eval_output = tf.sigmoid(tf.matmul(eval_h_fc5, w_fc6) + b_fc6)
with sess.as_default():
loss = tf.reduce_mean(-( (1 - y)*tf.log(1 - output) + y * tf.log(output) ))
optimizer = tf.train.AdamOptimizer(LEARNING_RATE)
#optimizer = tf.train.AdagradOptimizer(5e-3)
grads_and_vars = optimizer.compute_gradients(loss)
train_step = optimizer.apply_gradients(grads_and_vars)
sess.run(tf.initialize_all_variables())
epoch = 0
while True:
data_helper.shuffle_train_ins()
sys.stdout.write("Epoch %d\n#################################################" % epoch)
i = 0
training_loss = 0
training_ins_sz = len(data_helper.train_ins)
offset = 0
while offset < training_ins_sz:
if i % 50 == 0:
[ _w_embedding ] = sess.run([w_embedding])
eval_ins = data_helper.get_eval_ins_embedding(_w_embedding, EMBEDDING_SIZE)
with sess.as_default():
loss = tf.reduce_mean(-(
(1 - y) * tf.log(1 - output) + y * tf.log(output)))
optimizer = tf.train.AdamOptimizer(LEARNING_RATE)
#optimizer = tf.train.AdagradOptimizer(5e-3)
grads_and_vars = optimizer.compute_gradients(loss)
train_step = optimizer.apply_gradients(grads_and_vars)
sess.run(tf.initialize_all_variables())
epoch = 0
while True:
data_helper.shuffle_train_ins()
sys.stdout.write(
"Epoch %d\n#################################################"
% epoch)
i = 0
training_loss = 0
training_ins_sz = len(data_helper.train_ins)
eval_ins_sz = len(data_helper.eval_ins)
eval_batch_sz = 3000
offset = 0
while offset < training_ins_sz:
if i % 50 == 0:
