def formula_to_serialized_entities(match_parse, formula, tree,
sentence_number):
offset = match_parse.graph_parse.core_parse.image_segment_parse.diagram_image_segment.offset
grounded_formula = ground_formulas(match_parse, [formula])[0]
entities = []
zipped_formula = grounded_formula.zip(tree)
for zipped_node in zipped_formula:
formula_node, tree_node = zipped_node.nodes
if not isinstance(formula_node, FormulaNode):
continue
if len(formula_node.children) == 1 and not issubtype(
formula_node.return_type, 'entity'):
formula_node = formula_node.children[0]
if issubtype(formula_node.return_type, 'entity'):
coords = match_parse.graph_parse.core_parse.evaluate(formula_node)
if coords is not None:
coords = offset_coords(coords, formula_node.return_type,
offset)
content = tree_node.content.serialized()
content['signature']['return_type'] = formula_node.return_type
entity = {
"content": content,
"coords": serialize_entity(coords),
"sentence_number": sentence_number
}
entities.append(entity)
return entities
def val_func(parent_tag_rule, a_tag_rule, b_tag_rule):
valence = parent_tag_rule.signature.valence
if valence == 2:
a = issubtype(a_tag_rule.signature.return_type, parent_tag_rule.signature.arg_types[0])
b = issubtype(b_tag_rule.signature.return_type, parent_tag_rule.signature.arg_types[1])
return a and b
else:
return False
def val_func(p, a, b):
if p.signature.id != "Is":
return False
if not issubtype(a.signature.return_type, b.signature.return_type) and \
not issubtype(b.signature.return_type, a.signature.return_type):
return False
if not (issubtype(a.signature.return_type, 'number') or issubtype(a.signature.return_type, 'entity')):
return False
return BinaryRule.val_func(p, a, b)
def val_func(parent_tag_rule, a_tag_rule, b_tag_rule):
valence = parent_tag_rule.signature.valence
if valence == 2:
a = issubtype(a_tag_rule.signature.return_type,
parent_tag_rule.signature.arg_types[0])
b = issubtype(b_tag_rule.signature.return_type,
parent_tag_rule.signature.arg_types[1])
return a and b
else:
return False
def val_func(parent_tag_rule, child_tag_rule):
valence = parent_tag_rule.signature.valence
if valence == 0:
return False
elif valence == 1:
return issubtype(child_tag_rule.signature.return_type, parent_tag_rule.signature.arg_types[0])
elif valence == 2:
c1 = issubtype(child_tag_rule.signature.return_type, parent_tag_rule.signature.arg_types[0])
c2 = issubtype(parent_tag_rule.signature.arg_types[1], 'entity')
return c1 and c2
raise Exception()
def val_func(parent_tag_rule, child_tag_rule):
valence = parent_tag_rule.signature.valence
if valence == 0:
return False
elif valence == 1:
return issubtype(child_tag_rule.signature.return_type,
parent_tag_rule.signature.arg_types[0])
elif valence == 2:
c1 = issubtype(child_tag_rule.signature.return_type,
parent_tag_rule.signature.arg_types[0])
c2 = issubtype(parent_tag_rule.signature.arg_types[1], 'entity')
return c1 and c2
raise Exception()
def AreaOf(twod):
name = twod.__class__.__name__
assert issubtype(name, 'twod')
if name == "circle":
center, radius = twod
area = np.pi * radius**2
elif issubtype(name, 'polygon'):
# http://mathworld.wolfram.com/PolygonArea.html
area = area_of_polygon(twod)
elif name == 'arc':
circle, a, b = twod
else:
raise Exception()
return area
def Is(a, b):
if is_number(a) or is_number(b):
return Equals(a, b)
a_name = a.__class__.__name__
b_name = b.__class__.__name__
if issubtype(a_name, 'polygon') or issubtype(a_name, 'line'):
truth = set(a) == set(b)
else:
truth = a == b
if truth:
return TruthValue(0)
else:
return TruthValue(np.inf)
def semantic_tree_to_serialized_entities(match_parse, semantic_tree,
sentence_number, value_expr_formulas):
offset = match_parse.graph_parse.core_parse.image_segment_parse.diagram_image_segment.offset
formula = semantic_tree.to_formula()
entities = []
grounded_formula = ground_formulas(match_parse, [formula],
value_expr_formulas)[0]
zipped_formula = grounded_formula.zip(semantic_tree)
for zipped_node in zipped_formula:
formula_node, tree_node = zipped_node.nodes
if isinstance(formula_node, FormulaNode) and issubtype(
formula_node.return_type, 'entity'):
coords = match_parse.graph_parse.core_parse.evaluate(formula_node)
if coords is not None:
coords = offset_coords(coords,
tree_node.content.signature.return_type,
offset)
entity = {
"content": tree_node.content.serialized(),
"coords": serialize_entity(coords),
"sentence_number": sentence_number
}
entities.append(entity)
return entities
def parse_match_formulas(match_parse):
assert isinstance(match_parse, MatchParse)
match_atoms = []
for label, terms in match_parse.match_dict.iteritems():
for term in terms:
assert isinstance(term, FormulaNode)
if issubtype(term.return_type, 'entity'):
if term.signature.id == "Angle":
res = FormulaNode(signatures['Ge'], [FormulaNode(signatures['Pi'], []), FormulaNode(signatures['MeasureOf'], [term])])
match_atoms.append(res)
continue
# FIXME : to be obtained by tag model
left_term = prefix_to_formula(expression_parser.parse_prefix(label))
"""
if is_number(label):
left_term = FormulaNode(FunctionSignature(label, "number", []), [])
else:
vs = VariableSignature(label, 'number')
left_term = FormulaNode(vs, [])
"""
atom = FormulaNode(signatures['Equals'], [left_term, term])
match_atoms.append(atom)
if term.signature.id == "Div":
# TODO : this should be only constrained if the observed angle is < 180
# TODO : In fact, the labeling should be reorganized. (x --> x*\degree)
res = FormulaNode(signatures['Ge'], [180, left_term])
match_atoms.append(res)
return match_atoms
def offset_coords(coords, type_, offset):
coords = list(coords)
if issubtype(type_, 'point'):
coords = offset_point(coords, offset)
elif issubtype(type_, "line"):
coords[0] = offset_point(coords[0], offset)
coords[1] = offset_point(coords[1], offset)
elif issubtype(type_, 'circle'):
coords[0] = offset_point(coords[0], offset)
elif issubtype(type_, 'arc') or issubtype(type_, 'sector'):
coords[0][0] = offset_point(coords[0][0], offset)
coords[1] = offset_point(coords[1], offset)
coords[2] = offset_point(coords[2], offset)
else:
coords = [offset_point(point, offset) for point in coords]
return coords
def offset_coords(coords, type_, offset):
coords = list(coords)
if issubtype(type_, 'point'):
coords = offset_point(coords, offset)
elif issubtype(type_, "line"):
coords[0] = offset_point(coords[0], offset)
coords[1] = offset_point(coords[1], offset)
elif issubtype(type_, 'circle'):
coords[0] = offset_point(coords[0], offset)
elif issubtype(type_, 'arc') or issubtype(type_, 'sector'):
coords[0][0] = offset_point(coords[0][0], offset)
coords[1] = offset_point(coords[1], offset)
coords[2] = offset_point(coords[2], offset)
else:
coords = [offset_point(point, offset) for point in coords]
return coords
def is_valid_relation(parent_signature, child_signature, index):
parent_type = parent_signature.arg_types[index]
child_type = child_signature.return_type
if parent_type[0] != "*" and child_type == "*":
return False
if parent_type[0] == "*":
parent_type = parent_type[1:]
if child_type[0] == "*":
child_type = child_type[1:]
return issubtype(child_type, parent_type)
def Tangent(line, twod):
name = twod.__class__.__name__
if name == "circle":
d = perpendicular_distance_between_line_and_point(line, twod.center)
return Equals(d, twod.radius)
elif issubtype(name, 'polygon'):
out = reduce(operator.__or__,
(PointLiesOnLine(point, line) for point in twod), False)
return out
raise Exception()
def is_valid_relation(parent_signature, child_signature, index):
parent_type = parent_signature.arg_types[index]
child_type = child_signature.return_type
if parent_type[0] != "*" and child_type == "*":
return False
if parent_type[0] == "*":
parent_type = parent_type[1:]
if child_type[0] == "*":
child_type = child_type[1:]
return issubtype(child_type, parent_type)
def formula_to_serialized_entities(match_parse, formula, tree, sentence_number):
offset = match_parse.graph_parse.core_parse.image_segment_parse.diagram_image_segment.offset
grounded_formula = ground_formulas(match_parse, [formula])[0]
entities = []
zipped_formula = grounded_formula.zip(tree)
for zipped_node in zipped_formula:
formula_node, tree_node = zipped_node.nodes
if not isinstance(formula_node, FormulaNode):
continue
if len(formula_node.children) == 1 and not issubtype(formula_node.return_type, 'entity'):
formula_node = formula_node.children[0]
if issubtype(formula_node.return_type, 'entity'):
coords = match_parse.graph_parse.core_parse.evaluate(formula_node)
if coords is not None:
coords = offset_coords(coords, formula_node.return_type, offset)
content = tree_node.content.serialized()
content['signature']['return_type'] = formula_node.return_type
entity = {"content": content, "coords": serialize_entity(coords),
"sentence_number": sentence_number}
entities.append(entity)
return entities
def get_semantic_trees_by_type(self, return_type, terminator=None):
roots = [
node for node in self.node_dict.values()
if issubtype(node.tag_rule.signature.return_type, return_type)
and node.tag_rule.signature.return_type != 'ground'
]
semantic_trees = set(
itertools.chain(*[
self.get_semantic_trees_by_node(root, terminator)
for root in roots
]))
return semantic_trees
def IsCenterOf(point, twod):
name = twod.__class__.__name__
if name == 'circle':
return Equals(point[0], twod.center[0]) & Equals(
point[1], twod.center[1])
elif issubtype(name, 'polygon'):
distances = [distance_between_points(point, each) for each in twod]
reg = IsRegular(twod)
out = reduce(operator.__and__,
(Equals(distances[index - 1], distance)
for index, distance in enumerate(distances)), True)
return reg & out
else:
raise Exception()
def formula_to_semantic_tree(formula, syntax_parse, span):
"""
Create dummy semantic tree where each tag's syntax Parse and span is given
:param formula:
:param index:
:return:
"""
assert isinstance(formula, FormulaNode)
if issubtype(formula.signature.return_type, 'entity'):
new_sig = VariableSignature(formula.signature.id, formula.signature.return_type, name='temp')
tag_rule = TagRule(syntax_parse, span, new_sig)
return SemanticTreeNode(tag_rule, [])
tag_rule = TagRule(syntax_parse, span, formula.signature)
children = [formula_to_semantic_tree(child, syntax_parse, span) for child in formula.children]
semantic_tree = SemanticTreeNode(tag_rule, children)
return semantic_tree
def semantic_tree_to_serialized_entities(match_parse, semantic_tree, sentence_number, value_expr_formulas):
offset = match_parse.graph_parse.core_parse.image_segment_parse.diagram_image_segment.offset
formula = semantic_tree.to_formula()
entities = []
grounded_formula = ground_formulas(match_parse, [formula], value_expr_formulas)[0]
zipped_formula = grounded_formula.zip(semantic_tree)
for zipped_node in zipped_formula:
formula_node, tree_node = zipped_node.nodes
if isinstance(formula_node, FormulaNode) and issubtype(formula_node.return_type, 'entity'):
coords = match_parse.graph_parse.core_parse.evaluate(formula_node)
if coords is not None:
coords = offset_coords(coords, tree_node.content.signature.return_type, offset)
entity = {"content": tree_node.content.serialized(), "coords": serialize_entity(coords),
"sentence_number": sentence_number}
entities.append(entity)
return entities
def map(self, tr):
assert isinstance(tr, TagRule)
sp = tr.syntax_parse
out = []
for ref_rt in self.return_type_set:
out.append(int(issubtype(tr.signature.return_type, ref_rt)))
for ref_pos in self.pos_set:
out.append(int(ref_pos == sp.get_pos_by_span(tr.span)))
nbrs = sp.get_neighbors(tr.span)
pairs = set((rel, sp.get_word(key)) for key, rel in nbrs.iteritems())
for key_rel in self.key_rels:
for nbr in self.key_nbrs:
pair = (key_rel, nbr)
out.append(int(pair in pairs))
return tuple(out)
def map(self, tr):
assert isinstance(tr, TagRule)
sp = tr.syntax_parse
out = []
for ref_rt in self.return_type_set:
out.append(int(issubtype(tr.signature.return_type, ref_rt)))
for ref_pos in self.pos_set:
out.append(int(ref_pos == sp.get_pos_by_span(tr.span)))
nbrs = sp.get_neighbors(tr.span)
pairs = set((rel, sp.get_word(key)) for key, rel in nbrs.iteritems())
for key_rel in self.key_rels:
for nbr in self.key_nbrs:
pair = (key_rel, nbr)
out.append(int(pair in pairs))
return tuple(out)
def formula_to_semantic_tree(formula, syntax_parse, span):
"""
Create dummy semantic tree where each tag's syntax Parse and span is given
:param formula:
:param index:
:return:
"""
assert isinstance(formula, FormulaNode)
if issubtype(formula.signature.return_type, 'entity'):
new_sig = VariableSignature(formula.signature.id,
formula.signature.return_type,
name='temp')
tag_rule = TagRule(syntax_parse, span, new_sig)
return SemanticTreeNode(tag_rule, [])
tag_rule = TagRule(syntax_parse, span, formula.signature)
children = [
formula_to_semantic_tree(child, syntax_parse, span)
for child in formula.children
]
semantic_tree = SemanticTreeNode(tag_rule, children)
return semantic_tree
def evaluate(formula, assignment):
if not isinstance(formula, Node):
return formula
if not formula.is_grounded(assignment.keys()):
return None
if isinstance(formula, SetNode):
if issubtype(formula.head.return_type, 'boolean'):
out = reduce(operator.__and__, (evaluate(child, assignment)
for child in formula.children),
True)
return out
return formula
if isinstance(formula.signature, VariableSignature):
return assignment[formula.signature.id]
elif is_number(formula.signature.id):
return float(formula.signature.id)
else:
evaluated_args = []
for arg in formula.children:
if isinstance(arg, FormulaNode):
evaluated_args.append(evaluate(arg, assignment))
elif isinstance(arg, SetNode):
evaluated_args.append(
SetNode([
evaluate(arg_arg, assignment)
for arg_arg in arg.children
]))
else:
evaluated_args.append(arg)
# FIXME : rather than try/catch, check type matching
try:
out = getattr(this, formula.signature.id)(*evaluated_args)
return out
except:
return TruthValue(np.inf)
def parse_match_formulas(match_parse):
assert isinstance(match_parse, MatchParse)
match_atoms = []
for label, terms in match_parse.match_dict.iteritems():
for term in terms:
assert isinstance(term, FormulaNode)
if issubtype(term.return_type, 'entity'):
if term.signature.id == "Angle":
res = FormulaNode(signatures['Ge'], [
FormulaNode(signatures['Pi'], []),
FormulaNode(signatures['MeasureOf'], [term])
])
match_atoms.append(res)
continue
# FIXME : to be obtained by tag model
left_term = prefix_to_formula(
expression_parser.parse_prefix(label))
"""
if is_number(label):
left_term = FormulaNode(FunctionSignature(label, "number", []), [])
else:
vs = VariableSignature(label, 'number')
left_term = FormulaNode(vs, [])
"""
atom = FormulaNode(signatures['Equals'], [left_term, term])
