def __init__(
self,
entity_counts=(5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15),
train_entity_counts=(5, 6, 7, 8, 9, 10, 11, 12, 14),
validation_entity_counts=(13, ),
test_entity_counts=(15, ),
validation_combinations=(('square', 'red',
'solid'), ('triangle', 'green', 'solid'),
('circle', 'blue', 'solid')),
test_combinations=(('rectangle', 'yellow',
'solid'), ('cross', 'magenta', 'solid'),
('ellipse', 'cyan', 'solid')),
caption_size=12,
vocabulary=('.', 'a', 'all', 'an', 'are', 'blue', 'circle',
'circles', 'cross', 'crosses', 'cyan', 'eight',
'ellipse', 'ellipses', 'few', 'five', 'four', 'gray',
'green', 'half', 'is', 'magenta', 'most', 'no', 'none',
'of', 'pentagon', 'pentagons', 'quarter', 'quarters',
'rectangle', 'rectangles', 'red', 'semicircle',
'semicircles', 'seven', 'shape', 'shapes', 'six',
'square', 'squares', 'the', 'third', 'thirds', 'three',
'triangle', 'triangles', 'two', 'yellow'),
language=None):
# world_generator = LimitedAttributesGenerator(
# shapes_range=(2, 4),
# colors_range=(2, 4),
# textures_range=(1, 1),
world_generator = ReinforcedAttributesGenerator(
reinforcement_range=(1, 3),
entity_counts=entity_counts,
collision_tolerance=0.0,
boundary_tolerance=0.0,
train_entity_counts=train_entity_counts,
validation_entity_counts=validation_entity_counts,
test_entity_counts=test_entity_counts,
validation_combinations=validation_combinations,
test_combinations=test_combinations,
max_provoke_collision_rate=0.0)
quantifier_captioner = QuantifierCaptioner(
restrictor_captioner=RegularTypeCaptioner(
hypernym_rate=1.0, logical_tautology_rate=1.0),
body_captioner=AttributeTypeRelationCaptioner(
attribute_type_captioner=CaptionerMixer(
captioners=(RegularAttributeCaptioner(),
RegularTypeCaptioner()))),
quantifiers=('ratio', ))
number_bound_captioner = NumberBoundCaptioner(
quantifier_captioner=quantifier_captioner)
world_captioner = CaptionerMixer(captioners=(quantifier_captioner,
number_bound_captioner),
distribution=[1, 1])
super(QuantificationRatioSimple,
self).__init__(world_generator=world_generator,
world_captioner=world_captioner,
caption_size=caption_size,
vocabulary=vocabulary,
language=language)
def __init__(self,
world_size=64,
entity_counts=(5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15),
train_entity_counts=(5, 6, 7, 8, 9, 10, 11, 12, 14),
validation_entity_counts=(13, ),
test_entity_counts=(15, ),
max_provoke_collision_rate=0.33,
collision_tolerance=0.2,
boundary_tolerance=0.2,
pixel_noise_stddev=0.0):
world_generator = ReinforcedAttributesGenerator(
reinforcement_range=(1, 3),
entity_counts=entity_counts,
world_size=world_size,
train_entity_counts=train_entity_counts,
validation_entity_counts=validation_entity_counts,
test_entity_counts=test_entity_counts,
max_provoke_collision_rate=max_provoke_collision_rate,
collision_tolerance=collision_tolerance,
boundary_tolerance=boundary_tolerance)
num_classes = len(world_generator.shapes) * len(
world_generator.colors) * len(world_generator.textures)
super(Countshape, self).__init__(world_generator=world_generator,
num_classes=num_classes,
multi_class=True,
class_count=True,
pixel_noise_stddev=pixel_noise_stddev)
def __init__(
self,
entity_counts=(5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15),
train_entity_counts=(5, 6, 7, 8, 9, 10, 11, 12, 14),
validation_entity_counts=(13,),
test_entity_counts=(15,),
validation_combinations=(('square', 'red', 'solid'), ('triangle', 'green', 'solid'), ('circle', 'blue', 'solid')),
test_combinations=(('rectangle', 'yellow', 'solid'), ('cross', 'magenta', 'solid'), ('ellipse', 'cyan', 'solid')),
caption_size=14,
vocabulary=('.', 'a', 'an', 'biggest', 'blue', 'circle', 'cross', 'cyan', 'darkest', 'ellipse', 'gray', 'green', 'is', 'leftmost', 'lightest', 'lowermost', 'magenta', 'most', 'pentagon', 'rectangle', 'red', 'rightmost', 'semicircle', 'shape', 'smallest', 'square', 'topmost', 'triangle', 'yellow'),
language=None
):
world_generator = ReinforcedAttributesGenerator(
reinforcement_range=(1, 1),
entity_counts=entity_counts,
train_entity_counts=train_entity_counts,
validation_entity_counts=validation_entity_counts,
