def cell_create(self, scope_name):
with tf.compat.v1.variable_scope(scope_name):
if self.cell_type == 'tanh':
cells = rnn.MultiRNNCell([
rnn.BasicRNNCell(self.n_hidden[i])
for i in range(self.n_layers)
],
state_is_tuple=True)
elif self.cell_type == 'LSTM':
cells = rnn.MultiRNNCell([
rnn.BasicLSTMCell(self.n_hidden[i])
for i in range(self.n_layers)
],
state_is_tuple=True)
elif self.cell_type == 'GRU':
cells = rnn.MultiRNNCell([
rnn.GRUCell(self.n_hidden[i]) for i in range(self.n_layers)
],
state_is_tuple=True)
elif self.cell_type == 'LSTMP':
cells = rnn.MultiRNNCell([
rnn.LSTMCell(self.n_hidden[i])
for i in range(self.n_layers)
],
state_is_tuple=True)
cells = rnn.DropoutWrapper(cells,
input_keep_prob=self.dropout_ph,
output_keep_prob=self.dropout_ph)
return cells
def create_model(self, optimizer):
features = tf.placeholder(tf.int32, [None, self.seq_len], name='features')
labels = tf.placeholder(tf.int64, [None,], name='labels')
embs = tf.Variable(self.emb_arr, dtype=tf.float32, trainable=False)
x = tf.nn.embedding_lookup(embs, features)
forward_cell1, backward_cell1 = rnn.BasicGRUCell(64), rnn.BasicGRUCell(64)
forward_cell2, backward_cell2 = rnn.BasicGRUCell(32), rnn.BasicGRUCell(32)
(birnn_fw1, birnn_bw1), _, _ = dynamic_birnn(forward_cell1, backward_cell1, x, dtype=tf.float32)
print("shape:", birnn_fw1.shape, birnn_bw1.shape)
stacked_lstm = rnn.MultiRNNCell(
[rnn.BasicLSTMCell(self.n_hidden) for _ in range(2)])
outputs, _ = tf.nn.dynamic_rnn(stacked_lstm, x, dtype=tf.float32)
fc1 = tf.layers.dense(inputs=outputs[:,-1,:], units=30)
pred = tf.squeeze(tf.layers.dense(inputs=fc1, units=1))
loss = tf.losses.sigmoid_cross_entropy(multi_class_labels=labels, logits=pred)
#optimizer = tf.train.AdamOptimizer(learning_rate=self.lr)
grads_and_vars = optimizer.compute_gradients(loss)
grads, _ = zip(*grads_and_vars)
train_op = optimizer.apply_gradients(grads_and_vars, global_step=tf.train.get_global_step())
correct_pred = tf.equal(tf.to_int64(tf.greater(pred,0)), labels)
eval_metric_ops = tf.count_nonzero(correct_pred)
return features, labels, train_op, grads, eval_metric_ops, loss
def create_model(self, optimizer):
features = tf.placeholder(tf.int32, [None, self.seq_len])
embedding = tf.get_variable("embedding", [self.num_classes, 8])
x = tf.nn.embedding_lookup(embedding, features)
labels = tf.placeholder(tf.int32, [None, self.num_classes])
stacked_lstm = rnn.MultiRNNCell(
[rnn.BasicLSTMCell(self.n_hidden) for _ in range(2)])
outputs, _ = tf.nn.dynamic_rnn(stacked_lstm, x, dtype=tf.float32)
pred = tf.layers.dense(inputs=outputs[:, -1, :],
units=self.num_classes)
loss = tf.reduce_mean(
tf.nn.softmax_cross_entropy_with_logits_v2(logits=pred,
labels=labels))
grads_and_vars = optimizer.compute_gradients(loss)
grads, _ = zip(*grads_and_vars)
train_op = optimizer.apply_gradients(
grads_and_vars, global_step=tf.train.get_global_step())
correct_pred = tf.equal(tf.argmax(pred, 1), tf.argmax(labels, 1))
