def run_epoch(session, m, vectors_data, labels_data, eval_op, verbose=False):
"""Runs the model on the given data."""
epoch_size = ((len(vectors_data) // m.batch_size) - 1) // m.num_steps
start_time = time.time()
costs = 0.0
iters = 0
state = m.initial_state.eval()
true_count = 0
for step, (x, y) in enumerate(read_data.ptb_iterator(vectors_data, labels_data, m.batch_size,
m.num_steps)):
cost, state, out_logits, _ = session.run([m.cost, m.final_state, m.out_logits, eval_op],
{m.input_data: x,
m.targets: y,
m.initial_state: state})
costs += cost
iters += m.num_steps
labels = tf.reshape(y, [-1])
correct = tf.nn.in_top_k(tf.cast(out_logits, tf.float32), tf.cast(labels, tf.int32), 1)
eval_correct = tf.reduce_sum(tf.cast(correct, tf.int32))
true_count += session.run(eval_correct)
if verbose and step % (epoch_size // 10) == 10:
print("%.3f perplexity: %.3f speed: %.0f wps" %
(step * 1.0 / epoch_size, np.exp(costs / iters),
iters * m.batch_size / (time.time() - start_time)))
precision = true_count/(iters*m.batch_size)
print("precision:")
print(precision)
return np.exp(costs / iters)
def run_epoch(session, m, vectors_data, labels_data, eval_op, summary_writer, verbose=False):
"""Runs the model on the given data."""
epoch_size = ((len(vectors_data) // m.batch_size) - 1) // m.num_steps
start_time = time.time()
costs = 0.0
iters = 0
state = m.initial_state.eval()
true_count = 0
for step, (x, y) in enumerate(read_data.ptb_iterator(vectors_data, labels_data, m.batch_size,
m.num_steps)):
cost, state, out_logits, merged1, _ = session.run([m.cost, m.final_state, m.out_logits, m.merged, eval_op],
{m.input_data: x,
m.targets: y,
m.initial_state: state})
if (m.num_steps != 1):
summary_writer.add_summary(merged1,step)
costs += cost
iters += m.num_steps
labels = tf.reshape(y, [-1])
correct = tf.nn.in_top_k(tf.cast(out_logits, tf.float32), tf.cast(labels, tf.int32), 1, name = "correct")
eval_correct = tf.reduce_sum(tf.cast(correct, tf.int32))
true_count += session.run(eval_correct)
if verbose and step % (epoch_size // 10) == 10:
print("%.3f perplexity: %.3f speed: %.0f wps" %
(step * 1.0 / epoch_size, np.exp(costs / iters),
iters * m.batch_size / (time.time() - start_time)))
with tf.name_scope("perplexity") as scope:
perplexity = np.exp(costs / iters)
perplexity_summary = tf.scalar_summary("perplexity",perplexity)
with tf.name_scope("precision") as scope:
precision = true_count/(iters*m.batch_size)
print("precision:")
print(precision)
accuracy_summary = tf.scalar_summary("accuracy", precision)
merged_2 = tf.merge_summary([perplexity_summary,accuracy_summary])
merged2 = session.run(merged_2)
summary_writer.add_summary(merged2)
return perplexity