test_entity_counts=test_entity_counts,
validation_combinations=validation_combinations,
test_combinations=test_combinations
)
world_captioner = ExistentialCaptioner(
restrictor_captioner=MaxAttributeCaptioner(
scope_captioner=RegularTypeCaptioner(
hypernym_rate=1.0,
logical_tautology_rate=1.0
)
),
body_captioner=AttributeTypeRelationCaptioner(
attribute_type_captioner=CaptionerMixer(
captioners=(
RegularAttributeCaptioner(),
RegularTypeCaptioner()
)
)
)
)
super(Maxattr, self).__init__(
world_generator=world_generator,
world_captioner=world_captioner,
caption_size=caption_size,
vocabulary=vocabulary,
language=language
)
def __init__(
self,
entity_counts=(5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15),
train_entity_counts=(5, 6, 7, 8, 9, 10, 11, 12, 14),
validation_entity_counts=(13,),
test_entity_counts=(15,),
validation_combinations=(('square', 'red', 'solid'), ('triangle', 'green', 'solid'), ('circle', 'blue', 'solid')),
test_combinations=(('rectangle', 'yellow', 'solid'), ('cross', 'magenta', 'solid'), ('ellipse', 'cyan', 'solid')),
caption_size=14,
vocabulary=('.', 'a', 'above', 'an', 'behind', 'below', 'bigger', 'biggest', 'blue', 'circle', 'closer', 'closest', 'cross', 'cyan', 'darker', 'darkest', 'ellipse', 'farther', 'farthest', 'from', 'front', 'gray', 'green', 'in', 'is', 'left', 'leftmost', 'lighter', 'lightest', 'lowermost', 'magenta', 'most', 'of', 'pentagon', 'rectangle', 'red', 'right', 'rightmost', 'semicircle', 'shape', 'smaller', 'smallest', 'square', 'than', 'the', 'to', 'topmost', 'triangle', 'yellow'),
language=None
):
world_generator = ReinforcedAttributesGenerator(
reinforcement_range=(1, 1),
entity_counts=entity_counts,
train_entity_counts=train_entity_counts,
validation_entity_counts=validation_entity_counts,
test_entity_counts=test_entity_counts,
validation_combinations=validation_combinations,
test_combinations=test_combinations
)
world_captioner = ExistentialCaptioner(
restrictor_captioner=RegularTypeCaptioner(),
body_captioner=RelationCaptioner(
reference_captioner=RegularTypeCaptioner(),
comparison_captioner=RegularTypeCaptioner()
)
)
super(Relational, self).__init__(
world_generator=world_generator,
world_captioner=world_captioner,
caption_size=caption_size,
vocabulary=vocabulary,
language=language
)
def __init__(
self,
world_size=64,
world_color='black',
shapes=('square', 'rectangle', 'triangle', 'pentagon', 'cross',
'circle', 'semicircle', 'ellipse'),
colors=('red', 'green', 'blue', 'yellow', 'magenta', 'cyan',
'gray'),
textures=('solid', ),
rotation=True,
size_range=(0.1, 0.25),
distortion_range=(2.0, 3.0),
shade_range=0.4,
collision_tolerance=0.25,
collision_shade_difference=0.5,
boundary_tolerance=None,
entity_counts=None,
train_entity_counts=(5, 6, 7, 9, 11, 12, 14),
validation_entity_counts=(8, 13),
test_entity_counts=(10, 15),
validation_count_rate=0.5,
test_count_rate=0.5,
validation_combinations=(('square', 'red',
'solid'), ('triangle', 'green', 'solid'),
('circle', 'blue', 'solid')),
test_combinations=(('rectangle', 'yellow',
'solid'), ('cross', 'magenta', 'solid'),
('ellipse', 'cyan', 'solid')),
validation_space_rate_range=(0.0, 1.0),
test_space_rate_range=(0.0, 1.0),
validation_combination_rate=0.5,
test_combination_rate=0.5,
max_provoke_collision_rate=0.33,
reinforcement_range=(1, 3),
quantifiers=None,
number_bounds=None,
comparative_quantifiers=None,
caption_size=15,
vocabulary=('.', 'a', 'all', 'an', 'are', 'as', 'at', 'blue',
'but', 'circle', 'circles', 'cross', 'crosses', 'cyan',
'eight', 'ellipse', 'ellipses', 'exactly', 'five',
'four', 'gray', 'green', 'half', 'is', 'least', 'less',
'magenta', 'many', 'more', 'most', 'no', 'none', 'not',
'of', 'one', 'pentagon', 'pentagons', 'quarter',
'quarters', 'rectangle', 'rectangles', 'red',
'semicircle', 'semicircles', 'seven', 'shape',
'shapes', 'six', 'square', 'squares', 'than', 'the',
'third', 'thirds', 'three', 'triangle', 'triangles',
'twice', 'two', 'yellow', 'zero'),
correct_ratio=0.5,
train_correct_ratio=None,
validation_correct_ratio=None,
test_correct_ratio=None,
worlds_per_instance=1,
captions_per_instance=1,
