INNER CODE UNIT · Python

x_inp

ematvey/hierarchical-attention-networks · bn_lstm_test.py:18

x_inp = tf.expand_dims(x, -1)
lstm = BNLSTMCell(hidden_size, training) #LSTMCell(hidden_size)

#c, h
initialState = (
    tf.random_normal([batch_size, hidden_size], stddev=0.1),
    tf.random_normal([batch_size, hidden_size], stddev=0.1))

outputs, state = dynamic_rnn(lstm, x_inp, initial_state=initialState, dtype=tf.float32)

_, final_hidden = state

W = tf.get_variable('W', [hidden_size, 10], initializer=orthogonal_initializer())
b = tf.get_variable('b', [10])

y = tf.nn.softmax(tf.matmul(final_hidden, W) + b)

y_ = tf.placeholder(tf.float32, [None, 10])

View source record →