eval_metric_ops = tf.count_nonzero(correct_pred)
return features, labels, train_op, grads, eval_metric_ops, loss
def __init__(self, dim_bert, rnn_size, rnn_layer, batch_size,
input_embedding_size, dim_hidden, max_words_d,
vocabulary_size_d, drop_out_rate, method):
self.dim_bert = dim_bert
self.rnn_size = rnn_size
self.rnn_layer = rnn_layer
self.batch_size = batch_size
self.input_embedding_size = input_embedding_size
self.dim_hidden = dim_hidden
self.max_words_d = max_words_d
self.vocabulary_size_d = vocabulary_size_d
self.drop_out_rate = drop_out_rate
self.method = method
if self.method == 'lstm':
# print("\nmethod: {}\n".format(self.method))
# description-embedding
self.embed_desc_W = tf.Variable(tf.random.uniform(
[self.vocabulary_size_d, self.input_embedding_size], -0.08,
0.08),
name='embed_desc_W')
# encoder: RNN body
self.lstm_1_d = rnn_cell.LSTMCell(rnn_size,
input_embedding_size,
state_is_tuple=False)
self.lstm_dropout_1_d = rnn_cell.DropoutWrapper(
self.lstm_1_d, output_keep_prob=1 - self.drop_out_rate)
self.lstm_2_d = rnn_cell.LSTMCell(rnn_size,
rnn_size,
state_is_tuple=False)
self.lstm_dropout_2_d = rnn_cell.DropoutWrapper(
self.lstm_2_d, output_keep_prob=1 - self.drop_out_rate)
self.stacked_lstm_d = rnn_cell.MultiRNNCell(
[self.lstm_dropout_1_d, self.lstm_dropout_2_d])
# description state-embedding
self.embed_state_desc_W = tf.Variable(tf.random.uniform(
[2 * rnn_size * rnn_layer, self.dim_hidden], -0.08, 0.08),
name='embed_state_desc_W')
elif self.method == 'bert':
# print("\nmethod: {}\n".format(self.method))
# description state-embedding
self.embed_state_desc_W = tf.Variable(tf.random.uniform(
[dim_bert, self.dim_hidden], -0.08, 0.08),
name='embed_state_desc_W')
self.embed_state_desc_b = tf.Variable(tf.random.uniform(
[self.dim_hidden], -0.08, 0.08),
name='embed_state_desc_b')
# score embedding
self.embed_score_W = tf.Variable(tf.random.uniform(
[dim_hidden, vocabulary_size_d], -0.08, 0.08),
name='embed_score_W')
self.embed_score_b = tf.Variable(tf.random.uniform([vocabulary_size_d],
-0.08, 0.08),
name='embed_score_b')
def __init__(self, rnn_size, rnn_layer, batch_size, input_embedding_size, dim_image, dim_hidden, max_words_q, vocabulary_size, max_words_d, vocabulary_size_d, drop_out_rate, num_output):
self.rnn_size = rnn_size
self.rnn_layer = rnn_layer
self.batch_size = batch_size
self.input_embedding_size = input_embedding_size
self.dim_image = dim_image
self.dim_hidden = dim_hidden
self.max_words_q = max_words_q
self.vocabulary_size = vocabulary_size
self.max_words_d = max_words_d
self.vocabulary_size_d = vocabulary_size_d
self.drop_out_rate = drop_out_rate
self.num_output = num_output
# question-embedding
self.embed_ques_W = tf.Variable(tf.random.uniform([self.vocabulary_size, self.input_embedding_size], -0.08, 0.08), name='embed_ques_W')
# encoder: RNN body
self.lstm_1 = rnn_cell.LSTMCell(rnn_size, input_embedding_size, state_is_tuple=False)
self.lstm_dropout_1 = rnn_cell.DropoutWrapper(self.lstm_1, output_keep_prob = 1 - self.drop_out_rate)
