def train(train_loc, dev_loc, shape, settings):
train_texts1, train_texts2, train_labels = read_snli(train_loc)
dev_texts1, dev_texts2, dev_labels = read_snli(dev_loc)
print("Loading spaCy")
nlp = spacy.load("en_vectors_web_lg")
assert nlp.path is not None
print("Processing texts...")
train_X = create_dataset(nlp, train_texts1, train_texts2, 100, shape[0])
dev_X = create_dataset(nlp, dev_texts1, dev_texts2, 100, shape[0])
print("Compiling network")
model = build_model(get_embeddings(nlp.vocab), shape, settings)
print(settings)
model.fit(
train_X,
train_labels,
validation_data=(dev_X, dev_labels),
epochs=settings["nr_epoch"],
batch_size=settings["batch_size"],
)
if not (nlp.path / "similarity").exists():
(nlp.path / "similarity").mkdir()
print("Saving to", nlp.path / "similarity")
weights = model.get_weights()
# remove the embedding matrix. We can reconstruct it.
del weights[1]
with (nlp.path / "similarity" / "model").open("wb") as file_:
pickle.dump(weights, file_)
with (nlp.path / "similarity" / "config.json").open("w") as file_:
file_.write(model.to_json())
def train(train_loc, dev_loc, shape, settings):
train_texts1, train_texts2, train_labels = read_snli(train_loc)
dev_texts1, dev_texts2, dev_labels = read_snli(dev_loc)
print("Loading spaCy")
nlp = spacy.load('en')
assert nlp.path is not None
print("Compiling network")
model = build_model(get_embeddings(nlp.vocab), shape, settings)
print("Processing texts...")
Xs = []
for texts in (train_texts1, train_texts2, dev_texts1, dev_texts2):
Xs.append(get_word_ids(list(nlp.pipe(texts, n_threads=20, batch_size=20000)),
max_length=shape[0],
rnn_encode=settings['gru_encode'],
tree_truncate=settings['tree_truncate']))
train_X1, train_X2, dev_X1, dev_X2 = Xs
print(settings)
model.fit(
[train_X1, train_X2],
train_labels,
validation_data=([dev_X1, dev_X2], dev_labels),
nb_epoch=settings['nr_epoch'],
batch_size=settings['batch_size'])
if not (nlp.path / 'similarity').exists():
(nlp.path / 'similarity').mkdir()
print("Saving to", nlp.path / 'similarity')
weights = model.get_weights()
with (nlp.path / 'similarity' / 'model').open('wb') as file_:
pickle.dump(weights[1:], file_)
with (nlp.path / 'similarity' / 'config.json').open('wb') as file_:
file_.write(model.to_json())
def train(train_loc, dev_loc, shape, settings):
train_texts1, train_texts2, train_labels = read_snli(train_loc)
dev_texts1, dev_texts2, dev_labels = read_snli(dev_loc)
print("Loading spaCy")
nlp = spacy.load('en')
assert nlp.path is not None
print("Compiling network")
model = build_model(get_embeddings(nlp.vocab), shape, settings)
print("Processing texts...")
Xs = []
for texts in (train_texts1, train_texts2, dev_texts1, dev_texts2):
Xs.append(get_word_ids(list(nlp.pipe(texts, n_threads=20, batch_size=20000)),
max_length=shape[0],
rnn_encode=settings['gru_encode'],
tree_truncate=settings['tree_truncate']))
train_X1, train_X2, dev_X1, dev_X2 = Xs
print(settings)
model.fit(
[train_X1, train_X2],
train_labels,
validation_data=([dev_X1, dev_X2], dev_labels),
nb_epoch=settings['nr_epoch'],
batch_size=settings['batch_size'])
if not (nlp.path / 'similarity').exists():
(nlp.path / 'similarity').mkdir()
print("Saving to", nlp.path / 'similarity')
weights = model.get_weights()
with (nlp.path / 'similarity' / 'model').open('wb') as file_:
pickle.dump(weights[1:], file_)
with (nlp.path / 'similarity' / 'config.json').open('wb') as file_:
file_.write(model.to_json())
def train(model_dir, train_loc, dev_loc, shape, settings):
train_texts1, train_texts2, train_labels = read_snli(train_loc)
dev_texts1, dev_texts2, dev_labels = read_snli(dev_loc)
print("Loading spaCy")
nlp = spacy.load('en')
print("Compiling network")
model = build_model(get_embeddings(nlp.vocab), shape, settings)
print("Processing texts...")