pixel_noise_stddev=0.0,
caption_realizer='dmrs',
language=None):
world_generator = ReinforcedAttributesGenerator(
world_size=world_size,
world_color=world_color,
shapes=shapes,
colors=colors,
textures=textures,
rotation=rotation,
size_range=size_range,
distortion_range=distortion_range,
shade_range=shade_range,
collision_tolerance=collision_tolerance,
collision_shade_difference=collision_shade_difference,
boundary_tolerance=boundary_tolerance,
entity_counts=entity_counts,
train_entity_counts=train_entity_counts,
validation_entity_counts=validation_entity_counts,
validation_count_rate=validation_count_rate,
test_entity_counts=test_entity_counts,
test_count_rate=test_count_rate,
validation_combinations=validation_combinations,
validation_space_rate_range=validation_space_rate_range,
validation_combination_rate=validation_combination_rate,
test_combinations=test_combinations,
test_space_rate_range=test_space_rate_range,
test_combination_rate=test_combination_rate,
max_provoke_collision_rate=max_provoke_collision_rate,
reinforcement_range=reinforcement_range)
body_captioner = AttributeTypeRelationCaptioner(
attribute_type_captioner=CaptionerMixer(
captioners=(RegularAttributeCaptioner(),
RegularTypeCaptioner(hypernym_rate=0.0))))
quantifier_captioner = QuantifierCaptioner(
restrictor_captioner=RegularTypeCaptioner(hypernym_rate=1.0),
body_captioner=body_captioner,
quantifiers=quantifiers)
number_bound_captioner = NumberBoundCaptioner(
quantifier_captioner=quantifier_captioner,
number_bounds=number_bounds)
comparative_quantifier_captioner = ComparativeQuantifierCaptioner(
restrictor_captioner=RegularTypeCaptioner(hypernym_rate=1.0),
comparison_captioner=RegularTypeCaptioner(hypernym_rate=1.0),
body_captioner=body_captioner,
comparative_quantifiers=comparative_quantifiers)
world_captioner = CaptionerMixer(
captioners=(quantifier_captioner, number_bound_captioner,
comparative_quantifier_captioner),
distribution=(1, 1, 1))
super(QuantificationComplexDataset,
self).__init__(world_generator=world_generator,
world_captioner=world_captioner,
caption_size=caption_size,
vocabulary=vocabulary,
correct_ratio=correct_ratio,
train_correct_ratio=train_correct_ratio,
validation_correct_ratio=validation_correct_ratio,
test_correct_ratio=test_correct_ratio,
worlds_per_instance=worlds_per_instance,
captions_per_instance=captions_per_instance,
pixel_noise_stddev=pixel_noise_stddev,
caption_realizer=caption_realizer,
language=language)
def __init__(
self,
world_size=64,
world_color='black',
shapes=('square', 'rectangle', 'triangle', 'pentagon', 'cross',
'circle', 'semicircle', 'ellipse'),
colors=('red', 'green', 'blue', 'yellow', 'magenta', 'cyan',
'gray'),
textures=('solid', ),
rotation=True,
size_range=(0.1, 0.25),
distortion_range=(2.0, 3.0),
shade_range=0.4,
collision_tolerance=0.25,
collision_shade_difference=0.5,
boundary_tolerance=None,
entity_counts=(1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15),
train_entity_counts=None,
validation_entity_counts=None,
test_entity_counts=None,
validation_count_rate=0.5,
test_count_rate=0.5,
validation_combinations=None,
test_combinations=None,
validation_space_rate_range=(0.0, 1.0),
test_space_rate_range=(0.0, 1.0),
validation_combination_rate=0.5,
test_combination_rate=0.5,
max_provoke_collision_rate=0.33,
reinforcement_range=(1, 3),
generators=None,
captioners=None,
connectives=None,
caption_size=28,
vocabulary=('.', 'a', 'above', 'all', 'an', 'and', 'are', 'as',
'at', 'behind', 'below', 'bigger', 'blue', 'but',
'circle', 'circles', 'closer', 'color', 'cross',
'crosses', 'cyan', 'darker', 'different', 'eight',
'either', 'ellipse', 'ellipses', 'exactly', 'farther',
'few', 'five', 'four', 'from', 'front', 'gray',
'green', 'half', 'if', 'in', 'is', 'least', 'left',
'less', 'lighter', 'magenta', 'many', 'more', 'most',
'no', 'none', 'not', 'of', 'one', 'only', 'or',
'pentagon', 'pentagons', 'quarter', 'quarters',
'rectangle', 'rectangles', 'red', 'right', 'same',
'semicircle', 'semicircles', 'seven', 'shape',
'shapes', 'six', 'smaller', 'square', 'squares',
'than', 'the', 'there', 'third', 'thirds', 'three',
