TF

Import data

1. Dataset & iterator

tf.data.Dataset = a sequence of elements

tf.data.Iterator = extract elements from a dataset

2. Load data

dataset1 = tf.data.Dataset.from_tensor_slices(tf.random_uniform([4, 10]))
print(dataset1.output_types)  # ==> "tf.float32"
print(dataset1.output_shapes)  # ==> "(10,)"


dataset2 = tf.data.Dataset.from_tensor_slices(
       (
              tf.random_uniform([4]),
              tf.random_uniform([4, 100], maxval=100, dtype=tf.int32)
       )
)
print(dataset2.output_types)  # ==> "(tf.float32, tf.int32)"
print(dataset2.output_shapes)  # ==> "((), (100,))"

3. Transformation

dataset = (tf.data.Dataset.from_tensor_slices((tf.constant(filenames), tf.constant(labels)))
    .shuffle(num_samples)  # whole dataset into the buffer ensures good shuffling
    .map(parse_fn, num_parallel_calls=params.num_parallel_calls)
    .map(train_fn, num_parallel_calls=params.num_parallel_calls)
    .batch(params.batch_size)
    .prefetch(1)  # make sure you always have one batch ready to serve

4. Iterator

dataset = tf.data.Dataset.range(100)
iterator = dataset.make_one_shot_iterator()
next_element = iterator.get_next()

for i in range(100):
  value = sess.run(next_element)
  assert i == value

dataset = tf.data.Dataset.range(100)
iterator = dataset.make_initializable_iterator()
next_element = iterator.get_next()

# Initialize an iterator over a dataset with 10 elements.
sess.run(iterator.initializer)
for i in range(100):
  value = sess.run(next_element)
  assert i == value

Eager Execution

from __future__ import absolute_import, division, print_function

import tensorflow as tf

tf.enable_eager_execution()

tf.executing_eagerly()
>> True

Some functions

get_global_step

global_step = tf.train.get_global_step()

tf.train.Saver

last_saver = tf.train.Saver() # will keep last 5 epochs

with tf.Session() as sess:
    ...
    last_saver.restore(sess, restore_from)
    ...
    last_saver.save(sess, last_save_path, global_step=epoch + 1)

best_saver = tf.train.Saver(max_to_keep=1)  # only keep 1 best checkpoint

with tf.Session() as sess:
    ...
    best_save_path = best_saver.save(sess, best_save_path, global_step=epoch + 1)

Output

to screen

with tf.Session() as sess:
    sess.run(init)

    for step in range(11):
        sess.run(train_op, feed_dict=feed_dict)

        if step % 5 == 0:
            loss = sess.run(loss_op, feed_dict=feed_dict)
            print(loss)

tensorboard

with tf.name_scope('Layer'):
    ...
    outputs = activation_function(formula)
    tf.summary.histogram(name=filename, values=outputs)

with tf.name_scope('Loss'):
    ...
    loss = tf.reduce_mean(tf.square(y_feeds - output_layer))
    tf.summary.scalar('loss', loss)

merged = tf.summary.merge_all()

with tf.Session() as sess:
    sess.run(init)

    writer = tf.summary.FileWriter("TensorBoard/", graph=sess.graph) # add graph

    for step in range(11):
        sess.run(train, feed_dict=feed_dict)

        if step % 5 == 0:
            result = sess.run(merged, feed_dict=feed_dict)
            writer.add_summary(result, step) # add histogram & scalar