import matplotlib.pyplot as plt
import tensorflow as tf
import gym
import numpy as np
import math
import reinforcement_learning as rl
# Training
env_name = 'Breakout-v0'
#env_name = 'SpaceInvaders-v0'
rl.checkpoint_base_dir = 'checkpoints_tutorial16/'
rl.update_paths(env_name=env_name)
rl.maybe_download_checkpoint(env_name=env_name)
agent = rl.Agent(env_name=env_name,
training=True,
render=True,
use_logging=False)
model = agent.model
replay_memory = agent.replay_memory
agent.run(num_episodes=1)
# Testing
agent.epsilon_greedy.epsilon_testing
agent.training = False
agent.reset_episode_rewards()
agent.render = True
agent.run(num_episodes=1)
import numpy as np
import math
import reinforcement_learning as rl
env_name = 'Pong-v0'
# env_name = 'Breakout-v0'
# env_name = 'SpaceInvaders-v0'
rl.checkpoint_base_dir = 'checkpoints/'
rl.update_paths(env_name=env_name)
agent = rl.Agent(
env_name=env_name,
training=True, # set to False to see the agent play
render=
False, # set to True to render the game (maybe it speeds up training?)
use_logging=True)
model = agent.model
replay_memory = agent.replay_memory
# Hint for testing: put the epsilon value to 0.1 or even less (0.01), which mean that the agent will choose
# a random action from the action set with 10% (1%) probability, instead of the max Q-value action.
# You should find it in the agent definition.
agent.run(num_episodes=None) # None = it won't stop until you stop it
log_q_values = rl.LogQValues()
log_reward = rl.LogReward()
log_loss = rl.LogLoss(env_name)
log_q_values.read()
def main():
agent = reinforcement_learning.Agent(0, 0)
agent.load(reinforcement_learning.WEIGHTS_PATH)
test(agent)