self.lstm_2 = rnn_cell.LSTMCell(rnn_size, rnn_size, state_is_tuple=False)
self.lstm_dropout_2 = rnn_cell.DropoutWrapper(self.lstm_2, output_keep_prob = 1 - self.drop_out_rate)
self.stacked_lstm = rnn_cell.MultiRNNCell([self.lstm_dropout_1, self.lstm_dropout_2])
# state-embedding
self.embed_state_W = tf.Variable(tf.random.uniform([2*rnn_size*rnn_layer, self.dim_hidden], -0.08,0.08),name='embed_state_W')
self.embed_state_b = tf.Variable(tf.random.uniform([self.dim_hidden], -0.08, 0.08), name='embed_state_b')
# description-embedding
self.embed_desc_W = tf.Variable(tf.random.uniform([self.vocabulary_size_d, self.input_embedding_size], -0.08, 0.08), name='embed_desc_W')
# encoder: RNN body
self.lstm_1_d = rnn_cell.LSTMCell(rnn_size, input_embedding_size, state_is_tuple=False)
self.lstm_dropout_1_d = rnn_cell.DropoutWrapper(self.lstm_1_d, output_keep_prob = 1 - self.drop_out_rate)
self.lstm_2_d = rnn_cell.LSTMCell(rnn_size, rnn_size, state_is_tuple=False)
self.lstm_dropout_2_d = rnn_cell.DropoutWrapper(self.lstm_2_d, output_keep_prob = 1 - self.drop_out_rate)
self.stacked_lstm_d = rnn_cell.MultiRNNCell([self.lstm_dropout_1_d, self.lstm_dropout_2_d])
# description state-embedding
self.embed_state_desc_W = tf.Variable(tf.random.uniform([2*rnn_size*rnn_layer, self.dim_hidden], -0.08,0.08),name='embed_state_desc_W')
self.embed_state_desc_b = tf.Variable(tf.random.uniform([self.dim_hidden], -0.08, 0.08), name='embed_state_desc_b')
# image-embedding 1
self.embed_image_W = tf.Variable(tf.random.uniform([dim_image, self.dim_hidden], -0.08, 0.08), name='embed_image_W')
self.embed_image_b = tf.Variable(tf.random.uniform([dim_hidden], -0.08, 0.08), name='embed_image_b')
# my code
# image-embedding 2
self.embed_image2_W = tf.Variable(tf.random.uniform([dim_image, self.dim_hidden], -0.08, 0.08), name='embed_image2_W')
self.embed_image2_b = tf.Variable(tf.random.uniform([dim_hidden], -0.08, 0.08), name='embed_image2_b')
# image-embedding 3
self.embed_image3_W = tf.Variable(tf.random.uniform([dim_image, self.dim_hidden], -0.08, 0.08), name='embed_image3_W')
self.embed_image3_b = tf.Variable(tf.random.uniform([dim_hidden], -0.08, 0.08), name='embed_image3_b')
# image-embedding 4
self.embed_image4_W = tf.Variable(tf.random.uniform([dim_image, self.dim_hidden], -0.08, 0.08), name='embed_image4_W')
self.embed_image4_b = tf.Variable(tf.random.uniform([dim_hidden], -0.08, 0.08), name='embed_image4_b')
# image-embedding 5
self.embed_image5_W = tf.Variable(tf.random.uniform([dim_image, self.dim_hidden], -0.08, 0.08), name='embed_image5_W')
self.embed_image5_b = tf.Variable(tf.random.uniform([dim_hidden], -0.08, 0.08), name='embed_image5_b')
# end my code
# score-embedding
self.embed_scor_W = tf.Variable(tf.random.uniform([dim_hidden, self.num_output], -0.08, 0.08), name='embed_scor_W')
self.embed_scor_b = tf.Variable(tf.random.uniform([self.num_output], -0.08, 0.08), name='embed_scor_b')