'to', 'triangle', 'triangles', 'twice', 'two',
'yellow', 'zero'),
correct_ratio=0.5,
train_correct_ratio=None,
validation_correct_ratio=None,
test_correct_ratio=None,
worlds_per_instance=1,
captions_per_instance=1,
pixel_noise_stddev=0.0,
caption_realizer='dmrs',
language=None):
generator_list = list()
if generators is None or 'random' in generators:
random_generator = RandomAttributesGenerator(
world_size=world_size,
world_color=world_color,
shapes=shapes,
colors=colors,
textures=textures,
rotation=rotation,
size_range=size_range,
distortion_range=distortion_range,
shade_range=shade_range,
collision_tolerance=collision_tolerance,
collision_shade_difference=collision_shade_difference,
boundary_tolerance=boundary_tolerance,
entity_counts=entity_counts,
train_entity_counts=train_entity_counts,
validation_entity_counts=validation_entity_counts,
validation_count_rate=validation_count_rate,
test_entity_counts=test_entity_counts,
test_count_rate=test_count_rate,
validation_combinations=validation_combinations,
validation_space_rate_range=validation_space_rate_range,
validation_combination_rate=validation_combination_rate,
test_combinations=test_combinations,
test_space_rate_range=test_space_rate_range,
test_combination_rate=test_combination_rate,
max_provoke_collision_rate=max_provoke_collision_rate)
generator_list.append(random_generator)
if generators is None or 'reinforced' in generators:
reinforced_generator = ReinforcedAttributesGenerator(
world_size=world_size,
world_color=world_color,
shapes=shapes,
colors=colors,
textures=textures,
rotation=rotation,
size_range=size_range,
distortion_range=distortion_range,
shade_range=shade_range,
collision_tolerance=collision_tolerance,
collision_shade_difference=collision_shade_difference,
boundary_tolerance=boundary_tolerance,
entity_counts=entity_counts,
train_entity_counts=train_entity_counts,
validation_entity_counts=validation_entity_counts,
validation_count_rate=validation_count_rate,
test_entity_counts=test_entity_counts,
test_count_rate=test_count_rate,
validation_combinations=validation_combinations,
validation_space_rate_range=validation_space_rate_range,
validation_combination_rate=validation_combination_rate,
test_combinations=test_combinations,
test_space_rate_range=test_space_rate_range,
test_combination_rate=test_combination_rate,
max_provoke_collision_rate=max_provoke_collision_rate,
reinforcement_range=reinforcement_range)
generator_list.append(reinforced_generator)
world_generator = GeneratorMixer(generators=generator_list)
restrictor_captioner = CaptionerMixer(
captioners=(EmptyTypeCaptioner(),
RegularTypeCaptioner(hypernym_rate=1.0)))
body_captioner = AttributeTypeRelationCaptioner(
attribute_type_captioner=CaptionerMixer(
captioners=(RegularAttributeCaptioner(),
RegularTypeCaptioner(hypernym_rate=0.0))))
captioner_list = list()
if captioners is None or 'existential' in captioners:
existential_captioner = CaptionerMixer(
captioners=(RegularTypeCaptioner(),
ExistentialCaptioner(
restrictor_captioner=restrictor_captioner,
body_captioner=body_captioner)),
distribution=(1, 2))
captioner_list.append(existential_captioner)
if captioners is None or 'relational' in captioners:
relational_captioner = ExistentialCaptioner(
restrictor_captioner=RegularTypeCaptioner(),
body_captioner=RelationCaptioner(
reference_captioner=RegularTypeCaptioner(),
comparison_captioner=RegularTypeCaptioner()))
captioner_list.append(relational_captioner)
if captioners is None or 'quantification' in captioners:
quantification_captioner = QuantifierCaptioner(
restrictor_captioner=restrictor_captioner,
body_captioner=body_captioner)
captioner_list.append(quantification_captioner)
captioner = CaptionerMixer(captioners=captioner_list)
captioner_list = list()
if connectives is None or 'conjunction' in connectives:
captioner_list.append(ConjunctionCaptioner(captioner=captioner))
if connectives is None or 'disjunction' in connectives:
captioner_list.append(DisjunctionCaptioner(captioner=captioner))
if connectives is None or 'implication' in connectives:
captioner_list.append(ImplicationCaptioner(captioner=captioner))