Xs = []
for texts in (train_texts1, train_texts2, dev_texts1, dev_texts2):
Xs.append(
get_word_ids(list(nlp.pipe(texts, n_threads=20, batch_size=20000)),
max_length=shape[0],
rnn_encode=settings['gru_encode'],
tree_truncate=settings['tree_truncate']))
train_X1, train_X2, dev_X1, dev_X2 = Xs
print(settings)
model.fit([train_X1, train_X2],
train_labels,
validation_data=([dev_X1, dev_X2], dev_labels),
nb_epoch=settings['nr_epoch'],
batch_size=settings['batch_size'])
def train(model_dir, train_loc, dev_loc, shape, settings):
train_texts1, train_texts2, train_labels = read_snli(train_loc)
dev_texts1, dev_texts2, dev_labels = read_snli(dev_loc)
print("Loading spaCy")
nlp = spacy.load('en')
print("Compiling network")
model = build_model(get_embeddings(nlp.vocab), shape, settings)
print("Processing texts...")
Xs = []
for texts in (train_texts1, train_texts2, dev_texts1, dev_texts2):
Xs.append(get_word_ids(list(nlp.pipe(texts, n_threads=20, batch_size=20000)),
max_length=shape[0],
rnn_encode=settings['gru_encode'],
tree_truncate=settings['tree_truncate']))
train_X1, train_X2, dev_X1, dev_X2 = Xs
print(settings)
model.fit(
[train_X1, train_X2],
train_labels,
validation_data=([dev_X1, dev_X2], dev_labels),
nb_epoch=settings['nr_epoch'],
batch_size=settings['batch_size'])
def attention_foldrun(X, X2, y, name, Xte = None, Xte2 = None, start_fold = 0):
skf = StratifiedKFold(n_splits = 10, random_state = 111, shuffle = True)
if isinstance(X, pd.core.frame.DataFrame):
X = X.values
if isinstance(y, pd.core.frame.DataFrame):
y = y.is_duplicate.values
if isinstance(y, pd.core.frame.Series):
y = y.values
print('Running Decomposable Attention model with parameters:', settings)
i = 1
losses = []
train_splits = []
val_splits = []
for tr_index, val_index in skf.split(X, y):
train_splits.append(tr_index)
val_splits.append(val_index)
for i in range(start_fold, start_fold + 2):
X_trq1, X_valq1 = X[train_splits[i]], X[val_splits[i]]
X_trq2, X_valq2 = X2[train_splits[i]], X2[val_splits[i]]
y_tr, y_val = y[train_splits[i]], y[val_splits[i]]
y_tr = to_categorical(y_tr)
y_val = to_categorical(y_val)
t = time.time()
print('Start training on fold: {}'.format(i))
callbacks = [ModelCheckpoint('checks/decomposable_{}_10SKF_fold{}.h5'.format(i, name),
monitor='val_loss',
verbose = 0, save_best_only = True),
EarlyStopping(monitor='val_loss', patience = 4, verbose = 1)]
model = build_model(get_embeddings(nlp.vocab), shape, settings)
model.fit([X_trq1, X_trq2], y_tr, validation_data=([X_valq1, X_valq2], y_val),
nb_epoch=settings['nr_epoch'], batch_size=settings['batch_size'], callbacks = callbacks)
val_pred = model.predict([X_valq1, X_valq2], batch_size = 64)
score = log_loss(y_val, val_pred)
losses.append(score)
print('Predicting training set.')
