def execute(trial, n_epochs, timetags, word2vec, roles, coref):
train_stories, dev_stories, test_stories = load_data(coref)
all_stories = train_stories + test_stories + dev_stories
if timetags:
d = 556
else:
d = 300
model = WeakMemoryNetwork(
d, all_stories, timetags=timetags, word2vec=word2vec, roles=roles,
coref=coref, path=path)
pre_train_acc = compute_accuracy(train_stories, model)
pre_test_acc = compute_accuracy(test_stories, model)
for _ in range(n_epochs):
for story in train_stories + dev_stories:
model.train(story)
post_train_acc = compute_accuracy(train_stories, model)
post_test_acc = compute_accuracy(test_stories, model)
return {
'pre_train_acc': pre_train_acc,
'pre_test_acc': pre_test_acc,
'post_train_acc': post_train_acc,
'post_test_acc': post_test_acc,
}
def execute(trial, n_epochs, timetags, shift, word2vec, roles, coref, preinit):
train_stories, dev_stories, test_stories = load_data(coref)
all_stories = train_stories + test_stories + dev_stories
if coref:
pos_file = 'coref_pos.pkl'
else:
pos_file = 'pos.pkl'
model = WeakMemoryNetwork(
300, 256, all_stories, timetags=timetags, word2vec=word2vec,
roles=roles, coref=coref, preinit=preinit, shift=shift, path=path,
pos_file=pos_file)
pre_train_acc = compute_accuracy(train_stories, model)
pre_test_acc = compute_accuracy(test_stories, model)
for _ in range(n_epochs):
for story in train_stories + dev_stories:
model.train(story)
post_train_acc = compute_accuracy(train_stories, model)
post_test_acc = compute_accuracy(test_stories, model)
return {
'pre_train_acc': pre_train_acc,
'pre_test_acc': pre_test_acc,
'post_train_acc': post_train_acc,
'post_test_acc': post_test_acc,
}
with open('MCTest/mc160.dev.coref','rb') as f:
dev_stories = clean(pickle.load(f))
with open('MCTest/mc160.train.coref','rb') as f:
train_stories = clean(pickle.load(f))
with open('MCTest/mc160.test.coref','rb') as f:
test_stories = clean(pickle.load(f))
all_stories = train_stories + test_stories + dev_stories
# initialize with all stories to get full vocab
model = WeakMemoryNetwork(
300, 256, all_stories, timetags=True, word2vec=True, roles=True, coref=True,
preinit=False, pos_file='coref_pos.pkl')
print 'Training Accuracy prior to training: ', compute_accuracy(train_stories, model)
print 'Testing Accuracy prior to training: ', compute_accuracy(test_stories, model)
# Train for a certain number of epochs
count = 0
for i in range(30):
for story in train_stories + dev_stories:
model.train(story)
print 'Iteration ', count, ' complete!'
count += 1
# Test for good generalization