5eXn1vq6urk6FhYWKiIg46nnvu+8+FRcXOz4yMzMbMjYAAGgkB4ordc38Naqus+usLmG6b0TLuuW8ITTW/CQxQwEA0Fy9+MMu/WfTAbm5WPT6hH6KDfY2O1KjaNC3702YMEHDhw+vd+zcc8/VhAkTNGXKFEnSwIEDVVRUpLVr16pv376SpB9//FF2u13JyclHPa+Hh4c8PDwaMioAAGhkFTV1unreGuWVVqtzuJ9eGttbLtaGvROoJWis+UlihgIAoDn6bH2WZv6wS5L0+KU9NSA+yOREjeeUS6mysjKlpaU5Pt+7d6/Wr1+voKAgxcbGKji4/hoRbm5uioiIUOfOnSVJXbt21XnnnadrrrlGs2bNUm1trW666SaNHTuWnWMAAGgh7HZDd3ywQVuySxTs467/m9RPfi3wlvOTxfwEAABOxrqMQ7r7442SpGuHtddl/Vr2W+9P+e17a9asUVJSkpKSkiRJd9xxh5KSkvTAAw+c9DneffdddenSRWeffbZGjBihIUOG6I033jjVKAAAwEk9990OLd6SI3cXq16f0FcxQS3zlvOTxfwEAABOJKuoUtfMX6uaOruGdw3XPed1MTtSo7MYhmGYHeJUlZSUKCAgQMXFxfL39zc7DgAAOMKidft1+weHFzZ//vJEjerT1uREf01Lmjta0msBAKAlKa+u05hZK7TtQIm6RPhp4fWD5OPRoCsuNamTnTla1l6CAADAVGvTC3Xvx5skSTec0aHZF1IAAACNzW43dNsH67XtQIlCfN311uT+zbqQOhWUUgAAoEHsP1Sh6fPXqsZm17ndw3XX3zqbHQkAAMDpPf3NDn23NVfurla9MbGfogO9zI7UZCilAADAX1ZWXadpc9eooLxG3SL99cIVvWVlpz0AAIDj+mhNpmb9vFuS9MyYXuoT28bkRE2LUgoAAPwlNruhW99bpx25pQr189Bbk/vJ27113HIOAADwZ6XsK9TfFx1e9uDmszrqkt7RJidqepRSAADgL3ny6236YXuePFytenNiP0UGtJ5bzgEAAP6MjIIKXfv2WtXaDI3oGaHbh3cyO5IpKKUAAMCf9kFKht5culeS9OxlieodE2huIAAAACdXWlWrafNSVFheo57RAXrusta77AGlFAAA+FNW7inQPxZtliTdNjxBFyVGmZwIAADAudXZ7Lr5vXXalVemcH8PvTmxn7zcXcyOZRpKKQAAcMrSC8p13TtrVWc3dGGvSN16doLZkQAAAJze419t1393HJSnm1X/N7G/IgI8zY5kKkopAABwSoorazV1boqKKmqV2DZAz16WKIuldd5yDgAAcLIWrMrQ7GWHlz14/vLe6tk2wORE5qOUAgAAJ63OZtdNC1K1+2C5IgM89ebEfvJ0a723nAMAAJyM5Wn5euCzw8se3HlOJ43oGWlyIudAKQUAAE7av77cqqW78uXl5qI3J/ZTmH/rvuUcAADgRPYcLNP176aqzm7okt5RuumsjmZHchqUUgAA4KS8vWKf5q1IlyS9cEVv9YjmlnMAAIDjKa6o1dXz1qi4slZJsYF6anQvlj04AqUUAAA4oaW7DuqhL7ZKku45r7PO6xFhciIAAADnVlNn1w0L1mpPfrmiAjz1xgSWPfg9V7MDAAAA52S3G/olLV/zV+zTD9vzZBjSqD7Ruv70DmZHAwAAcFoHS6v1/uoMvbsqQzklVfJ2d9Fbk/sr1M/D7GhOh1IKAADUU1xZq4Vr9+vtlenam1/uOD6iZ4SeGNWTW84BAAB+xzAMpWYUaf6Kffpq0wHV2gxJUoivu569LFFdI/1NTuicKKUAAIAkaXtOieavSNei1CxV1tokSX4erhrTr60mnBan9qG+JicEAABwLlW1Nn2+PlvzVuzTluwSx/E+sYGaNKidzusRIQ9X3rJ3LJRSAAC0YrU2u77bmqt5y/dp1d5Cx/HO4X6aOChOI3tHy8eDcQEAAOBImYUVemdluj5Yk6miilpJkoerVZf0jtLEge3YEOYkMWUCANAK5ZVW6f3VmXp3VbpyS6olSS5Wi87tHq6JA9spOT6It+kBAAAcwW43tDQtX/OX79OPOw6vtylJbdt4acJpcbq8X4za+LibG7KZoZQCAKCVON5aB+MGxOrK5FhFBniZnBIAAMC5HGu9zWGdQjVpYJzO6BwmFysX8/4MSikAAFo41joAAAA4day32fgopQAAaKEyCir0zqp0fchaBwAAACel1mbXt1tyNW/FPq1mvc1Gx08SAIAWhLUOAAAATh3rbZqDUgoAgBaguLJWH6/dr7dX7NO+ggrHcdY6AAAAOLrD620e0rzl6fp685HrbXpo3IAY1ttsApRSAAA0Y0dd68DTVZf1jdFVp8Wy1gEAAMDvHGu9zb5xbTRxYJzO7xEpd1eriQlbD0opAACaGdY6AAAAOHW/rbf5QUqmiitZb9MZMLECANBM5JVW6b1VmVqwuv5aB+d1j9DEgXEawFoHAAAA9djthpbsOqj5K9L10xHrbcYEHV5v87K+rLdpJkopAACc2InWOhiXHKeIAE+TUwIAADgX1ttsHiilAABwQpU1Nn2+IUvzV6Sz1gEAAMBJ2nbg8Hqbn65jvc3mgFIKAAAnYRiGduaWaWHqftY6AAAAOEmlVbX6746DentlOuttNjP8UwEAwESHymv0S1q+luw8qKW78pVTUuV4jLUOAAAA/shuN7Q5u1hLdh7Ukp35Ss04pDr74SUOWG+zeaGUAgCgCdXa7FqfWfTrEHVQG7OKHQtuSpKnm1WDO4RoXHIsax0AAAD8KrekynER75e0fBWW19R7vF2wty5OjGK9zWaGUgoAgEaWWVihn38toVbsLlBpdV29x7tE+GlYp1ANTQhR/3ZB8nRzMSkpAACAc6iqtSllX6GW7jp8R/n2nNJ6j/t6uGpQh2AN6xSqYQmhig32Nikp/gpKKQAAGlh5dZ1W7C7Qkl2Hr+btzS+v93gbbzcNSQjVsIQQDesUqnB/ruYBAIDWzTAMpeWVacmvJdSqvQWqqrU7HrdYpF7RAb9eyAtVUmyg3FzY9KW5o5QCAOAvstsNbT1Qop93HtTSXQe1Nv2Qam3/e0+eq9WiPrFtNKzT4RKqe1QAb8sDAACtXnFF7RFrax5UdnFVvcfD/T00NCFUwzqFakjHEAWxxmaLQykFAMCfkFdapV9+vZL3S1q+8svqr2sQG+StYZ1CNDQhVIM6BMvP082kpAAAAM6hzmbXhv1F+nnn4Rlq4/4i2Y9YW9Pd1ark+CAN+7WI6hTuy0LlLRylFAAAJ6G6zqa1+w7p512Hd3nZdqCk3uM+7i4aeMS6Bu1CfExKCgAA4Dz2H6rQkl9LqGW781VaVX9tzYQw38PzU6dQJceztmZrQykFAMBRGIahPfnljl3yVu4pVGWtrd5zekT7O67k9YltI3dX1jUAAACtW0VNnVbuKThcRO06qD0H66+tGeDlpiEJIRqWcPiO8qhAL5OSwhlQSgEA8KviylotTzs8QC3Zma+sosp6j4f6eWhoQohO7xSqwR1DFOLrYVJSAAAA52AYh9fW/G2XvDX7DqnG9r8Fyl2sFvWOCfz1Ql6IerUNZG1NOFBKAQBaLZvd0Ib9Rb8urpmv9ZlFsh2xsIG7i1X949scXmAzIVRdI/1Y1wAAALR6+WXVjrU1l+zKV35Zdb3HowO9NKxTqE7vFKKBHUIU4MXamjg6SikAQKtisxtauadAi9Zl6butuSqurK33ePtQHw1LCNXpnUKV3D5I3u78VQkAAJBXUqXPN2Tr8w3Z2ri/uN5jXm6/rq2ZEKKhnULVPsSHC3k4KUzaAIBWYduBEn26Lkufrc9WTsn/thv283TVkI4hGtYpVEMTQtS2jbeJKQEAAJxHeXWdvt2ao09Ss7QsLb/eTnndIv01tFOITk8IVd92beThygLlOHWUUgCAFiunuEqfrc/SonVZ2p5T6jju7+mqC3pFaWTvKPWNayNXFxYoBwAAkKQ6m13Ldhfo03VZ+mZLjipq/rfRS1JsoEYlRevcHhEK8/M0MSVaCkopAECLUlpVq8Wbc/Tp+iwt310g49creu4uVp3VJUwjk6J1ZpdQruYBAAD8yjAMbcku0aJ1Wfp8Q7YOlv5vjai4YG9dmhStkb2j1S7Ex8SUaIkopQAAzV6tza6luw5q0bpsfbc1R1W1/9vxpX+7Nro0qa1G9IxQoLe7iSkBAACcy/5DFfpsfbY+XZelXXlljuNtvN10UWKURiZFKykmkPWh0GgopQAAzZJhGNqwv1ifrsvSFxuyVVBe43isfaiPRiVF65Le0YoJYo0oAACA3xRX1urrTQe0aF2WVu0tdBx3d7XqnG7hurR3tIZ1CpW7K8sboPFRSgEAmpXMwgotWpelT9dlaU9+ueN4sI+7LkqM0qg+0eoZHcAVPQAAgF/V1Nn13x15+nR9lr7flqeauv/dVX5a+yCNSmqr83pGyN/TzcSUaI0opQAATq+ookZfbjygT9dlaU36IcdxTzer/tYtQpf2idaQjiFyY8FyAAAASYfvKk/NKNKidfv15cYDKqqodTzWKdxXlya11cW9oxQd6GViSrR2lFIAAKdUXWfTT9vz9Elqln7akada2+EVyy0WaXCHEF36684vvh78VQYAAPCbvfnljrvKMworHMdD/Tx0SWKULu0TrW6R/txVDqfAJA8AcBp2u6E16Ye0aN1+/WfjAZVU1Tke6xrpr1FJ0booMUoRAWxBDAAA8JuCsmp9ufHwOlHrM4scx73dXXRe98N3lQ/qECIXK0UUnAulFADAdGl5ZVq0br8+XZetrKJKx/EIf09dkhSlS5Oi1SXC38SEAAAAzqWq1qbvt+VqUWqWft55UHX2w3eVWy3S0IRQjeoTrXO6hcvbnV/74bz4txMAYIqDpdX6YkO2Fq3L0qasYsdxXw9Xnd/j8BW95PhgrugBAAD8ym43tHJvgT5dl6WvN+WotPp/d5X3jA7QpUnRujAxUmF+3FWO5oFSCgDQZCpq6vTd1lx9kpqlX9LyZfv1ip6r1aLTO4Xq0j7RGt41XJ5uLiYnBQAAcB47ckq1aF2WPlufpQPFVY7j0YFeGvnrXeUdw/xMTAj8OZRSAIBGZRiGfknL16J1Wfpmc47Ka2yOx3rHBGpUn2hd0DNSwb4eJqYEAABwLgdLq/Xpuix9si5L2w6UOI77ebrqwl6RujSprfrFtZGVu8rRjFFKAQAajd1u6O6PN2ph6n7Hsdggb41MitalSdGKD/ExMR0AAIBz2pxVrCvfXKnSXzd9cXOx6MzOYRrVJ1pndA7jrnK0GJRSAIBG8+Ti7VqYul8uVouuHBCjS5Paqk9sIFsQAwAAHEN6Qbkmz1mt0qo6dQ7304SBcbqwV6QCvd3NjgY0OEopAECjeGPJbr2xZI8k6enRvTS6b1uTEwEAADi3vNIqTXhrtfLLatQt0l/vX3ua/D3dzI4FNBqr2QEAAC3PwrX