val_pred = pd.DataFrame(val_pred, index = val_splits[i])
val_pred.columns = ['attention_feat1', 'attention_feat2']
val_pred.to_pickle('OOF_preds/train_attentionpreds_fold{}.pkl'.format(i))
print(val_pred.head())
if Xte is not None:
print('Predicting test set.')
test_preds = model.predict([Xte, Xte2], batch_size = 64)
test_preds = pd.DataFrame(test_preds)
test_preds.columns = ['attention_feat1', 'attention_feat2']
test_preds.to_pickle('OOF_preds/test_attentionpreds_fold{}.pkl'.format(i))
del test_preds
gc.collect()
print('Final score for fold {} :'.format(i), score, '\n',
'Time it took to train and predict on fold:', time.time() - t, '\n')
del X_trq1, X_valq1, X_trq2, X_valq2, y_tr, y_val, val_pred
gc.collect()
i += 1
print('Mean logloss for model in 10-folds SKF:', np.array(losses).mean(axis = 0))
return
def train(train_loc, dev_loc, shape, settings):
train_texts1, train_texts2, train_labels, train_styling_arrays_1, train_styling_arrays_2, train_TWPs_1, train_TWPs_2 = read_snli(
train_loc)
dev_texts1, dev_texts2, dev_labels, dev_styling_arrays_1, dev_styling_arrays_2, dev_TWPs_1, dev_TWPs_2 = read_snli(
dev_loc)
print("Loading spaCy")
nlp = en_core_web_sm.load()
# en_vectors_web_lg.load(vocab=nlp.vocab)
path = '/home/ankesh/div_merging_models/alpha1/'
print("Compiling network")
# sense = sense2vec.load()
model = build_model(get_embeddings(nlp.vocab), shape, settings)
print("Processing texts...")
Xs = []
# train_texts1 = train_texts1[:1000]
# train_styling_arrays_1 = train_styling_arrays_1[:10]
# train_TWPs_1 = train_TWPs_1[:10]
# train_texts2 = train_texts2[:10]
# train_styling_arrays_2 = train_styling_arrays_2[:10]
# train_TWPs_2 = train_TWPs_2[:10]
# dev_texts1 = dev_texts1[:1]
# dev_styling_arrays_1 = dev_styling_arrays_1[:1]
# dev_TWPs_1 = dev_TWPs_1[:1]
# dev_texts2 = dev_texts2[:1]
# dev_styling_arrays_2 = dev_styling_arrays_2[:1]
# dev_TWPs_2 = dev_TWPs_2[:1]
# print (train_texts1[0])
# print (train_texts2[0])
for texts, styling_array, TWP in ((train_texts1, train_styling_arrays_1,
train_TWPs_1),
(train_texts2, train_styling_arrays_2,
train_TWPs_2),
(dev_texts1, dev_styling_arrays_1,
dev_TWPs_1), (dev_texts2,
dev_styling_arrays_2,
dev_TWPs_2)):
Xs.append(
get_word_ids(list(nlp.pipe(texts, n_threads=20, batch_size=20000)),
styling_array,
TWP,
max_length=shape[0],
rnn_encode=settings['gru_encode'],
tree_truncate=settings['tree_truncate']))
train_X1, train_X2, dev_X1, dev_X2 = Xs
# print (train_X1[0])
# print ('-'*10)
# print (train_X2[0])
# print ("shape of train X1", train_X1.shape)
# print("+"*40)
print(settings)
model.fit([train_X1, train_X2],
train_labels,
validation_data=([dev_X1, dev_X2], dev_labels),
nb_epoch=settings['nr_epoch'],
batch_size=settings['batch_size'])
# if not (nlp.path / 'similarity').exists():
# (nlp.path / 'similarity').mkdir()
print("Saving to", path + 'similarity')
weights = model.get_weights()
with open(path + 'similarity/' + 'model', 'wb') as file_:
pickle.dump(weights[1:], file_)
with open(path + 'similarity/' + 'config.json', 'wb') as file_:
file_.write(model.to_json())