def test_ideal1(self):
"""
First use case for the structure that has nesting equal 1
root
| \ \
A B C
"""
# Input information
MAX_TREE_DEPTH = 3
fillers_shapes = generate_shapes(max_tree_depth=MAX_TREE_DEPTH,
role_shape=SINGLE_ROLE_SHAPE,
filler_shape=SINGLE_FILLER_SHAPE)
keras_joiner = build_tree_joiner_network(roles=roles,
fillers_shapes=fillers_shapes)
t_V_r0_P_r1 = elementary_join(
joiner_network=keras_joiner,
input_structure_max_shape=fillers_shapes,
basic_roles=roles,
basic_fillers=fillers,
subtrees=(get_filler_by(name='V',
order=order_case_active,
fillers=fillers),
get_filler_by(name='P',
order=order_case_active,
fillers=fillers)))
t_V_r0_P_r1_expected = [
np.array([[
0.,
0.,
0.,
0.,
0.,
]], dtype=np.float32),
np.array([[
[0., 0.],
[40., 0.],
[0., 10.],
[0., 0.],
[0., 0.],
]],
dtype=np.float32),
np.array([[[[0., 0.], [0., 0.]], [[0., 0.], [0., 0.]],
[[0., 0.], [0., 0.]], [[0., 0.], [0., 0.]],
[[0., 0.], [0., 0.]]]],
dtype=np.float32)
]
for level_idx in range(len(t_V_r0_P_r1)):
np.testing.assert_array_equal(t_V_r0_P_r1[level_idx],
t_V_r0_P_r1_expected[level_idx])
def test_active_voice_sentence_ideal(self):
t_active_voice = encode_active_voice_sentence(
roles=roles, fillers=fillers, fillers_order=order_case_active)
keras_active_passive_network = build_active_passive_network(
roles=roles,
dual_roles=dual_basic_roles_case_1,
fillers=fillers,
tree_shape=t_active_voice)
res = check_active_case(ap_net=keras_active_passive_network,
encoded_sentence=t_active_voice,
roles=roles,
fillers=fillers,
dual_roles=dual_basic_roles_case_1)
for filler_name, filler_raw_encoded in res.items():
expected_filler_value = get_filler_by(name=filler_name,
order=order_case_active,
fillers=fillers)
filler_encoded = np.reshape(
filler_raw_encoded[0][:len(expected_filler_value)],
expected_filler_value.shape)
np.testing.assert_array_almost_equal(expected_filler_value,
filler_encoded)
def test_ideal2(self):
"""
First use case for the structure that has nesting equal 1
root
| \ \
A B C
"""
# Input information
fillers = np.array([
[7, 0, 0, 0, 0], # A
[0, 4, 0, 0, 0], # V
[0, 0, 2, 0, 0], # P
[0, 0, 0, 5, 0], # Aux
[0, 0, 0, 0, 3], # by
])
roles = np.array([
[10, 0], # r_0
[0, 5], # r_1
])
order_case_active = ['A', 'V', 'P']
MAX_TREE_DEPTH = 3
SINGLE_ROLE_SHAPE = roles[0].shape
SINGLE_FILLER_SHAPE = fillers[0].shape
fillers_shapes = generate_shapes(max_tree_depth=MAX_TREE_DEPTH,
role_shape=SINGLE_ROLE_SHAPE,
filler_shape=SINGLE_FILLER_SHAPE)
keras_joiner = build_tree_joiner_network(roles=roles, fillers_shapes=fillers_shapes)
t_V_r0_P_r1 = elementary_join(joiner_network=keras_joiner,
input_structure_max_shape=fillers_shapes,
basic_roles=roles,
basic_fillers=fillers,
subtrees=(
get_filler_by(name='V', order=order_case_active, fillers=fillers),
get_filler_by(name='P', order=order_case_active, fillers=fillers)
))
t_active_voice = elementary_join(joiner_network=keras_joiner,
input_structure_max_shape=fillers_shapes,
basic_roles=roles,
basic_fillers=fillers,
subtrees=(
get_filler_by(name='A', order=order_case_active, fillers=fillers),
t_V_r0_P_r1
))
dual_basic_roles_case_1 = np.linalg.inv(roles)
fillers_shapes_unshift = generate_shapes_for_unshift(max_tree_depth=MAX_TREE_DEPTH - 1,
role_shape=SINGLE_ROLE_SHAPE,
filler_shape=SINGLE_FILLER_SHAPE)
keras_ex0_unshifter = build_tree_unshifter_network(roles=dual_basic_roles_case_1,
fillers_shapes=fillers_shapes_unshift,
role_index=0)
prepared_for_unshift = reshape_to_satisfy_max_depth_after_unshift(t_active_voice,
MAX_TREE_DEPTH - 1,
SINGLE_ROLE_SHAPE,
SINGLE_FILLER_SHAPE)
extracted_child = keras_ex0_unshifter.predict_on_batch([
*prepared_for_unshift
])
extracted_child = extracted_child.reshape((*extracted_child.shape[1:],))
t_active_voice_left_child_after_unshift = extract_per_level_tensor_representation_after_unshift(
extracted_child,
max_tree_depth=MAX_TREE_DEPTH - 1,
role_shape=SINGLE_ROLE_SHAPE,
filler_shape=SINGLE_FILLER_SHAPE)
# for level_idx in range(len(t_V_r0_P_r1)-1):
np.testing.assert_array_equal(
fillers[0],
t_active_voice_left_child_after_unshift[0][0]
)