default=0,
type=int,
help="Render some episodes.")
parser.add_argument("--threads",
default=32,
type=int,
help="Maximum number of threads to use.")
parser.add_argument("--evaluate",
default=True,
type=bool,
help="Run evaluation phase.")
args = parser.parse_args()
# Create the environment
env = cart_pole_pixels_evaluator.environment()
# Construct the network
network = Network(threads=args.threads)
network.construct(args, env.state_shape, env.actions)
# Load the checkpoint if required
if args.checkpoint:
# Try extract it from embedded_data
try:
import embedded_data_cart_pole_pixels_rudolf_ha_reinforce_baseline
embedded_data_cart_pole_pixels_rudolf_ha_reinforce_baseline.extract(
)
# print("embedded_data extracted")
except:
pass
# ##################################################################################################################
args = parser.parse_args([] if "__file__" not in globals() else None)
# Fix random seeds and threads
np.random.seed(args.seed)
tf.random.set_seed(args.seed)
tf.config.threading.set_inter_op_parallelism_threads(args.threads)
tf.config.threading.set_intra_op_parallelism_threads(args.threads)
# Report only errors by default
if not args.verbose:
os.environ["TF_CPP_MIN_LOG_LEVEL"] = "3"
# Create the environment
env = cart_pole_pixels_evaluator.environment(seed=args.seed)
possible_actions = list(range(env.actions))
# Construct the network
network = Network(env, args)
# Training
for _ in range(args.episodes // args.batch_size):
batch_states, batch_actions, batch_returns = [], [], []
# Batch over multiple episodes (failed / finished)
for _ in range(args.batch_size):
# Perform episode
states, actions, rewards = [], [], []
state, done = env.reset(), False
while not done:
