def write_explanations_for_expressions(iteration_number):
print("Writing explanations")
DATA_FILE1 = 'data/expressions/expressions_train_list' + str(iteration_number) + '.pkl'
DATA_FILE2 = 'data/explanations/my_explanations_expressions' + str(iteration_number) + '.tsv'
with open(DATA_FILE1, 'rb') as f:
expressions_list = pickle.load(f)
index = 0
explanations = []
for expressions in progressbar.progressbar(expressions_list):
for expression in expressions:
explanation = Explanation(
name='LF_' + str(index),
label=expression[1],
condition=create_condition_for_expressions(expression[0]),
word=expression[0]
)
explanations.append(explanation)
index = index + 1
exp_io = ExplanationIO2()
exp_io.write(explanations, DATA_FILE2)
print("Done")
def apply_results():
best_iteration = get_best_iteration()
print(best_iteration)
DATA_FILE3 = 'data/tokens/correct_expressions_list' + str(best_iteration[1]) + '.pkl'
DATA_FILE4 = 'data/tokens/wrong_expressions_list' + str(best_iteration[1]) + '.pkl'
DATA_FILE5 = 'data/tokens/expressions_train_list' + str(best_iteration[1]) + '.pkl'
with open(DATA_FILE1, 'rb') as f:
Cs = pickle.load(f)
with open(DATA_FILE2, 'rb') as f:
Ys = pickle.load(f)
with open(DATA_FILE3, 'rb') as f:
correct_tokens_list = pickle.load(f)
with open(DATA_FILE4, 'rb') as f:
wrong_tokens_list = pickle.load(f)
with open(DATA_FILE5, 'rb') as f:
tokens_train_list = pickle.load(f)
tokens = []
for correct_tokens in correct_tokens_list:
for token in correct_tokens:
if token not in tokens:
tokens.append(token)
tokens2 = []
for tokens_train in tokens_train_list:
for token in tokens_train:
if token not in tokens2:
tokens2.append(token)
index = 0
explanations = []
for word in progressbar.progressbar(tokens2):
explanation = Explanation(
name='LF_' + str(index),
label=word[1],
condition=create_condition(word[0]),
word=word[0]
)
explanations.append(explanation)
index = index + 1
babbler = Babbler(Cs, Ys)
babbler.apply(explanations, split=0)
Ls = []
for split in [0, 1, 2]:
L = babbler.get_label_matrix(split)
Ls.append(L)
L_train = Ls[0].toarray()
L_test = Ls[2].toarray()
predicted_training_labels = []
predicted_test_labels = []
for line in L_train:
predicted_training_labels.append(most_frequent(line))
count = 0
for label in predicted_training_labels:
if label != 0:
count += 1
print(count)
for line in L_test:
predicted_test_labels.append(most_frequent(line))
len_wrong_train = calculate_number_wrong(Ys[0], predicted_training_labels)
len_wrong_test = calculate_number_wrong(Ys[2], predicted_test_labels)
training_accuracy = percentage(len(Ys[0]) - len_wrong_train, len(Ys[0]))
test_accuracy = percentage(len(Ys[2]) - len_wrong_test, len(Ys[2]))
print("Number of wrong in training set: " + str(len_wrong_train))
print("Number of wrong in test set: " + str(len_wrong_test))
print("Training Accuracy: " + str(training_accuracy))
print("Test Accuracy: " + str(test_accuracy))
def test2():
with open(DATA_FILE1, 'rb') as f:
Cs = pickle.load(f)
with open(DATA_FILE2, 'rb') as f:
Ys = pickle.load(f)
with open(DATA_FILE4, 'rb') as f:
correct_expressions_list = pickle.load(f)
index = 0
explanations = []
for expressions in progressbar.progressbar(correct_expressions_list):
for expression in expressions:
explanation = Explanation(
name='LF_' + str(index),
label=expression[1],
condition=create_condition_for_expressions(expression[0]),
word=expression[0]
)
explanations.append(explanation)
index = index + 1
babbler = Babbler(Cs, Ys)
babbler.apply(explanations, split=0)
Ls = []
for split in [0, 1, 2]:
L = babbler.get_label_matrix(split)
Ls.append(L)
babbler.commit()
parses = babbler.get_parses()
'''
parse = parses[222]
print(parse)
'''
L_train = Ls[0].toarray()
L_test = Ls[2].toarray()
predicted_training_labels = []
predicted_test_labels = []
for line in L_train:
predicted_training_labels.append(most_frequent(line))
for line in L_test:
predicted_test_labels.append(most_frequent(line))
len_wrong_train, training_accuracy = calculate_number_wrong_no_abstain(Ys[0], predicted_training_labels)
len_wrong_test, test_accuracy = calculate_number_wrong_no_abstain(Ys[2], predicted_test_labels)
print(predicted_training_labels)
print(predicted_test_labels)
print("Number of wrong in training set: " + str(len_wrong_train))
print("Number of wrong in test set: " + str(len_wrong_test))
print("Training Accuracy: " + str(training_accuracy))
print("Test Accuracy: " + str(test_accuracy))
'letters':['a','B','C'],
'smalls':['a','b','c','d'],
'luckies': [7, 8, 9],
'unluckies': [0, 13, 66],
}
# Test candidate (hash: 668761641257950361):
# "City land records show that GM President [Daniel Ammann] and his wife,
# [Pernilla Ammann], bought the 15-bedroom mansion on Balmoral Drive in
# the upscale historic neighborhood on July 31."
logic = [
# Base
Explanation(
condition="True",
label=1,
candidate=('foo', 'bar'),
semantics=('.root', ('.label', ('.int', 1), ('.bool', True)))),
# And
Explanation(
condition="True and True",
label=1,
candidate=('foo', 'bar'),
semantics=('.root', ('.label', ('.int', 1), ('.and', ('.bool', True), ('.bool', True))))),
# Or
Explanation(
condition="False or True",
label=1,
candidate=('foo', 'bar'),
semantics=('.root', ('.label', ('.int', 1), ('.or', ('.bool', False), ('.bool', True))))),
# Not boolean
'spouse': ['spouse', 'wife', 'husband', 'ex-wife', 'ex-husband'],
'family': [
'father', 'father', 'mother', 'sister', 'sisters', 'brother',
'brothers', 'son', 'sons', 'daughter', 'daughters', 'grandfather',
'grandmother', 'uncle', 'uncles', 'aunt', 'aunts', 'cousin', 'cousins'
],
'friend':
['boyfriend', 'girlfriend', 'boss', 'employee', 'secretary', 'co-worker'],
}
aliases['family'] += ["{}-in-law".format(f) for f in aliases['family']]
explanations = [
Explanation(
name='LF_and_married',
condition=
"the word 'and' is between X and Y and 'married' within five words of Y",
candidate=
'1bcd8648-8a80-47a3-82d4-38a4a594092f::span:1223:1228~~1bcd8648-8a80-47a3-82d4-38a4a594092f::span:1234:1238',
label=1),
Explanation(
name='LF_third_wheel',
condition="there is a person between X and Y",
candidate=
'6cd34ab0-653b-438e-b966-d7365a31651d::span:595:607~~6cd34ab0-653b-438e-b966-d7365a31651d::span:712:719',
label=2),
Explanation(
name='LF_married_two_people',
condition=
"the word 'married' is in the sentence and there are only two people in the sentence",
candidate=
'd535c921-f102-4d3b-9891-5a36ed93259e::span:823:830~~d535c921-f102-4d3b-9891-5a36ed93259e::span:836:839',