79fhX2yVJ953fhUIKAADgBEqqajV5dooyCisUG+StuVP7U0ihxaOUAgA0qB+35+qehRslSdcMjde1p3cwOREAAIBzq6q1afr8Ndp6oEQhvu56e9oAhfl5mh0LaHSnXEotWbJEF110kaKiomSxWPTpp586HqutrdW9996rnj17ysfHR1FRUZo4caKys7PrnaOwsFDjx4+Xv7+/AgMDNW3aNJWVlf3lFwMAMNfa9EO64d1U2eyGRiVF677zu5odCXAKzE8AgGOx2Q3d/sF6rdxTKF8PV82dMkBxwWwGg9bhlEup8vJyJSYm6pVXXvnDYxUVFUpNTdX999+v1NRUffLJJ9qxY4cuvvjies8bP368tmzZou+++05ffvmllixZounTp//5VwEAMN2u3FJNnZuiqlq7zugcqqfG9GKLYuBXzE8AgKMxDEMPfLZZX2/OkbuLVW9M6Kse0QFmxwKajMUwDONPf7HFokWLFmnkyJHHfE5KSooGDBig9PR0xcbGatu2berWrZtSUlLUr18/SdLixYs1YsQI7d+/X1FRUSf8viUlJQoICFBxcbH8/f3/bHwAQAPJLqrU6NeW60BxlZJiA/Xu1cnydmcvDbQMDT13mDU/NcZrAQD8NS9+v1Mvfr9LFov0yrg+GtEz0uxIQIM42Zmj0deUKi4ulsViUWBgoCRpxYoVCgwMdAxUkjR8+HBZrVatWrXqqOeorq5WSUlJvQ8AgHM4VF6jibNX60BxlTqE+mj2pP4UUsBf1BDzk8QMBQDO7J2V6Xrx+12SpEcu6UEhhVapUUupqqoq3XvvvbryyisdzVhOTo7CwsLqPc/V1VVBQUHKyck56nmeeOIJBQQEOD5iYmIaMzYA4CRV1NRp6rwUpeWVKTLAU/OnJauNj7vZsYBmraHmJ4kZCgCc1VebDuj+zzZLkm45O0ETToszORFgjkYrpWpra3X55ZfLMAy99tprf+lc9913n4qLix0fmZmZDZQSAPBn1drsuvHdVK3LKFKAl5vmTx2g6EAvs2MBzVpDzk8SMxQAOKPlu/N12/vrZRjSuORY3T48wexIgGka5f0Vvw1U6enp+vHHH+u9fzAiIkJ5eXn1nl9XV6fCwkJFREQc9XweHh7y8PBojKgAgD/Bbjd078KN+mnHQXm6WTV7cn8lhPuZHQto1hp6fpKYoQDA2WzOKtb0+WtVY7PrvO4R+tclPWSxsDEMWq8Gv1Pqt4Fq165d+v777xUcHFzv8YEDB6qoqEhr1651HPvxxx9lt9uVnJzc0HEAAI3gycXb9UlqllysFr06vo/6xrUxOxLQrDE/AUDLl15QrslzUlRWXafT2gfpxbG95cJOxWjlTvlOqbKyMqWlpTk+37t3r9avX6+goCBFRkZqzJgxSk1N1ZdffimbzeZY5yAoKEju7u7q2rWrzjvvPF1zzTWaNWuWamtrddNNN2ns2LEnvXMMAMA8byzZrTeW7JEkPTW6l87qEm5yIsD5MT8BQOuWV1qlCW+tVn5ZtbpG+uuNif3k6eZidizAdBbDMIxT+YL//ve/OvPMM/9wfNKkSXrooYcUHx9/1K/76aefdMYZZ0iSCgsLddNNN+mLL76Q1WrV6NGjNXPmTPn6+p5UBrYzBgBzLFy7X3d+tEGSdN/5XXTt6R1MTgQ0voaYO5xhfmqo1wIAODWlVbW64vWV2nqgRDFBXlp4/SCF+XmaHQtoVCc7c5xyKeUMGKgAoOn9tD1PV89fI5vd0NVD4vWPC7qyBgJahZY0d7Sk1wIAzUFVrU1T5qRoxZ4Chfi66+PrBqldiI/ZsYBGd7IzR6PtvgcAaDnWph/S9e+ulc1u6NKkaP19BIUUAADA8djshm7/YL1W7CmQr4er5k4ZQCEF/A6lFADguHbllmrq3BRV1dp1RudQPT2ml6wsygkAAHBMhmHogc826+vNOXJ3seqNCX3VIzrA7FiA06GUAgAcU3ZRpSbOXq3iylr1jgnUq+P7yM2FvzoAAACO56UfdundVRmyWKQXruitQR1DzI4EOCV+swAAHNWh8hpNnL1aB4qr1CHUR3Mm95e3+ylv2goAANCqvLMyXS9+v0uS9MjF3XVBr0iTEwHOi1IKAPAHFTV1mjovRWl5ZYrw99T8aclq4+NudiwAAACn9tWmA7r/s82SpFvOTtCEge3MDQQ4OUopAEA9tTa7bnw3VesyihTg5ab50wYoOtDL7FgAAABObfnufN32/noZhnTlgFjdPjzB7EiA06OUAgA42O2G7l24UT/tOChPN6tmT+6nTuF+ZscCAABwapuzijV9/lrV2Ow6r3uEHh3Zg52KgZNAKQUAcHhq8XZ9kpolF6tFr4zro75xQWZHAgAAcGrpBeWaPCdFZdV1So4P0otje8uFnYqBk0IpBQCQJL25ZI9eX7JHkvTkqJ46u2u4yYkAAACc28HSak2cvVr5ZdXqGumvNyf1k6ebi9mxgGaDUgoAoIVr9+uxr7ZJkmac30WX9YsxOREAAIBzK62q1eQ5q5VeUKGYIC/Nm9Jf/p5uZscCmhVKKQBo5X7anqd7Fm6UJF09JF7XDmtvciIAAADnVl1n0/T5a7Ulu0TBPu56e2qywvw9zY4FNDuUUgDQiqVmHNIN76bKZjd0aVK0/j6iK4tyAgAAHIfNbuj2D9ZrxZ4C+bi7aO6UAWoX4mN2LKBZopQCgFYqLa9UU+emqLLWptM7herpMb1kZVFOAACAYzIMQw9+vllfbcqRu4tVb0zsp55tA8yOBTRblFIA0AplF1VqwlurVVRRq8SYQL12VR+5ufBXAgAAwPHM/CFN76zMkMUiPX9FogZ3DDE7EtCs8RsIALQyRRU1mjR7tQ4UV6l9qI/mTO4vb3dXs2MBAAA4tXdWpuuF73dKkh6+uLsu7BVlciKg+aOUAoBWpLLGpqlzU7Qrr0wR/p56e1qygnzczY4FAADg1L7edED3f7ZZknTLWR01cWA7cwMBLQSlFAC0ErU2u25ckKrUjCL5e7pq/rQBig70MjsWAACAU1uxu0C3vr9ehiFdOSBWt5/TyexIQItBKQUArYBhGJqxcJN+3J4nD1erZk/ur07hfmbHAgAAcGpbsos1ff4a1djsOrd7uB4d2YOdioEGRCkFAK3Ak19v18LU/XKxWvTq+D7q1y7I7EgAAABOLb2gXJNmp6i0uk7J8UF6aWySXNipGGhQlFIA0MK9uWSPXl+yR5L05KieOrtruMmJAAAAnNvB0mpNnL1a+WXV6hrprzcn9ZOnm4vZsYAWh1IKAFqwT1L367GvtkmS7j2viy7rF2NyIgAAAOdWWlWryXNWK72gQjFBXpo3pb/8Pd3MjgW0SJRSANBC/bQjT/d8vFGSNG1IvK47vb3JiQAAAJxbdZ1N1769VluySxTs4675U5MV5u9pdiygxaKUAoAWKDXjkG54J1V1dkMje0fpHyO6signAADAcdjshm7/YL2W7y6Qj7uL5k4ZoPgQH7NjAS0apRQAtDBpeaWaOjdFlbU2DesUqqfHJMrKopwAAADHZBiGHvp8i77alCM3F4ten9BPPdsGmB0LaPEopQCgBckuqtTEt1arqKJWiTGBem18H7m78kc9AADA8cz8IU1vr0yXxSK9cEVvDUkIMTsS0CrwmwoAtBDFFbWaNHu1sour1D7UR3Mm95ePh6vZsQAAAJzaglUZeuH7nZKkhy7qrgt7RZmcCGg9KKUAoIV44PPN2pVXpgh/T82fOkBBPu5mRwIAAHBqO3JK9eDnmyVJN5/VUZMGtTM3ENDKUEoBQAvww7ZcfbY+W1aLNGtCX7Vt4212JAAAAKdmsxu6Z+FG1doMDe8apjvO6WR2JKDVoZQCgGautKpW/1h0+Arf1UPbq3dMoLmBAAAAmoE5y/ZqQ2aR/Dxc9ejInuxUDJiAUgoAmrknv96unJIqxQV76/bhXOEDAAA4kYyCCj377Q5J0t8v6KqIAE+TEwGtE6UUADRjK/cU6N1VGZKkJ0b1lJe7i8mJAAAAnJthGJrxyUZV1do1sH2wxvaPMTsS0GpRSgFAM1VVa9OMhRslSVcOiNWgDmxdDAAAcCIfrsnU8t0F8nSz6olRvG0PMBOlFAA0Uy98v1P7CioU7u+h+0Z0MTsOAACA08stqdKj/9kmSbrznM5qF+JjciKgdaOUAoBmaOP+Ir25ZI8k6bGRPeXv6WZyIgAAAOdmGIb++elmlVbVKbFtgKYMbmd2JKDVo5QCgGam1mbXPR9vlN2QLkqM0vBu4WZHAgAAcHpfbcrRd1tz5Wq16KkxveTqwq/DgNn4rxAAmpnXf96t7TmlauPtpgcv6mZ2HAAAAKd3qLxGD36+WZJ0w5kd1SXC3+REACRKKQBoVtLySjXzhzRJ0oMXdVeIr4fJiQAAAJzfv/6zVfllNUoI89WNZ3YwOw6AX1FKAUAzYbMbuufjjaqx2XVm51Bd0jvK7EgAAABO77878vRJapYsFumpMb3k4epidiQAv6KUAoBm4u0V+5SaUSQfdxc9dinbFwMAAJxIWXWd/rHo8Nv2pgyKV5/YNiYnAnAkSikAaAYyCyv09Dc7JEkzRnRVVKCXyYkAAACc3zOLtyurqFJt23jprnM7mR0HwO9QSgGAkzMMQ39ftEkVNTYNaBek8QNizY4EAADg9FL2FWr+ynRJ0pOjesnb3dXkRAB+j1IKAJzcwtQsLd2VL3dXq54c3VNWK2/bAwAAOJ6qWpvuXbhRhiFd3q+thiSEmB0JwFFQSgGAE8srrdK/vtwqSbp9eCe1D/U1OREAAIDze/nHXdpzsFyhfh76x4huZscBcAyUUgDgxB76fIuKK2vVI9pf1wyNNzsOAACA09uSXaxZP++RJP3rkh4K8HYzORGAY6GUAgAntXjzAX21KUcuVoueGt1Lri78kQ0AAHA8dTa77vl4o2x2QyN6Rui8HhFmRwJwHPyGAwBOqLiiVvd/tkWSdN3p7dU9KsDkRAAAAM7vzaV7tSW7RAFebnro4u5mxwFwApRSAOCEHvtqqw6WVqt9qI9uPivB7DgAAABOb8/BMr3w/U5J0v0XdlOYn6fJiQCcCKUUADiZX3bl68M1+2WxSE+P7iVPNxezIwEAADg1u93QjIWbVFNn19CEEI3uE212JAAngVIKAJxIRU2dZnyyUZI08bQ49WsXZHIiAAAA5/fu6gyt3lcob3cXPX5pT1ksFrMjATgJlFIA4ESe/Wan9h+qVHSgl+4+r4vZcQAAAJxeVlGlnvxqmyTpnnM7KybI2+REAE4WpRQAOInUjEOas3yvJOmxS3vI18PV5EQAAADOzTAM/WPRJpXX2NQ3ro0mDGxndiQAp4BSCgCcQHWdTfd+vFGGIY3qE60zOoeZHQkAAMDpfbY+W//dcVDuLlY9NbqnXKy8bQ9oTiilAMAJvPLTbu3KK1OIr7vuv6Cb2XEAAACcXn5ZtR7+Yosk6ZazO6pjmJ/JiQCcKkopADDZtgMlevWnNEnSwxf3UBsfd5MTAQAAOL+Hv9iqQxW16hLhp2tP72B2HAB/AqUUAJiozmbXvQs3qs5u6G/dwjWiZ4TZkQAAAJzed1tz9cWGbFkt0jNjEuXmwq+2QHPEf7kAYKI5y/Zp4/5i+Xm66l8je7B9MQAAwAmUVNXqn59ukiRdM6y9erYNMDkRgD+LUgoATLIvv1zPfbdDkvTPC7oq3N/T5EQAAADO74mvtiu3pFrtgr11+/BOZscB8Beccim1ZMkSXXTRRYqKipLFYtGnn35a73HDMPTAAw8oMjJSXl5eGj58uHbt2lXvOYWFhRo/frz8/f0VGBioadOmqays7C+9EABoTgzD0IxPNqqq1q5BHYJ1eb8YsyMBaETMTwDQMJbvztd7qzMkSU+O7iVPNxeTEwH4K065lCovL1diYqJeeeWVoz7+9NNPa+bMmZo1a5ZWrVolHx8fnXvuuaqqqnI8Z/z48dqyZYu+++47ffnll1qyZImmT5/+518FADQz76dkauWeQnm6WfXkqF68bQ9o4ZifAOCvq6yx6b5PDr9tb3xyrE5rH2xyIgB/lcUwDONPf7HFokWLFmnkyJGSDl/li4qK0p133qm77rpLklRcXKzw8HDNnTtXY8eO1bZt29StWzelpKSoX79+kqTFixdrxIgR2r9/v6Kiok74fUtKShQQEKDi4mL5+/v/2fgAYIqc4iqd8/zPKq2u0z8v6Kqrh7Y3OxKA42joucOs+akxXgsANKXHv9qmN5bsUWSAp769fZj8PN3MjgTgGE525mjQNaX27t2rnJwcDR8+3HEsICBAycnJWrFihSRpxYoVCgwMdAxUkjR8+HBZrVatWrXqqOetrq5WSUlJvQ8AaI4Mw9A/P92k0uo6JcYEasrgeLMjATBZY81PEjMUgJZjQ2aR/m/pHknSY5f2oJACWogGLaVycnIkSeHh4fWOh4eHOx7LyclRWFhYvcddXV0VFBTkeM7vPfHEEwoICHB8xMSw9gqA5unLjQf0/bY8ublY9PToXnKx8rY9oLVrrPlJYoYC0DLU1Nl178KNshvSJb2jdFaX8BN/EYBmoVnsvnffffepuLjY8ZGZmWl2JAA4ZYXlNXro8y2SpBvP7KjOEX4mJwLQ0jFDAWgJZv28W9tzShXk464HLuxmdhwADahBS6mIiAhJUm5ubr3jubm5jsciIiKUl5dX7/G6ujoVFhY6nvN7Hh4e8vf3r/cBAM3Nv77cqoLyGnUO99MNZ3Q0Ow4AJ9FY85PEDAWg+duVW6qXfzy8G+mDF3VTsK+HyYkANKQGLaXi4+MVERGhH374wXGspKREq1at0sCBAyVJAwcOVFFRkdauXet4zo8//ii73a7k5OSGjAMATuOn7XlatC5LVov01JhecndtFjeqAmgCzE8AcHQ2u6F7Fm5Urc3Q2V3CdHHiyW3qAKD5cD3VLygrK1NaWprj871792r9+vUKCgpSbGysbrvtNj366KNKSEhQfHy87r//fkVFRTl2mOnatavOO+88XXPNNZo1a5Zqa2t10003aezYsSe9cwwANCelVbX6x6LD2xdPHRyv3jGB5gYC0OSYnwDg1M1bvk/rMork6+GqRy/tIYuFtTiBluaUS6k1a9bozDPPdHx+xx13SJImTZqkuXPn6p577lF5ebmmT5+uoqIiDRkyRIsXL5anp6fja959913ddNNNOvvss2W1WjV69GjNnDmzAV4OADifpxfvUHZxlWKDvHXH3zqZHQeACZifAODUZBZW6JlvdkiS7hvRRZEBXiYnAtAYLIZhGGaHOFUlJSUKCAhQcXExayMAcGqr9xbq8tcPb+m+4OpkDeoYYnIiAKeqJc0dLem1AGi5DMPQhLdW65e0fCXHB+m9a06TlR2LgWblZGcOFjUBgEZSVWvTvQs3SpLG9o+hkAIAADgJH63dr1/S8uXhatWTo3tRSAEtGKUUADSSl37Ypb355Qrz89B9I7qaHQcAAMDp5ZVU6dEvt0qS7jink+JDfExOBKAxUUoBQCPYnFWsN5bskSQ9OrKHArzcTE4EAADg/B74bItKqurUMzpA04bEmx0HQCOjlAKABlZrs+uejzfKZjd0Qa9I/a17hNmRAAAAnN7Xmw5o8ZYcuVotemp0L7m68Osq0NLxXzkANLA3luzR1gMlCvR200MXdTc7DgAAgNMrqqjR/Z9tkSRdf0YHdYtiMwagNaCUAoAGlJZXppd+2CVJeuDCbgr18zA5EQAAgPN79D/blF9WrQ6hPrrprI5mxwHQRCilAKCB2O2GZizcqJo6u07vFKpLk6LNjgQAAOD0luw8qI/X7pfFIj09ppc8XF3MjgSgiVBKAUADeWdVutakH5KPu4seu7SHLBa2LwYAADie8uo63ffJJknSpIHt1DcuyOREAJoSpRQANID9hyr01NfbJUn3nt9Fbdt4m5wIAADA+T3zzQ5lFVUqOtBLd5/b2ew4AJoYpRQA/EWGYegfizarvMamfnFtdFVynNmRAAAAnN7a9ELNW7FPkvTEqJ7y8XA1NxCAJkcpBQB/0aJ1Wfp550G5u1r11Jheslp52x4AAMDxVNXadM/HG2UY0pi+bTWsU6jZkQCYgFIKAP6Cg6XVeuTLrZKkW89OUIdQX5MTAQAAOL9XfkrT7oPlCvH10D8v6Gp2HAAmoZQCgL/goS+2qKiiVt0i/TV9WHuz4wAAADi9rdkleu2/uyVJ/7qkuwK93U1OBMAslFIA8Cd9syVH/9l4QC5Wi54e00tuLvyRCgAAcDx1NrvuXbhRdXZD53WP0Pk9I82OBMBE/AYFAH9CcWWt7v90syRp+rD26hEdYHIiAAAA5/fWL3u1KatY/p6ueuSS7mbHAWAySikA+BOe+Gqb8kqr1T7ER7eenWB2HAAAAKe3N79cz3+3U5L0zwu7Kczf0+REAMxGKQUAp+irTQf0fkqmJOnJ0b3k6eZiciIAAADnVllj020frFd1nV1DOobosr5tzY4EwAlQSgHAKVixu0C3vb9ekjR1cLwGxAeZGwgAAMDJ1drsunFBqjZkFinAy01PjOopi8VidiwAToBSCgBO0tbsEk2fv0Y1NrvO7R6uf7B9MQAAwHEZhqG/f7JJP27Pk4erVbMn91NMkLfZsQA4CUopADgJmYUVmjRntUqr6zQgPkgvjU2Si5UrfAAAAMfzzDc79NHa/XKxWvTKuD7qG8dd5gD+h1IKAE6goKxaE2ev1sHSanWJ8NObE/uxjhQAAMAJzFm2V6/+d7ck6fFLe2h4t3CTEwFwNpRSAHAc5dV1mjo3RXvzyxUd6KV5UwcowMvN7FgAAABO7YsN2Xrky62SpLv+1klX9I81OREAZ0QpBQDHUGuz6/p3U7Vhf7HaeLtp3tQBCmfrYgAAgOP6ZVe+7vhwvQxDmjQwTjee2dHsSACcFKUUAByF3W7ono83asnOg/Jyc9Hsyf3VMczX7FgAAABObXNWsa59e41qbYYu6BmpBy7qzk57AI6JUgoAjuLJxdu1aF2WXKwWvXpVHyXFtjE7EgAAgFNLLyjX5DmrVV5j08D2wXr+ikQ2hgFwXJRSAPA7by7ZozeW7JEkPT26l87sHGZyIgAAAOd2sPTwxjD5ZTXqFumv1yf2lYcrG8MAOD5KKQA4wqJ1+/XYV9skSTPO76LRfduanAgAAMC5lVXXacrc1UovqFBMkJfmTu0vf082hgFwYpRSAPCrn3ce1N0fbZQkTRsSr2uHtTc5EQAAgHOrrrPp2rfXaHNWiYJ93DV/arLC/NgYBsDJoZQCAEkbMot0/TtrVWc3dEnvKP1jRFcW5QQAADgOu93QnR9u0LK0Anm7u2jOlP6KD/ExOxaAZoRSCkCrt+dgmabMTVFFjU1DE0L0zJhEWVmUEwAA4JgMw9C//rNVX248IFerRbOu6qtebQPNjgWgmaGUAtCq5ZVUaeLs1Sosr1HP6AC9dlVfubvyRyMAAMDxzPp5j+Ys2ydJeu7yRA3rFGpuIADNEr95AWi1SqpqNWlOivYfqlS7YG/NmdJfvh6uZscCAABwah+tydRTi7dLkv55QVdd0jva5EQAmitKKQCtUlWtTdPnr9G2AyUK8fXQ/KnJCvH1MDsWAACAU/txe65mfLJJknTt6e119VA2hgHw51FKAWh1bHZDd3y4Xiv3FMrXw1Vzp/RXbLC32bEAAACcWmrGId3wbqpsdkOj+kRrxnldzI4EoJmjlALQqhiGoYc+36KvNuXI3cWqNyb0VY/oALNjAQAAOLW0vFJNnZuiqlq7zugcqqdG92KnYgB/GaUUgFbl3z+m6e2V6bJYpOevSNSgjiFmRwIAAHBqB4orNfGt1SqqqFViTKBeHd9Hbi78Kgngr+NPEgCtxnurM/TcdzslSQ9d1F0X9ooyOREAAIBzK66o1aTZq5VdXKX2oT6aM7m/vN3ZGAZAw6CUAtAqfLslR/9YdHhRzpvO7KhJg9qZGwgAAMDJVdXadPX8FO3MLVO4v4fmTx2gIB93s2MBaEEopQC0eCn7CnXze+tkN6Qr+sXozr91MjsSAACAU6uz2XXLe+uUsu+Q/DxdNW/qALVtw8YwABoWpRSAFm1HTqmmzU1RdZ1dw7uG6bFLe7AoJwAAwHEYhqH7P9usb7fmyt3Vqv+b2E9dIvzNjgWgBaKUAtBiZRVVatLs1SqpqlPfuDZ6+co+cmVRTgAAgON64ftdem91pqwWaebY3kpuH2x2JAAtFL+dAWiRDpXXaNLs1copqVJCmK/emtRPXu4uZscCAABwam+vTNfMH3ZJkv41sofO6xFpciIALRmlFIAWp7LGpqnzUpSWV6bIAE/NmzpAgd4sygkAAHA8X286oAc+2yxJuvXsBI1PjjM5EYCWjlIKQItSa7PrxgWpWpdRpAAvN82bOkBRgV5mxwIAAHBqK3YX6Nb318swpHHJsbpteILZkQC0ApRSAFoMwzD090826cftefJwtWr25H7qFO5ndiwAAACntjW7RNPnr1GNza5zu4frX5ewMQyApkEpBaDFeOabHfpo7X65WC16ZVwf9Y0LMjsSAACAU8ssrNCkOatVWl2nAfFBemlsklysFFIAmgalFIAWYc6yvXr1v7slSY9f2kPDu4WbnAgAAMC5FZRVa9Ls1TpYWq0uEX56c2I/ebqxMQyApkMpBaDZ+2JDth75cqsk6a6/ddIV/WNNTgQAAODcyqvrNHVuivbklys60Evzpg5QgJeb2bEAtDKUUgCatWVp+brjw8OLck4aGKcbz+xodiQAAACnVmuz6/p3U7Vhf7HaeB/eGCbc39PsWABaIUopAM3W5qxiXfv2WtXaDF3QM1IPXNSdRTkBAACOw243dM/HG7Vk50F5ublo9uT+6hjma3YsAK0UpRSAZim9oFyT56xWWXWdBnUI1vNXJLIoJwAAwAk8uXi7Fq3LkovVolev6qOk2DZmRwLQilFKAWh2DpZWa+Ls1covq1G3SH+9PqGvPFxZlBMAAOB43lyyR28s2SNJenp0L53ZOczkRABaO0opAM1KWXWdpsxdrfSCCsUEeWnu1P7y82RRTgAAgONZtG6/HvtqmyRpxvldNLpvW5MTAQClFIBmpKbOruveXqvNWSUK9nHX/KnJCvNjUU4AAIDj+XnnQd390UZJ0rQh8bp2WHuTEwHAYZRSAJoFu93QnR9t0C9p+fJ2d9GcKf0VH+JjdiwAAACntiGzSNe/s1Z1dkOX9I7SP0Z0ZWMYAE6DUgqA0zMMQ//6z1Z9sSFbrlaLZl3VV73aBpodCwAAwKntOVimKXNTVFFj09CEED0zJlFWNoYB4EQopQA4vVk/79GcZfskSc9dnqhhnULNDQQAAODk8kqqNHH2ahWW16hndIBeu6qv3F359Q+Ac2nwP5VsNpvuv/9+xcfHy8vLSx06dNC//vUvGYbheI5hGHrggQcUGRkpLy8vDR8+XLt27WroKABagI/WZOqpxdslSf+8oKsu6R1tciIAaHjMTwAaUnFlrSbNSdH+Q5VqF+ytOVP6y9fD1exYAPAHDf4n01NPPaXXXntN8+bNU/fu3bVmzRpNmTJFAQEBuuWWWyRJTz/9tGbOnKl58+YpPj5e999/v84991xt3bpVnp4sWgy0dja7oR+25Wr2sr1auadQknTt6e119VAW5QTQMjE/AWgIB4orNX9FuhasylBxZa1CfD00f2qyQnw9zI4GAEdlMY68BNcALrzwQoWHh+utt95yHBs9erS8vLz0zjvvyDAMRUVF6c4779Rdd90lSSouLlZ4eLjmzp2rsWPHnvB7lJSUKCAgQMXFxfL392/I+ABMVFZdp4/WZGru8n1KL6iQJLlYLZo0sJ3uv5BFOQGYoynmjqaYn5rqtQBoeuszizT7l736atMB1dkP/3oXF+ytV8f3UfeoAJPTAWiNTnbmaPA7pQYNGqQ33nhDO3fuVKdOnbRhwwb98ssvev755yVJe/fuVU5OjoYPH+74moCAACUnJ2vFihUnPVQBaDkyCys0b/k+fZCSqdLqOklSgJebxiXHauLAOEUGeJmcEAAaF/MTgFNVZ7Prmy2H7yxfm37IcTw5PkjThsTr7K7hcmFRcwBOrsFLqRkzZqikpERdunSRi4uLbDabHnvsMY0fP16SlJOTI0kKDw+v93Xh4eGOx36vurpa1dXVjs9LSkoaOjaAJmYYhtakH9LsX/bqmy05+vWintqH+mjq4HiN6hMtb3fWPgDQOjTG/CQxQwEtUXFlrT5IydC85enKKqqUJLm5WHRxYrSmDG6nHtHcGQWg+Wjw3/g+/PBDvfvuu1qwYIG6d++u9evX67bbblNUVJQmTZr0p875xBNP6OGHH27gpADMUFNn11ebDmj2sr3auL/YcXxoQoimDonX6QmhbFUMoNVpjPlJYoYCWpK9+eWau2yvPlq7XxU1NklSsI+7xp8Wp6tOi1WYH2vLAWh+GryUuvvuuzVjxgzHbeQ9e/ZUenq6nnjiCU2aNEkRERGSpNzcXEVGRjq+Ljc3V7179z7qOe+77z7dcccdjs9LSkoUExPT0NEBNKLC8hq9tzpD81fsU27J4av2Hq5WjeoTrSmD49Up3M/khABgnsaYnyRmKKC5MwxDK3YX6K1f9urHHXn6bTXgLhF+mjo4Xhf3jpKnm4u5IQHgL2jwUqqiokJWq7XeMRcXF9ntdklSfHy8IiIi9MMPPziGqJKSEq1atUrXX3/9Uc/p4eEhDw92jACao525pZqzbK8+Sc1Sdd3hPwfC/Dw0cWCcrhwQq2B2gwGARpmfJGYooLmqqrXp8/XZmr1sr7bnlDqOn90lTFOHxGtQh2A2gAHQIjR4KXXRRRfpscceU2xsrLp3765169bp+eef19SpUyVJFotFt912mx599FElJCQ4tjSOiorSyJEjGzoOABPY7YZ+3nVQs3/Zq6W78h3He0YHaNqQeI3oGSl3V+txzgAArQvzEwBJyiut0jsrM/TuynQVlNdIkrzcXHRZv7aaPKid2of6mpwQABpWg5dSL7/8su6//37dcMMNysvLU1RUlK699lo98MADjufcc889Ki8v1/Tp01VUVKQhQ4Zo8eLF8vTkfdBAc1ZZY9PC1P2as2yvdh8slyRZLdLfukVo2tB49Ytrw1U9ADgK5iegdduSXazZv+zTFxuyVWM7fIdkVICnJg1qp7H9YxXg7WZyQgBoHBbD+O2dyc1HSUmJAgICVFxcLH9/f7PjAK3egeJKzV+RrgWrMlRcWStJ8vVw1RX9YzR5UDvFBHmbnBAA/ryWNHe0pNcCNHc2u6EftuVq9rK9Wrmn0HG8T2ygpg1pr3O7h8vVhTvLATRPJztzsN86gD9tfWaRZv+yV19tOqA6++F+OzbIW1MGt9OYvm3l58lVPQAAgCOVVdfpozWZmrt8n9ILKiRJrlaLRvSM1JTB7ZQU28bkhADQdCilAJySOptd32w5fFVvbfohx/HT2gdp6uB4nd01XC5W3qIHAABwpMzCCs1bvk8fpGSqtLpOkhTg5aZxybGaODBOkQFeJicEgKZHKQXgpBRX1ur91Rmat3yfsourJEnuLlZdlBilKYPbqUd0gMkJAQAAnIthGFqTfkhvLd2rb7fm6Ncby9U+1EdTB8drVJ9oebvzKxmA1os/AQEc1978cs1Ztlcfr92vihqbJCnYx13jT4vTVafFKsyPBXYBAACOVFNn1382ZWv2L/u0KavYcXxoQoimDonX6QmhsnJnOQBQSgH4I8MwtHx3gWb/slc/7sjTb9shdInw09TB8bq4d5Q83VzMDQkAAOBkCstrtGBVuuavSFdeabUkycPVqlF9ojVlcLw6hfuZnBAAnAulFACHqlqbPl+frdnL9mp7TqkkyWKRhncN09TB8RrYIVgWC1f1AAAAjrQzt1Rzlu3VJ6lZqq6zS5LC/Dw0aVA7XTkgVkE+7iYnBADnRCkFQFlFlXp/dYYWrMpQQXmNJMnb3UWX9W2ryYPjFR/iY3JCAAAA51Jns+vH7Xl6e2W6lu7KdxzvGR2gaUPiNaJnpNxdrSYmBADnRykFtFI2u6GftudpweoM/XdHnmPhzehAL00aFKcr+sUqwNvN3JAAAABOJruoUu+nZOrDlEzllBze/MVqkc7tHqGpQ+LVL64Nd5YDwEmilAJamZziKr2fkqEPUjJ14Ndd9CRpUIdgjU+O07ndw+XqwlU9AACA39jshn7emacFqzL04/b/XcwL8nHXZX3b6qrT4hQT5G1uSABohiilgFbAZje0ZNdBLViVoR+25ToGqTbebrqsX4zG9o9R+1Bfc0MCAAA4mdySKn2QkqkPUjKVVVTpOH5a+yCN+/Vinocrm78AwJ9FKQW0YHklVfpwTabeW11/kBoQH6TxybE6r0cEgxQAAMAR7HZDS9Py9e7KdP2wPU+2X6/mBXq7aUyftroyOVYduJgHAA2CUgpoYex2Q7+k5WvBqgx9vy1Xdb8OUgFebhrdp63GJceoYxjbEQMAABwpr7RKH63Zr/dTMpRZeMTFvHZBGvfrxTxPNy7mAUBDopQCWoiDpdX6aG2m3l+dqYzCCsfxfnFtNC45ViN6RjJIAQAAHMFuN7R8d4EWrE7Xt1v+dzHP39NVo/q01fjkWCWEczEPABoLpRTQjBmGoRW7C/Tu6gx9uyVHtbbDg5Sfp6tG92mrKwfEqnMEgxQAAMCRCsqq9dHa/XpvdYbSC/53Ma9PbKDGJcfpgp6R8nLnYh4ANDZKKaAZKiyv0cdrD68VtTe/3HG8d0ygxiXH6qJeUQxSAAAARzAMQyv2FGjBqgx9c+TFPA9XXdonWuOSY9Ulwt/klADQulBKAc2EYRhatbdQC1ZlaPHmHNXY7JIkXw9XjUyK0rgBceoWxSAFAABwpEPlNVqYul8LVmdoz8H/XcxLbBtw+GJeYpS83fm1CADMwJ++gJMrqqjRwtQsLViVrt1HDFK92gZo3IDDg5SPB/8pAwAA/MYwDKXsO6QFq9L11eYc1dQdvpjn4+6iS5KiNW5ArHpEB5icEgDAb7KAEzIMQ2vTD2nBqgx9uemAY5DydnfRJb2jNT6ZQQoAAOD3iitqtTD18FpRu/LKHMd7RPtr3IA4Xdw7Sr5czAMAp8GfyIATKa6s1aJfby/fmfu/QapbpL/GJcfqkt5R8vN0MzEhAACAczEMQ6kZRXp3Vbr+s/GAqn+9mOfl5qJLekdpXHKserUNNDckAOCoKKUAkxmGoXWZRYfvitqYraraw4OUp5tVFydGaVxynBLbBshisZicFAAAwHmUVNXq03VZWrAqQ9tzSh3Hu0T4aXxyrC5JipY/F/MAwKlRSgEmKa2q1afrs/XuyvQ/DFLjkmM1kkEKAACgHsMwtGF/sRasStcXGw6ostYm6fDFvAt7Hb4rKikmkIt5ANBMUEoBTai6zqbU9CJ9tj5Ln2/IVkXN4UHKw9WqC3pFanxyrPrEtmGQAgAAOEJWUaV+3J6n91ZlaOuBEsfxTuG+GjcgVpf2aasALy7mAUBzQykFNCKb3dCW7GItSyvQ8t35StlX6Hh7niR1DDs8SI3qE61Ab3cTkwIAADiPwvIardhdoGW787U8LV/7Ciocj7m7WnVBz0iNS45Vvzgu5gFAc0YpBTQgwzC0+2C5lu/O17K0fK3cU6jiytp6zwnx9dCwTiEa2z9W/dsxSAEAAJRX12n1vkItT8vXsrSCendDSZKL1aJebQN0Qc9Ije7TVm18uJgHAC0BpRTwFx0orjx8J1RavpbtzlduSXW9x/08XJXcPkiDOoRocMcQdQr3pYgCAACtWk2dXesyDmnZ7sMz1PrMItXZjXrP6Rzup0EdgzW4Q4gGtA9irU0AaIEopYBTVFRx5O3kBdqTX17vcXdXq/rGttGQhBAN6hCsntEBcnWxmpQWAADAfHa7oa0HSrQsLV/LdhcoZW+hY5Hy37Rt46XBHUI0qGOwBnUIUaifh0lpAQBNhVIKOIGKmjql7DvkuBNqS3aJjCMu5FktUs+2gRrcIViDO4aob1wbebq5mBcYAADAZIZhaG9+ueNOqBV7ClRUUX9Jg2Afdw38dX4a3CFEscHeJqUFAJiFUgr4nVqbXeszi7Q87fDdUOsyDqnWVv928oQwXw3uePhOqOT2wez2AgAAWr3ckqrDd0L9usHLgeKqeo/7uLsouX2wBv1aRHUO95PVypIGANCaUUqh1bPbDW3LKXGUUKv3Fqqipv7t5NGBXo4BalCHYIX5e5qUFgAAwDkUV9RqxZ4CxwYvuw/+bkkDF6v6xAX++pa8EPVqGyA3ljQAAByBUgqtjmEYSi+ocKwJtWJPgQrLa+o9p423mwb9uqbB4A4higv2ZnFyAADQqlXW2LQmvdBxJ9TmrGIduTa5xSL1jA74dXOXYPWLC5KXO0saAACOjVIKrUJeSZWW7y7QsrR8Ld9doKyiynqPe7u7KDk+yFFEdY3w53ZyAADQqtXa7Nq4v9ixrmZqepFqbPZ6z+kQ6vPrneQhGtg+WAHeLGkAADh5lFJosdLyyvTOynQtS8vXrryyeo+5uViUFNtGg3+9kpcYE8jt5AAAoNWz2Q19kJKpH7blatXeQpVV19V7PDLA03En1KAOIYoIYEkDAMCfRymFFmnlngJdM2+NSn8dpCwWqXuUv2NNg/7t2sjbnX/9AQAAflNdZ9MdH27QfzYecBwL9HbTwPbBGtQxRIM7BCs+xIclDQAADYbfytHifLslRze9t041dXb1i2ujaUPidVr7YLXxcTc7GgAAgFMqr67TtW+v1S9p+XJzsejmsxJ0VpcwdYtkSQMAQOOhlEKL8mFKpmZ8slF2QzqnW7hevjJJnm4ssAkAAHAsheU1mjJntTbsL5a3u4vemNBPQxJCzI4FAGgFKKXQYsz6ebee/Hq7JOnyfm31+KU95co6UQAAAMeUVVSpCW+t0p6D5Wrj7aa5UwYoMSbQ7FgAgFaCUgrNnmEYeuLr7XpjyR5J0rWnt9eM87qw3gEAAMBxpOWVasJbq3WguEpRAZ6aPy1ZHcN8zY4FAGhFKKXQrNXZ7Lp34SYtTN0vSfr7iC6aPqyDyakAAACc27qMQ5oyN0VFFbXqGOar+VMHKCrQy+xYAIBWhlIKzVZVrU03LUjV99vy5GK16KnRvTSmb1uzYwEAADi1JTsP6rp31qqixqbeMYGaM7k/G8IAAExBKYVmqbiyVtfMW6PV+wrl4WrVK+P6aHi3cLNjAQAAOLXPN2Trzg/Xq9ZmaGhCiGZd1Vc+HvxKAAAwB38DodnJK6nSxNmrtT2nVH6ernprUn8NiA8yOxYAAIBTm79inx78fIsMQ7ooMUrPXZYod1c2hQEAmIdSCs1KekG5Jry1WhmFFQr189C8KQPULcrf7FgAAABOyzAMvfj9Lr30wy5J0sSBcXroou6yWtkUBgBgLkopNBtbsos1aXaK8suqFRvkrXemJSs22NvsWAAAAE7LZjf00Odb9PbKdEnSbcMTdOvZCexSDABwCpRSaBZW7SnQ1fPWqLS6Tl0j/TVvan+F+XmaHQsAAMBp1dTZdceH6/XlxgOyWKRHLu6uCQPbmR0LAAAHSik4ve+25uqmBamqrrNrQHyQ/m9SP/l7upkdCwAAwGmVV9fpunfWaumufLm5WPTCFb11Ya8os2MBAFAPpRSc2odrMnXfJ5tksxsa3jVc/x6XJE83F7NjAQAAOK3C8hpNmZuiDZlF8nZ30esT+mpoQqjZsQAA+ANKKTit13/erSe+3i5JuqxvWz0xqqdcXdghBgAA4Fiyiio18a1V2n2wXG283TRnygD1jgk0OxYAAEdFKQWnYxiGnvh6u95YskeSdO2w9ppxfhcW5AQAADiOtLxSTXhrtQ4UVykywFNvTxugjmF+ZscCAOCYKKXgVOpsds34ZJM+XrtfknTf+V107ekdTE4FAADg3NZnFmnKnNU6VFGrDqE+entasqICvcyOBQDAcVFKwWlU1dp004J1+n5brlysFj0xqqcu7xdjdiwAAACntnTXQV379lpV1NiUGBOoOZP7K8jH3exYAACcEKUUnEJxZa2umbdGq/cVysPVqn+P66NzuoWbHQsAAMCpfbEhW3d8uF61NkNDE0I066q+8vFgxAcANA/8jQXT5ZVWadLsFG07UCI/D1f936R+Sm4fbHYsAAAAp/b2in164PMtMgzpwl6Rev7y3nJ3ZVMYAEDzQSkFU6UXlGvCW6uVUVihEF8PzZ86QN2i/M2OBQAA4LQMw9CL3+/SSz/skiRNOC1OD13cXS5WNoUBADQvlFIwzdbsEk2cvVr5ZdWKDfLW29MGKC7Yx+xYAAAATstuN/TQF1s0f0W6JOm24Qm69ewEdikGADRLlFIwxeq9hZo2L0WlVXXqEuGn+VMHKMzf0+xYAAAATqumzq47PlyvLzcekMUiPXxxd00c2M7sWAAA/GmUUmhy32/N1Y0LUlVdZ9eAdkF6c1I/BXi5mR0LAADAaZVX1+m6d9Zq6a58ublY9NzlvXVxYpTZsQAA+EsopdCkPlqTqRmfbJLNbmh413D9e1ySPN1czI4FAADgtArLazRlboo2ZBbJ291Fs67qq2GdQs2OBQDAX0YphSbz+s+79cTX2yVJY/q21ZOjesrVhR1iAAAAjiW7qFIT3lql3QfLFejtpjmT+yspto3ZsQAAaBCN0ghkZWXpqquuUnBwsLy8vNSzZ0+tWbPG8bhhGHrggQcUGRkpLy8vDR8+XLt27WqMKHAChmHoia+2OQqp6cPa65kxvSikAAA4AvMTfi8tr0xjXluu3QfLFRngqY+vG0ghBQBoURq8FTh06JAGDx4sNzc3ff3119q6dauee+45tWnzv79An376ac2cOVOzZs3SqlWr5OPjo3PPPVdVVVUNHQcmq7PZdc/HG/X6kj2SpPvO76K/j+jKDjEAAByB+Qm/tz6zSJfNWq7s4iq1D/XRx9cPUscwP7NjAQDQoCyGYRgNecIZM2Zo2bJlWrp06VEfNwxDUVFRuvPOO3XXXXdJkoqLixUeHq65c+dq7NixJ/weJSUlCggIUHFxsfz9/RsyPhpQVa1NN7+3Tt9tzZXVIj05upcu7xdjdiwAAE5JU8wdTTE/ScxQzcXSXQd17dtrVVFjU2LbAM2ZMkBBPu5mxwIA4KSd7MzR4HdKff755+rXr58uu+wyhYWFKSkpSW+++abj8b179yonJ0fDhw93HAsICFBycrJWrFhx1HNWV1erpKSk3gecW0lVrSbOXq3vtubK3dWqWVf1pZACAOAYGmN+kpihmqMvN2Zr6twUVdTYNDQhRAuuOY1CCgDQYjV4KbVnzx699tprSkhI0DfffKPrr79et9xyi+bNmydJysnJkSSFh4fX+7rw8HDHY7/3xBNPKCAgwPERE0O54czySqt0xesrtXpvofw8XPX21AH6W/cIs2MBAOC0GmN+kpihmpu3V6br5vfWqdZm6IJekfq/Sf3k48G+RACAlqvBSym73a4+ffro8ccfV1JSkqZPn65rrrlGs2bN+tPnvO+++1RcXOz4yMzMbMDEaEgZBRW6bNYKbTtQohBfD71/7WlKbh9sdiwAAJxaY8xPEjNUc2EYhl78fqfu/3SzDEO66rRYzRybJA9XF7OjAQDQqBq8lIqMjFS3bt3qHevatasyMjIkSRERh++Yyc3Nrfec3Nxcx2O/5+HhIX9//3ofcD7bDpRo9KzlSi+oUEyQlxZeP1DdowLMjgUAgNNrjPlJYoZqDux2Qw99vkUvfn94J8Vbz07Qvy7pIRcrm8IAAFq+Bi+lBg8erB07dtQ7tnPnTsXFxUmS4uPjFRERoR9++MHxeElJiVatWqWBAwc2dBw0kdV7C3X56yt0sLRaXSL8tPC6QYoL9jE7FgAAzQLzU+tUU2fXrR+s17wV6bJYpIcv7q7bz+nELsUAgFajwd+kfvvtt2vQoEF6/PHHdfnll2v16tV644039MYbb0iSLBaLbrvtNj366KNKSEhQfHy87r//fkVFRWnkyJENHQdN4OtNB3TbB+tVXWfXgHZBenNSPwV4uZkdCwCAZoP5qfUprqjVTe+laumufLm5WPTc5b11cWKU2bEAAGhSDV5K9e/fX4sWLdJ9992nRx55RPHx8XrxxRc1fvx4x3PuuecelZeXa/r06SoqKtKQIUO0ePFieXp6NnQcNKI6m13PfrtTs37eLUka3jVM/x7XR55urH8AAMCpYH5qXbYdKNF176xVekGFvNxc9PqEvhrWKdTsWAAANDmLYRiG2SFOVUlJiQICAlRcXMzaCCYpKKvWze+t0/LdBZKkq4fEa8b5XeTq0uDvCAUAwFQtae5oSa+lufp0XZZmfLJRVbV2tW3jpVlX9VWPaNbgBAC0LCc7c7DHLE7ZhswiXf/OWmUXV8nb3UVPj+mlC3txuzkAAMCx1NTZ9fhX2zR3+T5J0rBOoXrpit5q4+NubjAAAExEKYVT8t7qDD342RbV2OxqH+KjWRP6qlO4n9mxAAAAnFZuSZVufDdVa9IPSZJuOaujbh3eiR32AACtHqUUTkpVrU0PfrZFH6zJlCT9rVu4nr08Uf6eLGgOAABwLKv3FurGBak6WFotP09XvXB5bw3vFm52LAAAnAKlFE5o/6EK3fBuqjbuL5bVIt35t866/vQOsnJ1DwAA4KgMw9CcZfv0+FfbVGc31DncT7Mm9FV8iI/Z0QAAcBqUUjiuX3bl6+b3UnWoolZtvN0088okDU1gdxgAAIBjqaip032fbNJn67MlSRcnRunJ0T3l7c7oDQDAkfibEUdlGIZe+3m3nv1mh+yG1DM6QK9d1Udt23ibHQ0AAMBp7csv13XvrNX2nFK5Wi36+4iumjK4nSwW7jAHAOD3KKXwB6VVtbrzww36dmuuJOmKfjF6+JLu8nRzMTkZAACA8/p+a65u/3C9SqvqFOLroVfH99GA+CCzYwEA4LQopVDPrtxSXfv2Wu3JL5e7i1UPX9JdVw6INTsWAACA07LZDb34/U69/GOaJKlvXBu9Or6Pwv09TU4GAIBzo5SCw382HtDdH29QRY1NUQGeevWqvuodE2h2LAAAAKdVVFGjW99fr593HpQkTR7UTn8f0VXurlaTkwEA4PwopaA6m11PLd6uN5fulSQN6hCsl69MUrCvh8nJAAAAnNfmrGJd985a7T9UKU83q54Y1VOXJrU1OxYAAM0GpVQrd7C0WjctSNWqvYWSpGtPb6+7/9ZZri5c3QMAADiWj9fu1z8WbVJ1nV2xQd6adVVfdYvyNzsWAADNCqVUK5aacUg3vJOqnJIq+bi76NnLEnV+z0izYwEAADitmjq7Hvlyi95ZmSFJOqtLmF64vLcCvN1MTgYAQPNDKdUKGYahd1Zl6JEvtqjWZqhDqI9en9BPHcN8zY4GAADgtA4UV+qGd1O1LqNIFot029mddPNZHWW1WsyOBgBAs0Qp1cpU1dr0j0WbtTB1vyTp/B4ReuayRPl68K8CAADAsazYXaCb30tVflmN/D1d9dLYJJ3ZJczsWAAANGs0Ea1IZmGFrntnrbZkl8hqke49r4umD2svi4WrewAAAEdjGIb+b+lePbl4u2x2Q10j/TXrqj6KC/YxOxoAAM0epVQr8fPOg7rlvXUqrqxVsI+7Xr4ySYM6hpgdCwAAwGmVV9fpnoUb9Z+NByRJo5Ki9dilPeXl7mJyMgAAWgZKqRbObjf0yk9pev77nTIMKTEmUK+N76OoQC+zowEAADit3QfLdN3ba7Urr0yuVoseuKibJpwWxx3mAAA0IEqpFqy4slZ3frhe32/LkySNS47Vgxd1k4crV/cAAACOZfHmHN310QaVVdcpzM9Dr13VR33jgsyOBQBAi0Mp1UJtzynRdW+v1b6CCrm7WvXoyB66vF+M2bEAAACcls1u6Llvd+jV/+6WJA2ID9K/xyUpzM/T5GQAALRMlFIt0GfrszRj4SZV1toUHeilWVf1Vc+2AWbHAgAAcFqF5TW69f11WrorX5I0bUi8ZpzfRW4uVpOTAQDQclFKtSC1Nrse/2qb5izbJ0kamhCimWOT1MbH3dxgAAAATmzj/iJd/06qsooq5eXmoqfG9NLFiVFmxwIAoMWjlGoh8kqqdOOCVKXsOyRJuvHMDrrjnM5ysbIYJwAAwLF8kJKh+z/bopo6u+JDfDTrqr7qHOFndiwAAFoFSqkWYM2+Qt3wbqrySqvl5+Gq5y5P1N+6R5gdCwAAwGlV19n00Odb9N7qTEnS8K7hev6KRPl7upmcDACA1oNSqhkzDEPzlu/To//Zpjq7oU7hvpp1VV+1D/U1OxoAAIDTyiqq1A3vrNWG/cWyWKS7/tZZ15/eQVbuMAcAoElRSjVTlTU23ffJRn26PluSdGGvSD01upd8PPhHCgAAcCzL0vJ183vrVFheo0BvN700Nkmndwo1OxYAAK0SDUYzlF5QrmvfXqvtOaVysVp03/ldNG1IvCwWru4BAAAcjWEYmvXzHj3zzXbZDal7lL9mXdVXMUHeZkcDAKDVopRqZn7cnqvb3l+vkqo6hfi669/j+ui09sFmxwIAAHBapVW1uvujjVq8JUeSdFnftvrXyB7ydHMxORkAAK0bpVQz8um6LN32wXpJUp/YQL06vq8iAjzNDQUAAODEKmtsuuL1ldp6oERuLhY9dHF3jRsQyx3mAAA4AUqpZmLPwTL9fdEmSdKVA2L08MU95O5qNTkVAACAc3vky63aeqBEIb7uenNiPyXFtjE7EgAA+BWlVDNQXWfTze+tU0WNTae1D9KjI3vKhd1hAAAAjuurTQf03uoMWSzSS2OTKKQAAHAy3GrTDDyzeIe2ZJeojbebXrwiiUIKAADgBPYfqtCMhRslSded3kGDO4aYnAgAAPwepZST+2lHnv7vl72SpGfGJLKGFAAAwAnU2eyOjWF6xwTqjnM6mR0JAAAcBaWUE8srqdJdH26QJE0e1E7Du4WbnAgAAMD5zfxhl9akH5Kfh6tevjJJbi6MvAAAOCP+hnZSdruhOz7coILyGnWN9NeM87uYHQkAAMDprdhdoJd/SpMkPTaqp2KCvE1OBAAAjoVSykm9sXSPfknLl5ebi16+Mkmebi5mRwIAAHBqh8prdPsH62UY0uX92urixCizIwEAgOOglHJC6zIO6dlvdkiSHrq4mzqG+ZqcCAAAwLkZhqG7P96onJIqtQ/10UMXdzc7EgAAOAFKKSdTUlWrW95fpzq7oQt6ReryfjFmRwIAAHB6b69M1/fbcuXuYtXMsUnydnc1OxIAADgBSiknYhiG/rloszILKxUd6KXHL+0pi8VidiwAAACntu1AiR79zzZJ0ozzu6hHdIDJiQAAwMmglHIiH6/dr883ZMvFatHMK5MU4OVmdiQAAACnVlFTp5vfW6eaOrvO6hKmKYPbmR0JAACcJEopJ7H7YJke/HyLJOmOczqpb1wbkxMBAAA4v399uVVpeWUK8/PQM2N6cZc5AADNCKWUE6ius+mW99aposamge2Ddd3pHcyOBAAA4PT+s/GA3ludKYtFevGK3gr29TA7EgAAOAWUUk7gqa93aEt2idp4u+nFsb3lYuUKHwAAwPFkFlZoxicbJUk3nNFBgzqGmJwIAACcKkopk/24PVezl+2VJD17WaLC/T1NTgQAAODc6mx23fr+OpVW1SkpNlC3De9kdiQAAPAnUEqZKK+kSnd9dPgK3+RB7XR213CTEwEAADi/l37YpdSMIvl5uGrm2CS5uTDSAgDQHPE3uEnsdkO3f7heheU16hbpr/tGdDE7EgAAgNNbvjtf//4pTZL0+KieignyNjkRAAD4syilTDJryW4tSyuQl5uLZl6ZJA9XF7MjAQAAOLXC8hrd/sF6GYZ0Rb8YXZQYZXYkAADwF1BKmSA145Ce+3anJOnhi7urY5ivyYkAAACcm2EYuufjDcotqVaHUB89eHE3syMBAIC/iFKqiZVU1eqW99bJZjd0UWKULuvX1uxIAAAATm/e8n36flue3F2sevnKPvJ2dzU7EgAA+IsopZqQYRj6+yebtP9Qpdq28dJjl/aQxWIxOxYAAIBT25pdose/2i5J+vuILuoW5W9yIgAA0BAopZrQR2v368uNB+RitWjmlUny93QzOxIAAIBTq6ip083vparGZtfwrmGaNKid2ZEAAEADoZRqIml5ZXrwsy2SpDv/1kl9YtuYnAgAAMD5Pfz5Vu0+WK5wfw89PSaRu8wBAGhBKKWaQFWtTTe/t06VtTYN7his64Z1MDsSAACA0/tiQ7Y+WJMpi0V64YreCvJxNzsSAABoQJRSTeCpxdu17UCJgnzc9fzlvWW1coUPAADgeDILK/T3TzZJkm48o6MGdQgxOREAAGholFKN7IdtuZqzbJ8k6bnLEhXu72luIAAAACdXa7PrlvfXqbS6Tn1iA3Xb8ASzIwEAgEZAKdWIckuqdNdHGyRJUwfH68wuYSYnAgAAcH4vfr9T6zKK5OfpqpfGJsnVhZEVAICWiL/hG4nNbui299frUEWtukf5697zO5sdCQAAwOktT8vXq//dLUl6clQvxQR5m5wIAAA0FkqpRjLr591asadA3u4uevnKJHm4upgdCQAAwKkVlFXrtg/WyzCkKwfE6IJekWZHAgAAjajRS6knn3xSFotFt912m+NYVVWVbrzxRgUHB8vX11ejR49Wbm5uY0dpMmvTD+n573ZKkh6+uLvah/qanAgAADQnrXF+MgxDd3+8UXml1eoY5qsHLuxudiQAANDIGrWUSklJ0euvv65evXrVO3777bfriy++0EcffaSff/5Z2dnZGjVqVGNGaTLFlbW65b11stkNXZwYpTF925odCQAANCOtcX6SpLnL9+nH7Xlyd7Xq5SuT5OXOXeYAALR0jVZKlZWVafz48XrzzTfVpk0bx/Hi4mK99dZbev7553XWWWepb9++mjNnjpYvX66VK1c2VpwmYRiG/r5ok7KKKhUT5KXHLu0hi8VidiwAANBMtMb5SZI2ZxXria+2S5L+eUFXdY30NzkRAABoCo1WSt1444264IILNHz48HrH165dq9ra2nrHu3TpotjYWK1YseKo56qurlZJSUm9D2f04ZpM/WfjAblaLZo5Nkl+nm5mRwIAAM1IQ85PUvOYocqr63TLe+tUY7NreNdwTTgtzuxIAACgibg2xknff/99paamKiUl5Q+P5eTkyN3dXYGBgfWOh4eHKycn56jne+KJJ/Twww83RtQGk5ZXqoc+3ypJuvNvnZUU2+YEXwEAAPA/DT0/Sc1jhnr4iy3ak1+uCH9PPTOmF3eZAwDQijT4nVKZmZm69dZb9e6778rT07NBznnfffepuLjY8ZGZmdkg520oVbU23bRgnSprbRrSMUTXDmtvdiQAANCMNMb8JDn/DPX5hmx9uGa/LBbpxbG91cbH3exIAACgCTV4KbV27Vrl5eWpT58+cnV1laurq37++WfNnDlTrq6uCg8PV01NjYqKiup9XW5uriIiIo56Tg8PD/n7+9f7cCZPfr1d23NKFezjrucvT5TVyhU+AABw8hpjfpKce4bKKKjQPz7ZJEm6+cyOOq19sMmJAABAU2vwt++dffbZ2rRpU71jU6ZMUZcuXXTvvfcqJiZGbm5u+uGHHzR69GhJ0o4dO5SRkaGBAwc2dJxG9/3WXM1dvk+S9OxliQrzb7irmwAAoHVobfNTrc2uW95fp9LqOvWNa6Nbzk4wOxIAADBBg5dSfn5+6tGjR71jPj4+Cg4OdhyfNm2a7rjjDgUFBcnf318333yzBg4cqNNOO62h4zSqnOIq3f3xBknStCHxOrNLmMmJAABAc9Sa5idJev67nVqfWSR/T1e9NLa3XF0abe8dAADgxBplofMTeeGFF2S1WjV69GhVV1fr3HPP1auvvmpGlD/NZjd02wfrdKiiVj2i/XXPeZ3NjgQAAFqwljA/SdIvu/I16+fdkqQnR/dS2zbeJicCAABmsRiGYZgd4lSVlJQoICBAxcXFpq2N8O8fd+nZb3fK291FX948RO1DfU3JAQAAGpczzB0NxezXUlBWrfNeWqqDpdW6ckCsnhjVs8kzAACAxneyMwf3Sv8Ja9ML9cL3uyRJj1zSg0IKAADgBOx2Q3d9tEEHS6uVEOarBy7sZnYkAABgMkqpU1RcWatb3lsvm93QyN5RGt0n2uxIAAAATm/O8n36acdBubta9fK4JHm5u5gdCQAAmIxS6hQYhqG/f7JJWUWVig3y1r9G9pDFYjE7FgAAgFPbnFWsJ7/eJkm6/4Ku6hLRvN8GCQAAGgal1Cl4PyVT/9l0QK5Wi2ZemSQ/TzezIwEAADi18uo63fzeOtXaDP2tW7iuOi3O7EgAAMBJUEqdpF25pXr4iy2SpLvP7azeMYHmBgIAAGgGHvx8i/bmlysywFNPj+nFXeYAAMCBUuokVNXadPN761RVa9fQhBBdM7S92ZEAAACc3mfrs/Tx2v2yWqQXruitQG93syMBAAAnQil1Ep74apu255QqxNddz12eKKuVK3wAAADHk1FQoX8s2ixJuumsBJ3WPtjkRAAAwNlQSp3At1tyNG9FuiTp2csSFebnaXIiAAAA51Zrs+vm99eprLpO/du10S1ndTQ7EgAAcEKUUsdxoLhS9yzcKEm6Zmi8zugcZnIiAAAA5/fctzu1IbNI/p6uenFsklxdGDkBAMAfMSEcg81u6Lb316uoolY9owN097ldzI4EAADg9JbuOqhZP++WJD09ppeiA71MTgQAAJwVpdQxvPJTmlbtLZSPu4tmXpkkd1d+VAAAAMeTX1atOz7cIEkanxyr83pEmpwIAAA4M5qWo1izr1Avfr9TkvSvkT0UH+JjciIAAADnZrcbuvPDDTpYWq1O4b66/8JuZkcCAABOztXsAM4oo7BCri7/3979xMRVr2Ecf4ahDFwCtGAKTABF01xsi72tSK/FxEWJjSE1TaONDSqxi7uZRiiJodFgF9piazSm2FBx4cpqXVj/NHGBSDBNbMGOGIlKayS1kbTERDtTmlac+d2FgSuWwjnD7fmdke8nYTFnFvPkzeHkyTszZzLUUF2qrevKbMcBAADwvfi133X52u8KZWaoc/s6ZS8J2o4EAAB8jqXULLauK9PKcL7Klv3DdhQAAIC0UJCzREf/828Nj8X0z5I823EAAEAaYCl1A1Ul+bYjAAAApJXMYIb+Vb7UdgwAAJAmuKcUAAAAAAAAPMdSCgAAAAAAAJ5jKQUAAAAAAADPsZQCAAAAAACA51hKAQAAAAAAwHMspQAAAAAAAOA5llIAAAAAAADwHEspAAAAAAAAeI6lFAAAAAAAADzHUgoAAAAAAACeYykFAAAAAAAAz7GUAgAAAAAAgOdYSgEAAAAAAMBzLKUAAAAAAADgOZZSAAAAAAAA8Fym7QCpMMZIkmKxmOUkAADg726qb0z1j3RGhwIAAF5w2p/ScikVj8clSeXl5ZaTAACAxSIej6ugoMB2jAWhQwEAAC/N158CJg3f9ksmkxobG1NeXp4CgcBNeY1YLKby8nKdP39e+fn5N+U1/i6YlTvMyx3m5Q7zco5ZubOY52WMUTweVzgcVkZGet/54GZ3qMV8nqSCebnDvJxjVu4wL3eYlzuLdV5O+1NaflIqIyNDZWVlnrxWfn7+ojpxFoJZucO83GFe7jAv55iVO4t1Xun+CakpXnWoxXqepIp5ucO8nGNW7jAvd5iXO4txXk76U3q/3QcAAAAAAIC0xFIKAAAAAAAAnmMpdQOhUEh79uxRKBSyHcX3mJU7zMsd5uUO83KOWbnDvOAE54k7zMsd5uUcs3KHebnDvNxhXnNLyxudAwAAAAAAIL3xSSkAAAAAAAB4jqUUAAAAAAAAPMdSCgAAAAAAAJ5jKQUAAAAAAADPsZSaxaFDh3TbbbcpOztb69ev18DAgO1IvtTR0aF77rlHeXl5Wr58ubZs2aKRkRHbsdLCiy++qEAgoJaWFttRfOunn37SY489pqKiIuXk5Ki6ulpffPGF7Vi+lEgk1N7ersrKSuXk5OiOO+7Q888/L37H4g+fffaZNm/erHA4rEAgoPfff3/G88YYPffccyotLVVOTo7q6+t19uxZO2F9YK55TU5Oqq2tTdXV1crNzVU4HNYTTzyhsbExe4HhK3QoZ+hQqaNDzY8O5Rwdam50KOfoT6ljKfUXR48eVWtrq/bs2aNoNKo1a9Zo06ZNGh8ftx3Nd/r7+xWJRHTy5En19PRocnJSDzzwgCYmJmxH87XBwUG9/vrruuuuu2xH8a1ffvlFdXV1WrJkiT7++GN98803evnll7Vs2TLb0Xxp//796urq0muvvaZvv/1W+/fv14EDB9TZ2Wk7mi9MTExozZo1OnTo0KzPHzhwQAcPHtThw4d16tQp5ebmatOmTbp69arHSf1hrnlduXJF0WhU7e3tikajeu+99zQyMqKHHnrIQlL4DR3KOTpUauhQ86NDuUOHmhsdyjn60wIYzFBbW2sikcj040QiYcLhsOno6LCYKj2Mj48bSaa/v992FN+Kx+NmxYoVpqenx9x///2mubnZdiRfamtrM/fdd5/tGGmjoaHB7NixY8axrVu3msbGRkuJ/EuSOXbs2PTjZDJpSkpKzEsvvTR97NdffzWhUMi8/fbbFhL6y1/nNZuBgQEjyZw7d86bUPAtOlTq6FDzo0M5Q4dyhw7lHB3KOfqTO3xS6k9+++03nT59WvX19dPHMjIyVF9fr88//9xisvRw6dIlSVJhYaHlJP4ViUTU0NAw4xzD9T788EPV1NTokUce0fLly7V27Vq98cYbtmP51oYNG9Tb26szZ85Ikr766iudOHFCDz74oOVk/jc6OqoLFy7M+J8sKCjQ+vXrue47dOnSJQUCAS1dutR2FFhEh1oYOtT86FDO0KHcoUOljg61MPSn/8m0HcBPfv75ZyUSCRUXF884XlxcrO+++85SqvSQTCbV0tKiuro6rV692nYcX3rnnXcUjUY1ODhoO4rv/fDDD+rq6lJra6ueeeYZDQ4O6qmnnlJWVpaamppsx/Od3bt3KxaLqaqqSsFgUIlEQnv37lVjY6PtaL534cIFSZr1uj/1HG7s6tWramtr0/bt25Wfn287DiyiQ6WODjU/OpRzdCh36FCpo0Oljv40E0sp/F9EIhENDw/rxIkTtqP40vnz59Xc3Kyenh5lZ2fbjuN7yWRSNTU12rdvnyRp7dq1Gh4e1uHDhylUs3j33Xf11ltv6ciRI1q1apWGhobU0tKicDjMvHDTTE5Oatu2bTLGqKury3YcIG3RoeZGh3KHDuUOHQpeoz9dj6/v/cktt9yiYDCoixcvzjh+8eJFlZSUWErlfzt37tTx48fV19ensrIy23F86fTp0xofH9e6deuUmZmpzMxM9ff36+DBg8rMzFQikbAd0VdKS0u1cuXKGcfuvPNO/fjjj5YS+dvTTz+t3bt369FHH1V1dbUef/xx7dq1Sx0dHbaj+d7UtZ3rvjtThercuXPq6enhXT7QoVJEh5ofHcodOpQ7dKjU0aHcoz/NjqXUn2RlZenuu+9Wb2/v9LFkMqne3l7de++9FpP5kzFGO3fu1LFjx/Tpp5+qsrLSdiTf2rhxo77++msNDQ1N/9XU1KixsVFDQ0MKBoO2I/pKXV3ddT+NfebMGd16662WEvnblStXlJEx83IeDAaVTCYtJUoflZWVKikpmXHdj8ViOnXqFNf9G5gqVGfPntUnn3yioqIi25HgA3Qod+hQztGh3KFDuUOHSh0dyh36043x9b2/aG1tVVNTk2pqalRbW6tXX31VExMTevLJJ21H851IJKIjR47ogw8+UF5e3vR3hwsKCpSTk2M5nb/k5eVdd5+I3NxcFRUVcf+IWezatUsbNmzQvn37tG3bNg0MDKi7u1vd3d22o/nS5s2btXfvXlVUVGjVqlX68ssv9corr2jHjh22o/nC5cuX9f33308/Hh0d1dDQkAoLC1VRUaGWlha98MILWrFihSorK9Xe3q5wOKwtW7bYC23RXPMqLS3Vww8/rGg0quPHjyuRSExf+wsLC5WVlWUrNnyADuUcHco5OpQ7dCh36FBzo0M5R39aALs//udPnZ2dpqKiwmRlZZna2lpz8uRJ25F8SdKsf2+++abtaGmBnzOe20cffWRWr15tQqGQqaqqMt3d3bYj+VYsFjPNzc2moqLCZGdnm9tvv908++yz5tq1a7aj+UJfX9+s16qmpiZjzB8/adze3m6Ki4tNKBQyGzduNCMjI3ZDWzTXvEZHR2947e/r67MdHT5Ah3KGDrUwdKi50aGco0PNjQ7lHP0pdQFjjLk56y4AAAAAAABgdtxTCgAAAAAAAJ5jKQUAAAAAAADPsZQCAAAAAACA51hKAQAAAAAAwHMspQAAAAAAAOA5llIAAAAAAADwHEspAAAAAAAAeI6lFAAAAAAAADzHUgoAAAAAAACeYykFAAAAAAAAz7GUAgAAAAAAgOdYSgEAAAAAAMBz/wXui9WzyaFGVAAAAABJRU5ErkJggg==",
      "text/plain": [
       "<Figure size 1200x500 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Code used from: https://gymnasium.farama.org/tutorials/blackjack_tutorial/#visualizing-the-training\n",
    "fig, axs = plt.subplots(ncols=2, figsize=(12, 5))\n",
    "axs[0].set_title(\"Episode rewards\")\n",
    "reward_moving_average = (\n",
    "    np.convolve(\n",
    "        np.array(env_returns).mean(axis=1), np.ones(episode_window), mode=\"valid\"\n",
    "    )\n",
    "    / episode_window\n",
    ")\n",
    "axs[0].plot(range(len(reward_moving_average)), reward_moving_average)\n",
    "axs[1].set_title(\"Episode lengths\")\n",
    "length_moving_average = (\n",
    "    np.convolve(\n",
    "        np.array(env_steps).mean(axis=1), np.ones(episode_window), mode=\"valid\"\n",
    "    )\n",
    "    / episode_window\n",
    ")\n",
    "axs[1].plot(range(len(length_moving_average)), length_moving_average)\n",
    "plt.tight_layout()\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "ename": "NameError",
     "evalue": "name 'test_returns' is not defined",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mNameError\u001b[0m                                 Traceback (most recent call last)",
      "Cell \u001b[0;32mIn[13], line 3\u001b[0m\n\u001b[1;32m      1\u001b[0m fig, axs \u001b[38;5;241m=\u001b[39m plt\u001b[38;5;241m.\u001b[39msubplots(ncols\u001b[38;5;241m=\u001b[39m\u001b[38;5;241m2\u001b[39m, figsize\u001b[38;5;241m=\u001b[39m(\u001b[38;5;241m12\u001b[39m, \u001b[38;5;241m5\u001b[39m))\n\u001b[1;32m      2\u001b[0m axs[\u001b[38;5;241m0\u001b[39m]\u001b[38;5;241m.\u001b[39mset_title(\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mTest Returns\u001b[39m\u001b[38;5;124m\"\u001b[39m)\n\u001b[0;32m----> 3\u001b[0m axs[\u001b[38;5;241m0\u001b[39m]\u001b[38;5;241m.\u001b[39mplot(\u001b[38;5;28mrange\u001b[39m(\u001b[38;5;28mlen\u001b[39m(\u001b[43mtest_returns\u001b[49m)), test_returns)\n\u001b[1;32m      4\u001b[0m axs[\u001b[38;5;241m1\u001b[39m]\u001b[38;5;241m.\u001b[39mset_title(\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mTest Lengths\u001b[39m\u001b[38;5;124m\"\u001b[39m)\n\u001b[1;32m      5\u001b[0m axs[\u001b[38;5;241m1\u001b[39m]\u001b[38;5;241m.\u001b[39mplot(\u001b[38;5;28mrange\u001b[39m(\u001b[38;5;28mlen\u001b[39m(test_steps)), test_steps)\n",