match_atoms.append(atom)
if term.signature.id == "Div":
# TODO : this should be only constrained if the observed angle is < 180
# TODO : In fact, the labeling should be reorganized. (x --> x*\degree)
res = FormulaNode(signatures['Ge'], [180, left_term])
match_atoms.append(res)
return match_atoms
def parse_match_from_known_labels(graph_parse, known_labels):
assert isinstance(graph_parse, GraphParse)
match_dict = {}
point_key_dict = {}
offset = graph_parse.image_segment_parse.diagram_image_segment.offset
for idx, d in enumerate(known_labels):
label = d['label']
x = d['x'] - offset[0]
y = d['y'] - offset[1]
label_point = instantiators['point'](x, y)
type_ = d['type']
arr = type_.split(' ')
if len(arr) > 1:
type_ = arr[-1]
# Find closest type_ instance's key in graph_parse
instances = get_all_instances(graph_parse, type_)
if len(instances) == 0:
logging.error("no instance found of type %s" % type_)
continue
if len(arr) > 1 and type_ == 'line' and arr[0] == 'length':
distances = [(key,
label_distance_to_line(label_point, instance, True))
for key, instance in instances.iteritems()]
elif type_ == 'line':
distances = [(key,
label_distance_to_line(label_point, instance, False))
for key, instance in instances.iteritems()]
elif type_ == 'point':
distances = [(key, label_distance_to_point(label_point, instance))
for key, instance in instances.iteritems()]
elif type_ == 'arc':
distances = [(key, label_distance_to_arc(label_point, instance))
for key, instance in instances.iteritems()]
elif type_ == 'angle':
# filter subangles
# instances = {key: value for key, value in instances.iteritems() if all(x == value or not is_subangle(x, value) for x in instances.values())}
distances = [(key, label_distance_to_angle(label_point, instance))
for key, instance in instances.iteritems()]
# Then use the key to get corresponding variable in general graph
# Wrap the general instance in function nod3. If there are extra prefixes, add these as well the formula
argmin_key = min(distances, key=lambda pair: pair[1])[0]
if type_ == 'line':
a_key, b_key = argmin_key
a_point = graph_parse.point_variables[a_key]
b_point = graph_parse.point_variables[b_key]
formula = FormulaNode(signatures['Line'], [a_point, b_point])
if len(arr) > 1 and arr[0] == 'length':
formula = FormulaNode(signatures['LengthOf'], [formula])
elif type_ == 'point':
formula = graph_parse.point_variables[argmin_key]
point_key_dict[label] = argmin_key
elif type_ == 'angle':
a_key, b_key, c_key = argmin_key
a_point = graph_parse.point_variables[a_key]
b_point = graph_parse.point_variables[b_key]
c_point = graph_parse.point_variables[c_key]
formula = FormulaNode(signatures['Angle'],
[a_point, b_point, c_point])
if len(arr) > 1 and arr[0] == 'angle':
formula = FormulaNode(signatures['MeasureOf'], [formula])
formula = FormulaNode(
signatures['Div'],
[formula, FormulaNode(signatures['Degree'], [])])
elif type_ == 'arc':
(center_key, radius_key), a_key, b_key = argmin_key
center_point = graph_parse.point_variables[center_key]
radius = graph_parse.radius_variables[center_key][radius_key]
circle = FormulaNode(signatures['Circle'], [center_point, radius])
a_point = graph_parse.point_variables[a_key]
b_point = graph_parse.point_variables[b_key]
formula = FormulaNode(signatures['Arc'],
[circle, a_point, b_point])
if label not in match_dict:
match_dict[label] = []
elif issubtype(formula.return_type, 'entity'):
raise Exception()
match_dict[label].append(formula)
match_parse = MatchParse(graph_parse, match_dict, point_key_dict)
return match_parse
def get_semantic_trees_by_type(self, return_type, terminator=None):
roots = [node for node in self.node_dict.values() if issubtype(node.tag_rule.signature.return_type, return_type)
