def cut_batch(self, step, is_train = True, sampling = 1): n = step % self.train_batch begin = self.batch_size * n end = self.batch_size * (n + 1) if is_train == True: self.x_batch = self.x[begin:end] self.y_batch = self.y[begin:end] else: self.x_batch = self.x_valid[begin:end] self.y_batch = self.y_valid[begin:end] count = 0.0 # number of total words need to be computed loss in one batch, self.mask = self.mask_zero[:] for i in range(self.batch_size): j = np.where(self.y_batch[i] == self.EOS)[0][0] count += j+1 self.mask[i][:j+1] = 1 self.count_words = count self.x_batch = seq2seq_lib.one_hot(\ self.x_batch, self.batch_size, self.num_steps, self.vocab_size) self.y_batch = seq2seq_lib.one_hot(\ self.y_batch, self.batch_size, self.num_steps, self.vocab_size) return
def load_test(self):
self.x_test = seq2seq_lib.preprocess_testing_data(self.num_steps)
print self.x_test
if self.x_test.shape[0] <= 64:
self.x_test_vocab = seq2seq_lib.one_hot(\
self.x_test, self.x_test.shape[0], self.num_steps, self.vocab_size)
else:
self.x_test_vocab = seq2seq_lib.one_hot(\
self.x_test[:64], 64, self.num_steps, self.vocab_size)
self.x_test_vocab_1 = seq2seq_lib.one_hot(\
self.x_test[64:], self.x_test.shape[0]-64, self.num_steps, self.vocab_size)
def load_test(self): self.x_test = seq2seq_lib.preprocess_testing_data(self.num_steps) self.x_test_vocab = seq2seq_lib.one_hot(\ self.x_test, self.x_test.shape[0], self.num_steps, self.vocab_size)