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
img_channels = 3 # Color Channels
# Convert image into Black and white
trace_length = 4 # RNN states
state_size = (trace_length, img_rows, img_cols, img_channels)
agent = A2CAgent(state_size, action_size, trace_length)
agent.model = Networks.a2c_lstm(state_size, action_size, agent.value_size,
agent.learning_rate)
# 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