and node.tag_rule.signature.return_type != 'ground']
semantic_trees = set(itertools.chain(*[self.get_semantic_trees_by_node(root, terminator) for root in roots]))
return semantic_trees
def parse_match_from_known_labels(graph_parse, known_labels):
assert isinstance(graph_parse, GraphParse)
match_dict = {}
point_key_dict = {}
offset = graph_parse.image_segment_parse.diagram_image_segment.offset
for idx, d in enumerate(known_labels):
label = d['label']
x = d['x'] - offset[0]
y = d['y'] - offset[1]
label_point = instantiators['point'](x, y)
type_ = d['type']
arr = type_.split(' ')
if len(arr) > 1:
type_ = arr[-1]
# Find closest type_ instance's key in graph_parse
instances = get_all_instances(graph_parse, type_)
if len(instances) == 0:
logging.error("no instance found of type %s" % type_)
continue
if len(arr) > 1 and type_ == 'line' and arr[0] == 'length':
distances = [(key, label_distance_to_line(label_point, instance, True)) for key, instance in instances.iteritems()]
elif type_ == 'line':
distances = [(key, label_distance_to_line(label_point, instance, False)) for key, instance in instances.iteritems()]
elif type_ == 'point':
distances = [(key, label_distance_to_point(label_point, instance)) for key, instance in instances.iteritems()]
elif type_ == 'arc':
distances = [(key, label_distance_to_arc(label_point, instance)) for key, instance in instances.iteritems()]
elif type_ == 'angle':
# filter subangles
# instances = {key: value for key, value in instances.iteritems() if all(x == value or not is_subangle(x, value) for x in instances.values())}
distances = [(key, label_distance_to_angle(label_point, instance)) for key, instance in instances.iteritems()]
# Then use the key to get corresponding variable in general graph
# Wrap the general instance in function nod3. If there are extra prefixes, add these as well the formula
argmin_key = min(distances, key=lambda pair: pair[1])[0]
if type_ == 'line':
a_key, b_key = argmin_key
a_point = graph_parse.point_variables[a_key]
b_point = graph_parse.point_variables[b_key]
formula = FormulaNode(signatures['Line'], [a_point, b_point])
if len(arr) > 1 and arr[0] == 'length':
formula = FormulaNode(signatures['LengthOf'], [formula])
elif type_ == 'point':
formula = graph_parse.point_variables[argmin_key]
point_key_dict[label] = argmin_key
elif type_ == 'angle':
a_key, b_key, c_key = argmin_key
a_point = graph_parse.point_variables[a_key]
b_point = graph_parse.point_variables[b_key]
c_point = graph_parse.point_variables[c_key]
formula = FormulaNode(signatures['Angle'], [a_point, b_point, c_point])
if len(arr) > 1 and arr[0] == 'angle':
formula = FormulaNode(signatures['MeasureOf'], [formula])
formula = FormulaNode(signatures['Div'], [formula, FormulaNode(signatures['Degree'], [])])
elif type_ == 'arc':
(center_key, radius_key), a_key, b_key = argmin_key
center_point = graph_parse.point_variables[center_key]
radius = graph_parse.radius_variables[center_key][radius_key]
circle = FormulaNode(signatures['Circle'], [center_point, radius])
a_point = graph_parse.point_variables[a_key]
b_point = graph_parse.point_variables[b_key]
formula = FormulaNode(signatures['Arc'], [circle, a_point, b_point])
if len(arr) > 0 and arr[0] == 'angle':
formula = FormulaNode(signatures['MeasureOf'], [formula])
formula = FormulaNode(signatures['Div'], [formula, FormulaNode(signatures['Degree'], [])])
if label not in match_dict:
match_dict[label] = []
elif issubtype(formula.return_type, 'entity'):
raise Exception()
match_dict[label].append(formula)
match_parse = MatchParse(graph_parse, match_dict, point_key_dict)
return match_parse
