overlapping-ranges.py

import tensorflow as tf

window_size = 10
batch_size = 10
data = tf.range(100)

indices = tf.data.Dataset.range(90)

window_fn = lambda x: data[x:x+window_size]

train = indices.filter(lambda x:tf.less(tf.mod(x,10),7)).map(window_fn).batch(batch_size).repeat()
test  = indices.filter(lambda x:tf.logical_and(tf.greater_equal(tf.mod(x,10), 7), tf.less(tf.mod(x,10), 9))).map(window_fn).batch(batch_size).repeat()
validation = indices.filter(lambda x:tf.equal(tf.mod(x, 10),9)).map(window_fn).batch(batch_size).repeat() 

sess = tf.Session()

sess.run(tf.global_variables_initializer())


train_iterator = train.make_one_shot_iterator()
test_iterator  = test.make_one_shot_iterator()
validation_iterator = validation.make_one_shot_iterator()

next_train_element      = train_iterator.get_next()
next_test_element       = test_iterator.get_next()
next_validation_element = validation_iterator.get_next()

print ('train')
for i in range(10):
    print(sess.run(next_train_element))
print ('test')
for i in range(10):
    print(sess.run(next_test_element))
print ('validation')
for i in range(10):
    print(sess.run(next_validation_element))