print("Starting the training.")
time_start = time()
for epoch in range(epochs):
print("\nEpoch %d\n-------" % (epoch + 1))
train_episodes_finished = 0
# training
print("Training...")
train_scores = []
losses = []
game.new_episode()
for learning_step in trange(learning_steps_per_epoch, leave=False):
loss = model.step(training=True)
losses.append(loss)
if game.is_episode_finished():
score = game.get_total_reward()
train_scores.append(score)
game.new_episode()
train_episodes_finished += 1
train_scores = np.array(train_scores)
losses = np.array(losses)
print("%d training episodes played." % train_episodes_finished)
print("Current size of the memory buffer:", sys.getsizeof(model.memory))
print(
"Results: mean score: %.1f +/- %.1f," %
(train_scores.mean(), train_scores.std()),
"min: %.1f," % train_scores.min(), "max: %.1f," % train_scores.max())
actions,
file_name,
ddqn=use_ddqn,
parameter_exploration=use_parameter_exploration,
gpu=use_gpu,
loading=1)
print("======================================")
print("Testing trained neural network.")
print("Testing...")
test_scores = []
for _ in range(episodes_to_watch):
game.new_episode()
while not game.is_episode_finished():
model.step(training=False, showing=True)
# Sleep between episodes
sleep(1.0)
score = game.get_total_reward()
test_scores.append(score)
print("Total score: ", score)
test_scores = np.array(test_scores)
print("%d test episodes played." % episodes_to_watch)
print(
"Results: mean: %.1f +/- %.1f," % (test_scores.mean(), test_scores.std()),
"min: %.1f," % test_scores.min(), "max: %.1f," % test_scores.max())