def diagonal_lstm(inputs, scope='diagonal_lstm'):
with tf.compat.v1.variable_scope(scope):
skewed_inputs = skew(inputs, scope="skewed_i")
input_to_state = conv2d(skewed_inputs,
64, [1, 1],
mask_type="b",
scope="i_to_s")
column_wise_inputs = tf.transpose(input_to_state, perm=[0, 2, 1, 3])
batch, width, height, channel = column_wise_inputs.get_shape().as_list(
)
rnn_inputs = tf.reshape(column_wise_inputs,
[-1, width, height * channel])
cell = DiagonalLSTMCell(16, height, channel)
outputs, states = tf.compat.v1.nn.dynamic_rnn(cell,
inputs=rnn_inputs,
dtype=tf.float32)
width_first_outputs = tf.reshape(outputs, [-1, width, height, 16])
skewed_outputs = tf.transpose(width_first_outputs, perm=[0, 2, 1, 3])
outputs = unskew(skewed_outputs)
return outputs
def skew(inputs, scope="skew"):
with tf.compat.v1.name_scope(scope):
batch, height, width, channel = inputs.get_shape().as_list()
rows = tf.split(inputs, height, 1)
new_width = width + height - 1
new_rows = []
for idx, row in enumerate(rows):
transposed_row = tf.transpose(tf.squeeze(row, [1]), perm=[0, 2, 1])
squeezed_row = tf.reshape(transposed_row, [-1, width])
padded_row = tf.pad(squeezed_row, paddings=((0, 0), (idx, height - 1 - idx)))
unsqueezed_row = tf.reshape(padded_row, [-1, channel, new_width])
untransposed_row = tf.transpose(unsqueezed_row, perm=[0, 2, 1])
new_rows.append(untransposed_row)
outputs = tf.stack(new_rows, axis=1, name="output")
return outputs
def LSTM_Sentiment(input_tensor):
# Reference Paper: https://www.bioinf.jku.at/publications/older/2604.pdf
lstmCell = tf.compat.v1.nn.rnn_cell.BasicLSTMCell(1024)
output_rnn, _ = tf.compat.v1.nn.dynamic_rnn(lstmCell, input_tensor, dtype=tf.float32)
W_fc = tf.Variable(tf.random.truncated_normal([1024, 2]))
b_fc = tf.Variable(tf.constant(0.1, shape=[2]))
output_transposed = tf.transpose(output_rnn, perm=[1, 0, 2])
output = tf.gather(output_transposed, int(output_transposed.get_shape()[0]) - 1)
return tf.identity(tf.matmul(output, W_fc) + b_fc, name="output")