def test_fit_without_trusted_region(self):
gcr = GaussianCNNRegressorWithModel(
input_shape=(10, 10, 3),
num_filters=(3, 6),
filter_dims=(3, 3),
strides=(1, 1),
padding='SAME',
hidden_sizes=(32, ),
output_dim=1,
adaptive_std=False,
use_trust_region=False)
train_data, test_data = get_train_test_data()
observations, returns = train_data
for _ in range(20):
gcr.fit(observations, returns)
paths, expected = test_data
prediction = gcr.predict(paths['observations'])
average_error = 0.0
for i in range(len(expected)):
average_error += np.abs(expected[i] - prediction[i])
average_error /= len(expected)
assert average_error <= 0.05
def test_is_pickleable(self):
input_shape = (28, 28, 3)
gcr = GaussianCNNRegressorWithModel(
input_shape=input_shape,
num_filters=(3, 6),
filter_dims=(3, 3),
strides=(1, 1),
padding='SAME',
hidden_sizes=(32, ),
output_dim=1,
adaptive_std=False,
use_trust_region=False)
with tf.variable_scope(
'GaussianCNNRegressorWithModel/GaussianCNNRegressorModel',
reuse=True):
bias = tf.get_variable('dist_params/mean_network/hidden_0/bias')
bias.load(tf.ones_like(bias).eval())
result1 = gcr.predict([np.ones(input_shape)])
h = pickle.dumps(gcr)
with tf.Session(graph=tf.Graph()):
gcr_pickled = pickle.loads(h)
result2 = gcr_pickled.predict([np.ones(input_shape)])
assert np.array_equal(result1, result2)
class GaussianCNNBaselineWithModel(Baseline):
"""
GaussianCNNBaseline With Model.
It fits the input data to a gaussian distribution estimated by a CNN.
Args:
env_spec (garage.envs.env_spec.EnvSpec): Environment specification.
subsample_factor (float): The factor to subsample the data. By
default it is 1.0, which means using all the data.
regressor_args (dict): Arguments for regressor.
"""
def __init__(
self,
env_spec,
subsample_factor=1.,
regressor_args=None,
name='GaussianCNNBaselineWithModel',
):
super().__init__(env_spec)
if regressor_args is None:
regressor_args = dict()
self._regressor = GaussianCNNRegressorWithModel(
input_shape=(env_spec.observation_space.shape),
output_dim=1,
name=name,
**regressor_args)
self.name = name
@overrides
def fit(self, paths):
"""Fit regressor based on paths."""
observations = np.concatenate([p['observations'] for p in paths])
returns = np.concatenate([p['returns'] for p in paths])
self._regressor.fit(observations, returns.reshape((-1, 1)))
@overrides
def predict(self, path):
"""Predict value based on paths."""
return self._regressor.predict(path['observations']).flatten()
@overrides
def get_param_values(self, **tags):
"""Get parameter values."""
return self._regressor.get_param_values(**tags)
@overrides
def set_param_values(self, flattened_params, **tags):
"""Set parameter values to val."""
self._regressor.set_param_values(flattened_params, **tags)
@overrides
def get_params_internal(self, **tags):
"""Get parameter values."""
return self._regressor.get_params_internal(**tags)
def test_fit_unnormalized(self):
gcr = GaussianCNNRegressorWithModel(
input_shape=(10, 10, 3),
num_filters=(3, 6),
filter_dims=(3, 3),
strides=(1, 1),
padding='SAME',
hidden_sizes=(32, ),
output_dim=1,
adaptive_std=True,
normalize_inputs=False,
normalize_outputs=False)
train_data, test_data = get_train_test_data()
observations, returns = train_data
for _ in range(20):
gcr.fit(observations, returns)
paths, expected = test_data
prediction = gcr.predict(paths['observations'])
average_error = 0.0
for i in range(len(expected)):
average_error += np.abs(expected[i] - prediction[i])
average_error /= len(expected)
assert average_error <= 0.1
x_mean = self.sess.run(gcr.model.networks['default'].x_mean)
x_mean_expected = np.zeros_like(x_mean)
x_std = self.sess.run(gcr.model.networks['default'].x_std)
x_std_expected = np.ones_like(x_std)
assert np.array_equal(x_mean, x_mean_expected)
assert np.array_equal(x_std, x_std_expected)
y_mean = self.sess.run(gcr.model.networks['default'].y_mean)
y_mean_expected = np.zeros_like(y_mean)
y_std = self.sess.run(gcr.model.networks['default'].y_std)
y_std_expected = np.ones_like(y_std)
assert np.allclose(y_mean, y_mean_expected)
assert np.allclose(y_std, y_std_expected)