def predict(self, a_rel, a_x):
"""Method for predicting sense of single relation.
Args:
a_rel (dict):
input relation to classify
a_x (np.array):
(submodels x class) array of input predictions
Returns:
str:
most probable sense of discourse relation
"""
ret = np.mean(a_x, axis=0)
return (np.argmax(ret), ret)
if is_explicit(a_rel):
return self.explicit.predict(a_x)
return self.implicit.predict(a_x)
def predict(self, a_rel, a_x):
"""Method for predicting sense of single relation.
Args:
a_rel (dict):
input relation to classify
a_x (np.array):
(submodels x class) array of input predictions
Returns:
str:
most probable sense of discourse relation
"""
ret = np.mean(a_x, axis=0)
return (np.argmax(ret), ret)
if is_explicit(a_rel):
return self.explicit.predict(a_x)
return self.implicit.predict(a_x)
def _divide_data(self, a_data):
"""Separate dataset into explicit and implicit instances.
Args:
a_data (2-tuple(dict, dict)):
list of gold relations and dict with parses
Returns:
(2-tuple(list, list)):
lists of explicit and implicit training instances
"""
if a_data is None:
return ((), ())
explicit_instances = []
implicit_instances = []
for x_i, irel, y_i in a_data:
if is_explicit(irel):
explicit_instances.append((x_i, y_i))
else:
implicit_instances.append((x_i, y_i))
return (explicit_instances, implicit_instances)
def _divide_data(self, a_data):
"""Separate dataset into explicit and implicit instances.
Args:
a_data (2-tuple(dict, dict)):
list of gold relations and dict with parses
Returns:
(2-tuple(list, list)):
lists of explicit and implicit training instances
"""
if a_data is None:
return ((), ())
explicit_instances = []
implicit_instances = []
for x_i, irel, y_i in a_data:
if is_explicit(irel):
explicit_instances.append((x_i, y_i))
else:
implicit_instances.append((x_i, y_i))
return (explicit_instances, implicit_instances)
def _divide_ds(self, a_ds):
"""Separate dataset into explicit and implicit instances.
Args:
a_ds (tuple):
list of gold relations and dict with parses
Returns:
tuple:
trainings set with explicit and implicit connectives
"""
if not a_ds:
return (([], {}), ([], {}))
explicit_instances = []
implicit_instances = []
for i, irel in enumerate(a_ds[0]):
if is_explicit(irel):
explicit_instances.append((i, irel))
else:
implicit_instances.append((i, irel))
ret = ((explicit_instances, a_ds[1]), (implicit_instances, a_ds[1]))
return ret
def predict(self, a_rel, a_data, a_ret, a_i):
"""Method for predicting sense of single relation.
Args:
a_rel (dict):
discourse relation whose sense should be predicted
a_data (2-tuple(dict, dict)):
list of input JSON data
a_ret (np.array):
output prediction vector
a_i (int):
row index in the output vector
Returns:
void:
Note:
modifies ``a_ret[a_i]`` in place
"""
if is_explicit(a_rel):
return self.explicit.predict(a_rel, a_data, a_ret, a_i)
return self.implicit.predict(a_rel, a_data, a_ret, a_i)
def predict(self, a_rel, a_data, a_ret, a_i):
"""Method for predicting sense of single relation.
Args:
a_rel (dict):
discourse relation whose sense should be predicted
a_data (2-tuple(dict, dict)):
list of input JSON data
a_ret (np.array):
output prediction vector
a_i (int):
row index in the output vector
Returns:
void:
Note:
modifies ``a_ret[a_i]`` in place
"""
if is_explicit(a_rel):
return self.explicit.predict(a_rel, a_data, a_ret, a_i)
return self.implicit.predict(a_rel, a_data, a_ret, a_i)
def _divide_ds(self, a_ds):
"""Separate dataset into explicit and implicit instances.
Args:
a_ds (tuple):
list of gold relations and dict with parses
Returns:
tuple:
trainings set with explicit and implicit connectives
"""
if not a_ds:
return (([], {}), ([], {}))
explicit_instances = []
implicit_instances = []
for i, irel in enumerate(a_ds[0]):
if is_explicit(irel):
explicit_instances.append((i, irel))
else:
implicit_instances.append((i, irel))
ret = ((explicit_instances, a_ds[1]),
(implicit_instances, a_ds[1]))
return ret
def test_is_explicit():
assert is_explicit({CONNECTIVE: {TOK_LIST: [1, 2, 3]}})
assert is_explicit({CONNECTIVE: {TOK_LIST: []}}) is False