def predict_final_word_models(models, vocabulary, filename):
id_list = []
prev_tokens_list = []
prev_tokens_lens = []
with open(filename, "r") as fin:
fin.readline()
for line in fin:
id_, prev_sent, grt_last_token = line.strip().split(",")
id_list.append(id_)
prev_tokens = prev_sent.split()
prev_tokens_list.append(prev_tokens)
prev_tokens_lens.append(len(prev_tokens))
X = np.array([
build_input_data(t, vocabulary)[0][0].tolist()
for t in prev_tokens_list
])
y_prob = 0
ind = 0
for model in models:
if ind == 0:
att = 0.6
elif ind == 1:
att = 0.4
y_prob = y_prob + att * model.predict(X, batch_size=32)
ind = ind + 1
last_token_probs = np.array([
y_prob[b, prev_tokens_lens[b] - 1, :] for b in range(y_prob.shape[0])
])
return dict(zip(id_list, last_token_probs))
def predict_final_word(model_list, vocabulary, filename):
list_prob = []
id_list = []
for model in model_list:
id_list = []
prev_tokens_list = []
prev_tokens_lens = []
with open(filename, "r") as fin:
fin.readline()
for line in fin:
id_, prev_sent, grt_last_token = line.strip().split(",")
id_list.append(id_)
prev_tokens = prev_sent.split()
prev_tokens_list.append(prev_tokens)
prev_tokens_lens.append(len(prev_tokens))
X = np.array([
build_input_data(t, vocabulary)[0][0].tolist()
for t in prev_tokens_list
])
y_prob = (model.predict(X, batch_size=32))
last_token_probs = np.array([
y_prob[b, prev_tokens_lens[b] - 1, :]
for b in range(y_prob.shape[0])
])
list_prob.append(last_token_probs)
list_sum = 0
for a in list_prob:
list_sum = list_sum + a
last_token_probs = list_sum / (len(list_prob))
return dict(zip(id_list, last_token_probs))
