def set_graph(self, embedding_matrix):
with self.graph.as_default():
# tf Graph input
tf.set_random_seed(1)
self.x = tf.placeholder(tf.int32, shape=(self.cfg.bsize, self.cfg.max_seq_len), name="x")
self.y = tf.placeholder(tf.float32, shape=(self.cfg.bsize,6), name="y")
self.em = tf.placeholder(tf.float32, shape=(embedding_matrix.shape[0], embedding_matrix.shape[1]), name="em")
self.keep_prob = tf.placeholder(dtype=tf.float32, name="keep_prob")
self.output = model_baseline(self.em,self.x,self.keep_prob) #self.cfg.bsize
with tf.variable_scope('logits'):
self.logits = self.output
with tf.variable_scope('loss'):
self.loss = binary_crossentropy(self.y,self.logits)
self.cost = tf.losses.log_loss(predictions=self.logits, labels=self.y)
(_, self.auc_update_op) = tf.metrics.auc(predictions=self.logits,labels=self.y,curve='ROC')
self.global_step = tf.Variable(0, trainable=False)
self.learning_rate = tf.train.exponential_decay(self.cfg.lr, self.global_step,self.cfg.decay_steps, self.cfg.decay, staircase=True)
with tf.variable_scope('optim'):
self.optimizer = tf.train.RMSPropOptimizer(learning_rate=self.learning_rate).minimize(self.loss,global_step=self.global_step)
with tf.variable_scope('saver'):
self.saver = tf.train.Saver(max_to_keep=self.cfg.max_models_to_keep)
X_valid = X[:split_at]
Y_valid = Y[:split_at]
X_train = X[split_at:]
Y_train = Y[split_at:]
graph = tf.Graph()
with graph.as_default():
with tf.variable_scope('Input'):
x = tf.placeholder(dtype=tf.int32,shape=(None,maxlen))
y = tf.placeholder(dtype=tf.float32,shape=(None,6))
logits =
loss = binary_crossentropy(y, logits)
cost = tf.losses.log_loss(labels=y, predictions=logits)
optimizer = tf.train.RMSPropOptimizer(learning_rate=0.01).minimize(loss)
(_, auc_update_op) = tf.contrib.metrics.streaming_auc(
predictions=logits,
labels=y,
curve='ROC')
train_iters = len(X_train) - 2*bsize
valid_iters = len(X_valid) - 2*bsize
with tf.Session(graph=graph) as sess:
init = tf.global_variables_initializer()
sess.run(init)
for epoch in range(EPOCHS):