#coding:utf-8
import tensorflow as tf
import numpy as np
def dynamic_bi_rnn(inputs, seqlen, n_hidden, keep_prob, cell_name='', reuse=False):
batch_size = tf.shape(inputs)[0]
with tf.variable_scope(cell_name + 'fw', initializer=tf.contrib.layers.xavier_initializer(), dtype = tf.float32, reuse=reuse):
fw_cell = tf.contrib.rnn.GRUCell(n_hidden)
fw_cell = tf.contrib.rnn.DropoutWrapper(fw_cell, output_keep_prob = keep_prob)
fw_initial_state = fw_cell.zero_state(batch_size, tf.float32)
with tf.variable_scope(cell_name + 'bw', initializer=tf.contrib.layers.xavier_initializer(), dtype = tf.float32, reuse=reuse):
bw_cell = tf.contrib.rnn.GRUCell(n_hidden)
bw_cell = tf.contrib.rnn.DropoutWrapper(bw_cell, output_keep_prob = keep_prob)
bw_initial_state = bw_cell.zero_state(batch_size, tf.float32)
with tf.variable_scope(cell_name, reuse=reuse):
outputs, output_states = tf.nn.bidirectional_dynamic_rnn(fw_cell, bw_cell, inputs,
initial_state_fw=fw_initial_state,
initial_state_bw=bw_initial_state,
sequence_length=seqlen)
return outputs, output_states
def get_matrix_tree(r, A):
L = tf.reduce_sum(A, 1)
L = tf.matrix_diag(L)
L = L - A
r_diag = tf.matrix_diag(r)
LL = L + r_diag
LL_inv = tf.matrix_inverse(LL) #batch_l, doc_l, doc_l
LL_inv_diag_ = tf.matrix_diag_part(LL_inv)
d0 = tf.multiply(r, LL_inv_diag_)
LL_inv_diag = tf.expand_dims(LL_inv_diag_, 2)
tmp1 = tf.multiply(A, tf.matrix_transpose(LL_inv_diag))
tmp2 = tf.multiply(A, tf.matrix_transpose(LL_inv))
d = tmp1 - tmp2
d = tf.concat([tf.expand_dims(d0,[1]), d], 1)
return d
def discourse_rank(str_scores, damp):
batch_l = tf.shape(str_scores)[0]
max_doc_l = tf.shape(str_scores)[2]
str_scores_root = str_scores[:, 0, :]
str_scores_words = str_scores[:, 1:, :]
str_root = tf.expand_dims(tf.one_hot(indices=0, depth=(max_doc_l+1),
on_value=0.0, off_value=1.0/tf.cast(max_doc_l, tf.float32), dtype=tf.float32), 0)
str_roots = tf.expand_dims(tf.tile(str_root, [batch_l, 1]), -1)
adj_scores = tf.concat([str_roots, str_scores], 2)
eye = tf.expand_dims(tf.diag(tf.ones(max_doc_l+1)), 0)
eye_words = tf.tile(eye, [batch_l, 1, 1])
eig_matrix_inv = tf.matrix_inverse(eye_words - damp*adj_scores)
damp_vec = tf.ones([batch_l, (max_doc_l+1), 1]) / tf.cast((max_doc_l+1), tf.float32)
eig_scores_ = tf.multiply((1-damp), tf.matmul(eig_matrix_inv, damp_vec))
eig_scores_ = tf.squeeze(eig_scores_, 2)[:, 1:]
eig_scores = tf.nn.softmax(eig_scores_, 1)
eig_str_scores_root = tf.expand_dims(tf.multiply(eig_scores, str_scores_root), 1)
eig_str_scores = tf.concat([eig_str_scores_root, str_scores_words], 1)
return eig_str_scores