if connectives is None or 'equivalence' in connectives:
captioner_list.append(EquivalenceCaptioner(captioner=captioner))
world_captioner = CaptionerMixer(captioners=captioner_list)
super(LogicalDataset,
self).__init__(world_generator=world_generator,
world_captioner=world_captioner,
caption_size=caption_size,
vocabulary=vocabulary,
correct_ratio=correct_ratio,
train_correct_ratio=train_correct_ratio,
validation_correct_ratio=validation_correct_ratio,
test_correct_ratio=test_correct_ratio,
worlds_per_instance=worlds_per_instance,
captions_per_instance=captions_per_instance,
pixel_noise_stddev=pixel_noise_stddev,
caption_realizer=caption_realizer,
language=language)
def __init__(self,
world_size=64,
world_colors=('black', ),
shapes=('square', 'rectangle', 'triangle', 'pentagon',
'cross', 'circle', 'semicircle', 'ellipse'),
colors=('red', 'green', 'blue', 'yellow', 'magenta', 'cyan',
'gray'),
textures=('solid', ),
rotation=True,
size_range=(0.1, 0.25),
distortion_range=(2.0, 3.0),
shade_range=0.4,
collision_tolerance=0.25,
collision_shade_difference=0.5,
boundary_tolerance=None,
entity_counts=(3, 4, 5, 6, 7, 8, 9, 10),
train_entity_counts=None,
validation_entity_counts=None,
test_entity_counts=None,
validation_count_rate=0.5,
test_count_rate=0.5,
validation_combinations=None,
test_combinations=None,
validation_space_rate_range=(0.0, 1.0),
test_space_rate_range=(0.0, 1.0),
validation_combination_rate=0.5,
test_combination_rate=0.5,
max_provoke_collision_rate=0.33,
relations=None,
negation=True,
existential_incorrect_distribution=(1, 1),
relation_incorrect_distribution=(2, 1, 1),
type_existing_attribute_rate=1.0,
type_incorrect_distribution=(1, 1, 1, 1),
caption_size=15,
vocabulary=('.', 'a', 'above', 'an', 'as', 'behind', 'below',
'besides', 'bigger', 'blue', 'circle', 'closer',
'color', 'cross', 'cyan', 'darker', 'different',
'does', 'ellipse', 'exist', 'exists', 'farther',
'from', 'front', 'gray', 'green', 'in', 'is',
'left', 'lighter', 'magenta', 'not', 'of',
'pentagon', 'rectangle', 'red', 'right', 'same',
'semicircle', 'shape', 'smaller', 'square',
'than', 'the', 'to', 'triangle', 'yellow'),
correct_ratio=0.5,
train_correct_ratio=None,
validation_correct_ratio=None,
test_correct_ratio=None,
worlds_per_instance=1,
captions_per_instance=1,
pixel_noise_stddev=None,
caption_realizer='dmrs',
language=None):
world_generator = ReinforcedAttributesGenerator(
world_size=world_size,
world_colors=world_colors,
shapes=shapes,
colors=colors,
textures=textures,
rotation=rotation,
size_range=size_range,
distortion_range=distortion_range,
shade_range=shade_range,
collision_tolerance=collision_tolerance,
collision_shade_difference=collision_shade_difference,
boundary_tolerance=boundary_tolerance,
entity_counts=entity_counts,
train_entity_counts=train_entity_counts,
validation_entity_counts=validation_entity_counts,
validation_count_rate=validation_count_rate,
test_entity_counts=test_entity_counts,
test_count_rate=test_count_rate,
validation_combinations=validation_combinations,
validation_space_rate_range=validation_space_rate_range,
validation_combination_rate=validation_combination_rate,
test_combinations=test_combinations,
test_space_rate_range=test_space_rate_range,
test_combination_rate=test_combination_rate,
max_provoke_collision_rate=max_provoke_collision_rate,
reinforcement_range=(1, 1))
relation_captioner = RelationCaptioner(
reference_captioner=RegularTypeCaptioner(
existing_attribute_rate=type_existing_attribute_rate,
incorrect_distribution=type_incorrect_distribution),
comparison_captioner=UniqueTypeCaptioner(),
relations=relations,
incorrect_distribution=relation_incorrect_distribution)
if negation:
relation_captioner = NegationRelationCaptioner(
relation_captioner=relation_captioner)
world_captioner = ExistentialCaptioner(
restrictor_captioner=RegularTypeCaptioner(
existing_attribute_rate=type_existing_attribute_rate,
incorrect_distribution=type_incorrect_distribution),
body_captioner=relation_captioner,
incorrect_distribution=existential_incorrect_distribution)
super(RelationalDataset,
self).__init__(world_generator=world_generator,
world_captioner=world_captioner,
