def get_history_sum_embedded(self):
# TODO: add mask info for this operation
his_days =['one', 'two', 'three', 'four','five','six','seven','eight','nine',
'ten','eleven','twelve','thirteen','fourteen']
for fir in his_days:
key = "history_" + fir + "_chap_ph"
embed_key = "history_" + fir + "_chap_embedded"
setattr(self, embed_key,get_mask_zero_embedded(self.chapters_embeddings_var,
getattr(self, key)))
for fir in his_days:
key = "history_" + fir + "_sec_ph"
embed_key = "history_" + fir + "_sec_embedded"
setattr(self, embed_key,get_mask_zero_embedded(self.sections_embeddings_var,
getattr(self, key)))
chap = tf.reduce_mean(self.history_one_chap_embedded, axis=-2)
#b*x*128 b*128 b*(128*14)
sec = tf.reduce_mean(self.history_one_sec_embedded, axis=-2)
for fir in his_days[:0:-1]:
key_c = "history_" + fir + "_chap_embedded"
chap = tf.concat([chap, tf.reduce_mean(getattr(self, key_c), axis=-2)], axis=-1)
key_s = "history_" + fir + "_sec_embedded"
sec = tf.concat([sec, tf.reduce_mean(getattr(self, key_s), axis=-2)], axis=-1)
history_chap_emb = tf.reshape(chap, [-1, HIS_DAYS, EMBEDDING_DIM])
history_sec_emb = tf.reshape(sec, [-1, HIS_DAYS, EMBEDDING_DIM])
chap_mean = tf.reduce_mean(history_chap_emb, axis=-2)
sec_mean = tf.reduce_mean(history_sec_emb, axis=-2)
return chap_mean, sec_mean
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 __call__(self, i_to_s, state, scope="DiagonalBiLSTMCell"):
c_prev = tf.slice(state, [0, 0], [-1, self._num_units])
h_prev = tf.slice(state, [0, self._num_units], [-1, self._num_units])
with tf.compat.v1.variable_scope(scope):
conv1d_inputs = tf.reshape(
h_prev, [-1, self._height, 1, self._hidden_dims],
name='conv1d_inputs')
conv_s_to_s = conv1d(conv1d_inputs,
4 * self._hidden_dims,
2,
scope='s_to_s')
s_to_s = tf.reshape(conv_s_to_s,
[-1, self._height * self._hidden_dims * 4])
lstm_matrix = tf.sigmoid(s_to_s + i_to_s)
i, g, f, o = tf.split(lstm_matrix, 4, 1)
c = f * c_prev + i * g
h = tf.multiply(o, tf.tanh(c), name='hid')
new_state = tf.concat([c, h], 1)
return h, new_state
def get_history_sum_embedded(self):
# TODO: add mask info for this operation
his_days =['one', 'two', 'three', 'four','five','six','seven','eight','nine',
'ten','eleven','twelve','thirteen','fourteen']
for fir in his_days:
key_c = "history_" + fir + "_chap_ph"
embed_key_c = "history_" + fir + "_chap_embedded"
setattr(self, embed_key_c,get_mask_zero_embedded(self.chapters_embeddings_var,
getattr(self, key_c)))
key_s = "history_" + fir + "_sec_ph"
embed_key_s = "history_" + fir + "_sec_embedded"
setattr(self, embed_key_s,get_mask_zero_embedded(self.sections_embeddings_var,
getattr(self, key_s)))
key_st = "style_" + fir + "_ph"
embed_key_st = "history_" + fir + "_sty_embedded"
setattr(self, embed_key_st,tf.nn.embedding_lookup(self.style_embeddings_var,
getattr(self, key_st)))
chap = get_rnn_sum(self.history_one_chap_embedded,"rnncha")
#b*x*128 b*128 b*(128*14)
sec = get_rnn_sum(self.history_one_sec_embedded, "rnnsec")
sty = self.history_one_sty_embedded
for fir in his_days[:0:-1]:
key_c = "history_" + fir + "_chap_embedded"
chap = tf.concat([chap, get_rnn_sum(getattr(self, key_c), "rnncha")], axis=-1)
key_s = "history_" + fir + "_sec_embedded"
sec = tf.concat([sec, get_rnn_sum(getattr(self, key_s), "rnnsec")], axis=-1)
key_st = "history_" + fir + "_sty_embedded"
sty = tf.concat([sty, getattr(self, key_st)], axis=-1)
history_chap_emb = tf.reshape(chap, [-1, HIS_DAYS, EMBEDDING_DIM])
history_sec_emb = tf.reshape(sec, [-1, HIS_DAYS, EMBEDDING_DIM])
history_sty_emb = tf.reshape(sty, [-1, HIS_DAYS, EMBEDDING_DIM])
chap_mean = tf.reduce_mean(history_chap_emb, axis=-2)
sec_mean = tf.reduce_mean(history_sec_emb, axis=-2)
sty_mean = tf.reduce_mean(history_sty_emb, axis=-2)
#return chap_mean, sec_mean
return tf.concat([chap_mean, sec_mean,sty_mean], axis=-1)
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