data, table = chains.process_corpus("train.csv")
table.write("all.json")
print("Successfully built the model")
sys.exit(0)
else:
if not path.exists("all.json"):
print("Load the model first using --load")
# load the pre-built model
with open("all.json") as fp:
table = chains.ProbabilityTable(json.load(fp))
total, correct = 0, 0
# load testing data
test = chains.load_data("test.csv")
with open("answers.txt", "w") as f:
# header for the csv
f.write("InputStoryid,AnswerRightEnding\n")
test_tqdm = tqdm.tqdm(test)
for t in test_tqdm:
one, two = parse_test_instance(t)
_, one_deps = chains.extract_dependency_pairs(one)
_, two_deps = chains.extract_dependency_pairs(two)
# logic to choose between one and two
prog_one = chains.protagonist(one)
prog_two = chains.protagonist(two)
return [
chains.ParsedStory(id, id, chains.nlp(" ".join(story[2:6] + [a])),
*(sentences + [chains.nlp(a)]))
for a in alternatives
]
def story_answer(story):
"""Tells you the correct answer. Return (storyid, index). 1 for the first ending, 2 for the second ending"""
#obviously you can't use this information until you've chosen your answer!
return story.InputStoryid, story.AnswerRightEnding
# Load training data and build the model
#data, table = chains.process_corpus("train.csv", 100)
#print(table.pmi("move", "nsubj", "move", "nsubj"))
# load the pre-built model
with open("all.json") as fp:
table = chains.ProbabilityTable(json.load(fp))
# load testing data
test = chains.load_data("val.csv")
for t in test:
one, two = parse_test_instance(t)
one_deps = chains.extract_dependency_pairs(one)
pprint(one[2:])
pprint(two[2:])
# logic to choose between one and two
pprint("answer:" + str(story_answer(t)))