game.new_episode()
game_state = game.get_state()
misc = game_state.game_variables # [KILLCOUNT, AMMO, HEALTH]
prev_misc = misc
action_size = game.get_available_buttons_size()
img_rows, img_cols = 64, 64
# Convert image into Black and white
img_channels = 4 # We stack 4 frames
state_size = (img_rows, img_cols, img_channels)
agent = A2CAgent(state_size, action_size)
agent.actor = Networks.actor_network(state_size, action_size,
agent.actor_lr)
agent.critic = Networks.critic_network(state_size, agent.value_size,
agent.critic_lr)
# Start training
GAME = 0
t = 0
max_life = 0 # Maximum episode life (Proxy for agent performance)
# Buffer to compute rolling statistics
life_buffer, ammo_buffer, kills_buffer = [], [], []
for i in range(max_episodes):
game.new_episode()
game_state = game.get_state()
misc = game_state.game_variables
prev_misc = misc
