def test_create_inputs(encoder_type, normalize, num_vector, num_visual):
vec_obs_shape = (5,)
vis_obs_shape = (84, 84, 3)
obs_shapes = []
for _ in range(num_vector):
obs_shapes.append(vec_obs_shape)
for _ in range(num_visual):
obs_shapes.append(vis_obs_shape)
h_size = 128
obs_spec = create_observation_specs_with_shapes(obs_shapes)
encoders, embedding_sizes = ModelUtils.create_input_processors(
obs_spec, h_size, encoder_type, h_size, normalize
)
total_output = sum(embedding_sizes)
vec_enc = []
vis_enc = []
for i, enc in enumerate(encoders):
if len(obs_shapes[i]) == 1:
vec_enc.append(enc)
else:
vis_enc.append(enc)
assert len(vec_enc) == num_vector
assert len(vis_enc) == num_visual
assert total_output == int(num_visual * h_size + vec_obs_shape[0] * num_vector)
if num_vector > 0:
assert isinstance(vec_enc[0], VectorInput)
for enc in vis_enc:
assert isinstance(enc, ModelUtils.get_encoder_for_type(encoder_type))
def test_create_encoders(encoder_type, normalize, num_vector, num_visual,
unnormalized_inputs):
vec_obs_shape = (5, )
vis_obs_shape = (84, 84, 3)
obs_shapes = []
for _ in range(num_vector):
obs_shapes.append(vec_obs_shape)
for _ in range(num_visual):
obs_shapes.append(vis_obs_shape)
h_size = 128
num_layers = 3
unnormalized_inputs = 1
vis_enc, vec_enc = ModelUtils.create_encoders(obs_shapes, h_size,
num_layers, encoder_type,
unnormalized_inputs,
normalize)
vec_enc = list(vec_enc)
vis_enc = list(vis_enc)
assert len(vec_enc) == (1 if unnormalized_inputs + num_vector > 0 else 0
) # There's always at most one vector encoder.
assert len(vis_enc) == num_visual
if unnormalized_inputs > 0:
assert isinstance(vec_enc[0], VectorAndUnnormalizedInputEncoder)
elif num_vector > 0:
assert isinstance(vec_enc[0], VectorEncoder)
for enc in vis_enc:
assert isinstance(enc, ModelUtils.get_encoder_for_type(encoder_type))
def test_min_visual_size():
# Make sure each EncoderType has an entry in MIS_RESOLUTION_FOR_ENCODER
assert set(ModelUtils.MIN_RESOLUTION_FOR_ENCODER.keys()) == set(EncoderType)
for encoder_type in EncoderType:
good_size = ModelUtils.MIN_RESOLUTION_FOR_ENCODER[encoder_type]
vis_input = torch.ones((1, good_size, good_size, 3))
ModelUtils._check_resolution_for_encoder(good_size, good_size, encoder_type)
enc_func = ModelUtils.get_encoder_for_type(encoder_type)
enc = enc_func(good_size, good_size, 3, 1)
enc.forward(vis_input)
# Anything under the min size should raise an exception. If not, decrease the min size!
with pytest.raises(Exception):
bad_size = ModelUtils.MIN_RESOLUTION_FOR_ENCODER[encoder_type] - 1
vis_input = torch.ones((1, bad_size, bad_size, 3))
with pytest.raises(UnityTrainerException):
# Make sure we'd hit a friendly error during model setup time.
ModelUtils._check_resolution_for_encoder(
bad_size, bad_size, encoder_type
)
enc = enc_func(bad_size, bad_size, 3, 1)
enc.forward(vis_input)
def test_create_inputs(encoder_type, normalize, num_vector, num_visual):
vec_obs_shape = (5, )
vis_obs_shape = (84, 84, 3)
obs_shapes = []
for _ in range(num_vector):
obs_shapes.append(vec_obs_shape)
for _ in range(num_visual):
obs_shapes.append(vis_obs_shape)
h_size = 128
vis_enc, vec_enc, total_output = ModelUtils.create_input_processors(
obs_shapes, h_size, encoder_type, normalize)
vec_enc = list(vec_enc)
vis_enc = list(vis_enc)
assert len(vec_enc) == (1 if num_vector >= 1 else 0)
assert len(vis_enc) == num_visual
assert total_output == int(num_visual * h_size +
vec_obs_shape[0] * num_vector)
if num_vector > 0:
assert isinstance(vec_enc[0], VectorInput)
for enc in vis_enc:
assert isinstance(enc, ModelUtils.get_encoder_for_type(encoder_type))
