def refactored_pipeline():
"""You should import the refactored methods here"""
# Step 1
# You should put the `load` method here:
X_train, X_test, y_train, y_test = src.load()
# Step 2
# You should put the `train` method here:
model, train_score = src.train(X_train, y_train)
# Step 3
# You should put the `evaluate` method here:
test_score = src.evaluate(model, X_test, y_test)
# Final Step
# Return the train_score and test_score
return train_score, test_score
agent = A2C(env_pool, net, device, optimizer, args.entropy_reg, writer,
args.rollout_len, args.normalize_advantage, args.gamma)
elif args.agent == 'PPO':
agent = PPO(env_pool, net, device, optimizer, args.entropy_reg, writer,
args.rollout_len, args.normalize_advantage,
args.ppo_batch_size, args.num_ppo_epochs, args.ppo_eps, args.gae_lambda,
args.gamma)
else:
raise BaseException('wrong agent')
# training:
if args.train_steps > 0:
print('========================= training ===========================')
agent.train(args.train_steps)
try:
os.mkdir('./checkpoints/' + args.environment + '/')
except FileExistsError:
pass
checkpoint = 'checkpoints/' + args.environment + '/' + args.agent + '.pth'
save(agent, checkpoint)
print('Training done. Checkpoint saved in \'{}\''.format(checkpoint))
if args.watch > 0:
if args.train_steps == 0:
checkpoint = 'checkpoints/' + args.environment + '/' + args.agent + '.pth'
load(checkpoint, agent)
print('========================= watching ===========================')
for _ in range(args.watch):
print(play_episode(environment, agent, True))
environment.close()
def test_load_return_length():
"""Test that load return an iterable of length 4"""
output = src.load()
assert output
def test_evaluate():
"""Test that evaluate returns a float"""
X_train, X_test, y_train, y_test = src.load()
clf, score = src.train(X_train, y_train)
test_score = src.evaluate(clf, X_test, y_test)
assert isinstance(test_score, float)
def test_train(C, gamma):
"""Test that train returns the expected types"""
X_train, X_test, y_train, y_test = src.load()
clf, score = src.train(X_train, y_train)
assert isinstance(clf, svm.SVC)
assert isinstance(score, float)
def test_load_return_shape(size):
"""Test that the return shapes are correct"""
X_train, X_test, y_train, y_test = src.load(train_size=size)
num_samples = 1797
assert X_train.shape == (int(num_samples * size), 64)
assert X_test.shape == (int(num_samples * (1 - size)) + 1, 64)