caption_size=caption_size,
vocabulary=vocabulary,
correct_ratio=correct_ratio,
train_correct_ratio=train_correct_ratio,
validation_correct_ratio=validation_correct_ratio,
test_correct_ratio=test_correct_ratio,
worlds_per_instance=worlds_per_instance,
captions_per_instance=captions_per_instance,
pixel_noise_stddev=pixel_noise_stddev,
caption_realizer=caption_realizer,
language=language)
def __init__(
self,
entity_counts=(5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15),
train_entity_counts=(5, 6, 7, 8, 9, 10, 11, 12, 14),
validation_entity_counts=(13,),
test_entity_counts=(15,),
validation_combinations=(('square', 'red', 'solid'), ('triangle', 'green', 'solid'), ('circle', 'blue', 'solid')),
test_combinations=(('rectangle', 'yellow', 'solid'), ('cross', 'magenta', 'solid'), ('ellipse', 'cyan', 'solid')),
caption_size=19,
vocabulary=('.', 'a', 'above', 'all', 'an', 'are', 'as', 'at', 'behind', 'below', 'bigger', 'biggest', 'blue', 'both', 'but', 'circle', 'circles', 'closer', 'closest', 'cross', 'crosses', 'cyan', 'darker', 'darkest', 'eight', 'ellipse', 'ellipses', 'exactly', 'farther', 'farthest', 'five', 'four', 'from', 'front', 'gray', 'green', 'half', 'in', 'is', 'least', 'left', 'leftmost', 'less', 'lighter', 'lightest', 'lowermost', 'magenta', 'many', 'more', 'most', 'not', 'of', 'one', 'pentagon', 'pentagons', 'rectangle', 'rectangles', 'red', 'right', 'rightmost', 'semicircle', 'semicircles', 'seven', 'shape', 'shapes', 'six', 'smaller', 'smallest', 'square', 'squares', 'than', 'the', 'three', 'to', 'topmost', 'triangle', 'triangles', 'twice', 'two', 'yellow', 'zero'),
language=None
):
world_generator = ReinforcedAttributesGenerator(
reinforcement_range=(1, 3),
entity_counts=entity_counts,
train_entity_counts=train_entity_counts,
validation_entity_counts=validation_entity_counts,
test_entity_counts=test_entity_counts,
validation_combinations=validation_combinations,
test_combinations=test_combinations
)
body_captioner = CaptionerMixer(
captioners=(
AttributeTypeRelationCaptioner(
attribute_type_captioner=CaptionerMixer(
captioners=(
RegularAttributeCaptioner(),
RegularTypeCaptioner()
)
)
),
RelationCaptioner(
reference_captioner=RegularTypeCaptioner(),
comparison_captioner=RegularTypeCaptioner()
)
),
distribution=[1, 2]
)
quantifier_captioner = QuantifierCaptioner(
restrictor_captioner=RegularTypeCaptioner(
hypernym_rate=1.0,
logical_tautology_rate=1.0
),
body_captioner=body_captioner,
quantifiers=('count',)
)
number_bound_captioner = NumberBoundCaptioner(
quantifier_captioner=quantifier_captioner
)
comparative_quantifier_captioner = ComparativeQuantifierCaptioner(
restrictor_captioner=RegularTypeCaptioner(
hypernym_rate=1.0
),
comparison_captioner=RegularTypeCaptioner(
hypernym_rate=1.0
),
body_captioner=body_captioner
)
world_captioner = CaptionerMixer(
captioners=(quantifier_captioner, number_bound_captioner, comparative_quantifier_captioner),
distribution=[1, 1, 1]
)
super(QuantificationCount, self).__init__(
world_generator=world_generator,
world_captioner=world_captioner,
caption_size=caption_size,
vocabulary=vocabulary,
language=language
)
def __init__(
self,
world_size=64,
world_color='black',
shapes=('square', 'rectangle', 'triangle', 'pentagon', 'cross',
'circle', 'semicircle', 'ellipse'),
colors=('red', 'green', 'blue', 'yellow', 'magenta', 'cyan',
'gray'),
textures=('solid', ),
rotation=True,
size_range=(0.1, 0.25),
distortion_range=(2.0, 3.0),
shade_range=0.4,
collision_tolerance=0.25,
collision_shade_difference=0.5,
boundary_tolerance=0.25,
entity_counts=(1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15),
train_entity_counts=(1, 2, 4, 6, 7, 9, 11, 12, 14),
validation_entity_counts=(3, 8, 13),
validation_count_rate=0.5,
test_entity_counts=(5, 10, 15),
test_count_rate=0.5,
validation_combinations=(('square', 'red',
'solid'), ('triangle', 'green', 'solid'),
('circle', 'blue', 'solid')),
validation_space_rate_range=(0.0, 1.0),
validation_combination_rate=0.5,
test_combinations=(('rectangle', 'yellow',
'solid'), ('cross', 'magenta', 'solid'),
('ellipse', 'cyan', 'solid')),
test_space_rate_range=(0.0, 1.0),
