def evaluate():
# Read vocabulary
vocab, rev_vocab = _load_vocabulary(FLAGS.vocab_fname)
with tf.Graph().as_default() as g:
#Enque data for evaluation
num_examples_per_epoch, tower_img_embedding, tower_context_length, \
tower_caption_length, tower_context_id, tower_caption_id, \
tower_answer_id, tower_context_mask, \
tower_caption_mask = enqueue(True)
tower_argmax = []
# Calculate the gradients for each model tower.
with tf.variable_scope(tf.get_variable_scope()) as scope:
for i in xrange(FLAGS.num_gpus):
with tf.device('/gpu:%d' % i):
with tf.name_scope('%s_%d' % (TOWER_NAME, i)) as scope:
inputs = [
tower_img_embedding[i],
tower_context_length[i],
tower_caption_length[i],
tower_context_id[i],
tower_caption_id[i],
tower_answer_id[i],
tower_context_mask[i],
tower_caption_mask[i]
]
net = CSMN(inputs, ModelConfig(FLAGS), is_training= False)
argmax = net.argmax
# Reuse variables for the next tower.
tf.get_variable_scope().reuse_variables()
# Keep track of the gradients across all towers.
tower_argmax.append(argmax)
argmaxs = tf.concat(tower_argmax, 0)
answer_ids = tf.concat(tower_answer_id, 0)
saver = tf.train.Saver(tf.global_variables())
# Build the summary operation based on the TF collection of Summaries.
summary_op = tf.summary.merge_all()
summary_writer = tf.summary.FileWriter(FLAGS.eval_dir, g)
#Don't evaluate again for the same checkpoint.
b_g_s = "0"
while True:
c_g_s = _eval_once(
saver, summary_writer, argmaxs, answer_ids, vocab,
rev_vocab, num_examples_per_epoch, b_g_s
)
b_g_s = c_g_s
if FLAGS.run_once:
break
time.sleep(FLAGS.eval_interval_secs)
def tower_loss(inputs, scope):
'''
Runs the data through the model and returns loss.
Inputs:
inputs - tf.data.Dataset iterator. It is dictionary containing input data.
scope - scope
Outputs:
loss - model loss
'''
net = CSMN(inputs, ModelConfig(FLAGS))
loss = net.loss
tf.summary.scalar(scope + 'loss', loss)
return loss
def evaluate():
vocab, rev_vocab = load_vocabulary(FLAGS.vocab_fname)
with tf.Graph().as_default() as g:
n, data_iter = dr.get_data('test2.txt', False)
tower_argmax = []
tower_ans_id = []
with tf.variable_scope(tf.get_variable_scope()) as scope:
for i in range(FLAGS.NUM_GPUS):
with tf.device('/gpu:%d' % i):
with tf.name_scope('%s_%d' % (TOWER_NAME, i)) as var_scope:
inp = data_iter.get_next()
net = CSMN(inp, ModelConfig(FLAGS), is_training=False)
argmax = net.argmax
tf.get_variable_scope().reuse_variables()
tower_argmax.append(argmax)
tower_ans_id.append(inp['answer_id'])
argmaxs = tf.concat(tower_argmax, 0)
answer_ids = tf.concat(tower_ans_id, 0)
saver = tf.train.Saver(tf.global_variables())
summary_op = tf.summary.merge_all()
summary_writer = tf.summary.FileWriter(FLAGS.eval_dir, g)
b_g_s = "0"
while True:
c_g_s = eval_once(saver, summary_writer, argmaxs, answer_ids,
vocab, rev_vocab, n, b_g_s)
b_g_s = c_g_s
if FLAGS.run_once:
break
time.sleep(FLAGS.eval_interval_secs)
def _tower_loss(inputs, scope):
net = CSMN(inputs, ModelConfig(FLAGS))
loss = net.loss
tf.summary.scalar(scope + 'loss', loss)
return loss