      "\u001b[0;31mNameError\u001b[0m: name 'test_returns' is not defined"
     ]
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 1200x500 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig, axs = plt.subplots(ncols=2, figsize=(12, 5))\n",
    "axs[0].set_title(\"Test Returns\")\n",
    "axs[0].plot(range(len(test_returns)), test_returns)\n",
    "axs[1].set_title(\"Test Lengths\")\n",
    "axs[1].plot(range(len(test_steps)), test_steps)\n",
    "plt.tight_layout()\n",
    "plt.show()"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Testing Videos"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Test Episode 0"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Test Episode 0 Return: 23.0\n"
     ]
    }
   ],
   "source": [
    "test_episode = 0\n",
    "print(f\"Test Episode {test_episode} Return: {test_returns[test_episode]}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<video src=\"./videos/gae_discrete_monte-carlo//CartPole-v1-episode-0.mp4\" controls  >\n",
       "      Your browser does not support the <code>video</code> element.\n",
       "    </video>"
      ],
      "text/plain": [
       "<IPython.core.display.Video object>"
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "Video(f\"{VIDEOS_ROOT_DIR}/{ENV_TYPE}-episode-{test_episode}.mp4\")"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Test Episode 1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Test Episode 1 Return: 29.0\n"
     ]
    }
   ],
   "source": [
    "test_episode = 1\n",
    "print(f\"Test Episode {test_episode} Return: {test_returns[test_episode]}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<video src=\"./videos/gae_discrete_monte-carlo//CartPole-v1-episode-1.mp4\" controls  >\n",
       "      Your browser does not support the <code>video</code> element.\n",
       "    </video>"
      ],
      "text/plain": [
       "<IPython.core.display.Video object>"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "Video(f\"{VIDEOS_ROOT_DIR}/{ENV_TYPE}-episode-{test_episode}.mp4\")"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Test Episode 2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Test Episode 2 Return: 70.0\n"
     ]
    }
   ],
   "source": [
    "test_episode = 2\n",
    "print(f\"Test Episode {test_episode} Return: {test_returns[test_episode]}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<video src=\"./videos/gae_discrete_monte-carlo//CartPole-v1-episode-2.mp4\" controls  >\n",
       "      Your browser does not support the <code>video</code> element.\n",
       "    </video>"
      ],
      "text/plain": [
       "<IPython.core.display.Video object>"
      ]
     },
     "execution_count": 19,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "Video(f\"{VIDEOS_ROOT_DIR}/{ENV_TYPE}-episode-{test_episode}.mp4\")"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Test Episode 3"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Test Episode 3 Return: 130.0\n"
     ]
    }
   ],
   "source": [
    "test_episode = 3\n",
    "print(f\"Test Episode {test_episode} Return: {test_returns[test_episode]}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<video src=\"./videos/gae_discrete_monte-carlo//CartPole-v1-episode-3.mp4\" controls  >\n",
       "      Your browser does not support the <code>video</code> element.\n",
       "    </video>"
      ],
      "text/plain": [
       "<IPython.core.display.Video object>"
      ]
     },
     "execution_count": 21,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "Video(f\"{VIDEOS_ROOT_DIR}/{ENV_TYPE}-episode-{test_episode}.mp4\")"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Test Episode 4"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Test Episode 4 Return: 187.0\n"
     ]
    }
   ],
   "source": [
    "test_episode = 4\n",
    "print(f\"Test Episode {test_episode} Return: {test_returns[test_episode]}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<video src=\"./videos/gae_discrete_monte-carlo//CartPole-v1-episode-4.mp4\" controls  >\n",
       "      Your browser does not support the <code>video</code> element.\n",
       "    </video>"
      ],
      "text/plain": [
       "<IPython.core.display.Video object>"
      ]
     },
     "execution_count": 23,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "Video(f\"{VIDEOS_ROOT_DIR}/{ENV_TYPE}-episode-{test_episode}.mp4\")"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Test Episode 5"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Test Episode 5 Return: 96.0\n"
     ]
    }
   ],
   "source": [
    "test_episode = 5\n",
    "print(f\"Test Episode {test_episode} Return: {test_returns[test_episode]}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<video src=\"./videos/gae_discrete_monte-carlo//CartPole-v1-episode-5.mp4\" controls  >\n",
       "      Your browser does not support the <code>video</code> element.\n",
       "    </video>"
      ],
      "text/plain": [
       "<IPython.core.display.Video object>"
      ]
     },
     "execution_count": 25,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "Video(f\"{VIDEOS_ROOT_DIR}/{ENV_TYPE}-episode-{test_episode}.mp4\")"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "flax",
   "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": "3.10.12"
  },
  "orig_nbformat": 4,
  "vscode": {
   "interpreter": {
    "hash": "0a8d1f2580cdfde5c5829808ec6fccc81a351d243fb8b1925f7928e44ccf575b"
   }
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}