test_combination_rate=0.5,
max_provoke_collision_rate=0.33,
reinforcement_range=(1, 3),
caption_size=28,
vocabulary=('.', 'a', 'above', 'all', 'an', 'and', 'are', 'as',
'at', 'behind', 'below', 'bigger', 'blue', 'but',
'circle', 'circles', 'closer', 'cross', 'crosses',
'cyan', 'darker', 'eight', 'either', 'ellipse',
'ellipses', 'exactly', 'farther', 'few', 'five',
'four', 'from', 'front', 'gray', 'green', 'half', 'in',
'is', 'least', 'left', 'less', 'lighter', 'magenta',
'many', 'more', 'most', 'no', 'none', 'not', 'of',
'one', 'or', 'pentagon', 'pentagons', 'quarter',
'quarters', 'rectangle', 'rectangles', 'red', 'right',
'semicircle', 'semicircles', 'seven', 'shape',
'shapes', 'six', 'smaller', 'square', 'squares',
'than', 'the', 'there', 'third', 'thirds', 'three',
'to', 'triangle', 'triangles', 'twice', 'two',
'yellow', 'zero'),
correct_ratio=0.5,
train_correct_ratio=None,
validation_correct_ratio=None,
test_correct_ratio=None,
worlds_per_instance=1,
captions_per_instance=1,
pixel_noise_stddev=0.0,
caption_realizer='dmrs',
language=None):
random_generator = RandomAttributesGenerator(
world_size=world_size,
world_color=world_color,
shapes=shapes,
colors=colors,
textures=textures,
rotation=rotation,
size_range=size_range,
distortion_range=distortion_range,
shade_range=shade_range,
collision_tolerance=collision_tolerance,
collision_shade_difference=collision_shade_difference,
boundary_tolerance=boundary_tolerance,
entity_counts=entity_counts,
train_entity_counts=train_entity_counts,
validation_entity_counts=validation_entity_counts,
test_entity_counts=test_entity_counts,
validation_combinations=validation_combinations,
test_combinations=test_combinations,
max_provoke_collision_rate=max_provoke_collision_rate)
reinforced_attributes_generator = ReinforcedAttributesGenerator(
world_size=world_size,
world_color=world_color,
shapes=shapes,
colors=colors,
textures=textures,
rotation=rotation,
size_range=size_range,
distortion_range=distortion_range,
shade_range=shade_range,
collision_tolerance=collision_tolerance,
collision_shade_difference=collision_shade_difference,
boundary_tolerance=boundary_tolerance,
entity_counts=entity_counts,
train_entity_counts=train_entity_counts,
validation_entity_counts=validation_entity_counts,
test_entity_counts=test_entity_counts,
validation_combinations=validation_combinations,
test_combinations=test_combinations,
max_provoke_collision_rate=max_provoke_collision_rate,
reinforcement_range=reinforcement_range)
world_generator = GeneratorMixer(
generators=(random_generator, reinforced_attributes_generator))
body_captioner = AttributeTypeRelationCaptioner(
attribute_type_captioner=CaptionerMixer(
captioners=(RegularAttributeCaptioner(),
RegularTypeCaptioner())))
existential_captioner = CaptionerMixer(captioners=(
RegularTypeCaptioner(),
ExistentialCaptioner(restrictor_captioner=RegularTypeCaptioner(
hypernym_rate=1.0, logical_tautology_rate=1.0),
body_captioner=body_captioner)))
relation_captioner = ExistentialCaptioner(
restrictor_captioner=RegularTypeCaptioner(),
body_captioner=RelationCaptioner(
reference_captioner=RegularTypeCaptioner(),
comparison_captioner=RegularTypeCaptioner()))
quantifier_captioner = QuantifierCaptioner(
restrictor_captioner=RegularTypeCaptioner(
hypernym_rate=1.0, logical_tautology_rate=1.0),
body_captioner=body_captioner)
number_bound_captioner = NumberBoundCaptioner(
quantifier_captioner=quantifier_captioner)
comparative_quantifier_captioner = ComparativeQuantifierCaptioner(
restrictor_captioner=RegularTypeCaptioner(hypernym_rate=1.0),
comparison_captioner=RegularTypeCaptioner(hypernym_rate=1.0),
body_captioner=body_captioner)
quantification_captioner = CaptionerMixer(
captioners=(quantifier_captioner, number_bound_captioner,
comparative_quantifier_captioner),
distribution=[2, 2, 1])
world_captioner = CaptionerMixer(captioners=(
ConjunctionCaptioner(captioners=(existential_captioner,
relation_captioner,
quantification_captioner)),
DisjunctionCaptioner(captioners=(existential_captioner,
relation_captioner,
quantification_captioner))))
super(Combination,
self).__init__(world_generator=world_generator,
