# PLE takes our game and the state_preprocessor. It will process the state
# for our agent.
game = Catcher(width=128, height=128)
env = PLE(game, fps=60, state_preprocessor=nv_state_preprocessor)
agent = Agent(env,
batch_size,
num_frames,
frame_skip,
lr,
discount,
rng,
optimizer="sgd_nesterov")
agent.build_model()
memory = ReplayMemory(max_memory_size, min_memory_size)
env.init()
for epoch in range(1, num_epochs + 1):
steps, num_episodes = 0, 0
losses, rewards = [], []
env.display_screen = False
# training loop
while steps < num_steps_train:
episode_reward = 0.0
agent.start_episode()
while env.game_over() == False and steps < num_steps_train:
state = env.getGameState()
#memory settings
max_memory_size = 100000
min_memory_size = 1000 #number needed before model training starts
epsilon_rate = (epsilon - epsilon_min) / epsilon_steps
#PLE takes our game and the state_preprocessor. It will process the state for our agent.
game = Catcher(width=128, height=128)
env = PLE(game, fps=60, state_preprocessor=nv_state_preprocessor)
agent = Agent(env, batch_size, num_frames, frame_skip, lr,
discount, rng, optimizer="sgd_nesterov")
agent.build_model()
memory = ReplayMemory(max_memory_size, min_memory_size)
env.init()
for epoch in range(1, num_epochs+1):
steps, num_episodes = 0, 0
losses, rewards = [], []
env.display_screen = False
#training loop
while steps < num_steps_train:
episode_reward = 0.0
agent.start_episode()
while env.game_over() == False and steps < num_steps_train:
state = env.getGameState()