world_captioner=world_captioner,
caption_size=caption_size,
vocabulary=vocabulary,
correct_ratio=correct_ratio,
train_correct_ratio=train_correct_ratio,
validation_correct_ratio=validation_correct_ratio,
test_correct_ratio=test_correct_ratio,
worlds_per_instance=worlds_per_instance,
captions_per_instance=captions_per_instance,
pixel_noise_stddev=pixel_noise_stddev,
caption_realizer=caption_realizer,
language=language)
def __init__(
self,
world_size=64,
world_colors=('black', ),
shapes=('square', 'rectangle', 'triangle', 'pentagon', 'cross',
'circle', 'semicircle', 'ellipse'),
colors=('red', 'green', 'blue', 'yellow', 'magenta', 'cyan',
'gray'),
textures=('solid', ),
rotation=True,
size_range=(0.1, 0.25),
distortion_range=(2.0, 3.0),
shade_range=0.4,
collision_tolerance=0.25,
collision_shade_difference=0.5,
boundary_tolerance=None,
entity_counts=(4, 5, 6, 7, 8, 9, 10),
train_entity_counts=None,
validation_entity_counts=None,
test_entity_counts=None,
validation_count_rate=0.5,
test_count_rate=0.5,
validation_combinations=None,
test_combinations=None,
validation_space_rate_range=(0.0, 1.0),
test_space_rate_range=(0.0, 1.0),
validation_combination_rate=0.5,
test_combination_rate=0.5,
max_provoke_collision_rate=0.33,
allow_empty_scope=True,
selectors=None,
caption_size=14,
vocabulary=('.', 'a', 'an', 'are', 'bigger', 'biggest', 'blue',
'circle', 'circles', 'closer', 'closest', 'cross',
'crosses', 'cyan', 'darker', 'darkest', 'ellipse',
'ellipses', 'farther', 'farthest', 'five', 'four',
'from', 'gray', 'green', 'is', 'left', 'leftmost',
'lighter', 'lightest', 'lower', 'lowermost', 'magenta',
'one', 'pentagon', 'pentagons', 'rectangle',
'rectangles', 'red', 'right', 'rightmost',
'semicircle', 'semicircles', 'shape', 'shapes',
'smaller', 'smallest', 'square', 'squares', 'the',
'three', 'to', 'triangle', 'triangles', 'two', 'upper',
'uppermost', 'yellow'),
correct_ratio=0.5,
train_correct_ratio=None,
validation_correct_ratio=None,
test_correct_ratio=None,
worlds_per_instance=1,
captions_per_instance=1,
pixel_noise_stddev=None,
caption_realizer='dmrs',
language=None):
world_generator = ReinforcedAttributesGenerator(
world_size=world_size,
world_colors=world_colors,
shapes=shapes,
colors=colors,
textures=textures,
rotation=rotation,
size_range=size_range,
distortion_range=distortion_range,
shade_range=shade_range,
collision_tolerance=collision_tolerance,
collision_shade_difference=collision_shade_difference,
boundary_tolerance=boundary_tolerance,
entity_counts=entity_counts,
train_entity_counts=train_entity_counts,
validation_entity_counts=validation_entity_counts,
validation_count_rate=validation_count_rate,
test_entity_counts=test_entity_counts,
test_count_rate=test_count_rate,
validation_combinations=validation_combinations,
validation_space_rate_range=validation_space_rate_range,
validation_combination_rate=validation_combination_rate,
test_combinations=test_combinations,
test_space_rate_range=test_space_rate_range,
test_combination_rate=test_combination_rate,
max_provoke_collision_rate=max_provoke_collision_rate,
reinforcement_range=(1, 1))
scope_captioners = [RegularTypeCaptioner(hypernym_rate=1.0)]
if allow_empty_scope:
scope_captioners.append(EmptyTypeCaptioner())
world_captioner = ExistentialCaptioner(
restrictor_captioner=SelectorCaptioner(
scope_captioner=CaptionerMixer(captioners=scope_captioners),
comparison_captioner=UniqueTypeCaptioner(),
selectors=selectors),
body_captioner=AttributeTypeRelationCaptioner(
attribute_type_captioner=CaptionerMixer(
captioners=(RegularAttributeCaptioner(),
RegularTypeCaptioner(hypernym_rate=0.0)))))
super(SelectionDataset,
self).__init__(world_generator=world_generator,
world_captioner=world_captioner,
caption_size=caption_size,
vocabulary=vocabulary,
correct_ratio=correct_ratio,
train_correct_ratio=train_correct_ratio,
validation_correct_ratio=validation_correct_ratio,
test_correct_ratio=test_correct_ratio,
worlds_per_instance=worlds_per_instance,
captions_per_instance=captions_per_instance,
pixel_noise_stddev=pixel_noise_stddev,
caption_realizer=caption_realizer,
language=language)