train_dataset, val_dataset = train_utils.get_datasets()
types = train_dataset.output_types
shapes = train_dataset.output_shapes

# LOOK HERE #
train_iterator = train_dataset.make_one_shot_iterator()
val_iterator = val_dataset.make_one_shot_iterator()
handle = tf.placeholder(tf.string, shape=[], name='dataset_handle')
iterator = tf.data.Iterator.from_string_handle(handle, types, shapes)

inputs, outputs = iterator.get_next()

model = MyModel(inputs)
loss = MyLoss(model.outputs, outputs)

train_step = ...
val_step = ...

with tf.Session() as sess:
    train_handle = sess.run(train_iterator.string_handle())
    val_handle = sess.run(val_iterator.string_handle())

    sess.graph.finalize()
    while True:
        sess.run(train_step, {handle: train_handle})
        sess.run(val_step, {handle: val_handle})

train_dataset, val_dataset = train_utils.get_datasets()
types = train_dataset.output_types
shapes = train_dataset.output_shapes

# LOOK HERE #
iterator = tf.data.Iterator.from_structure(types, shapes)
train_init_op = iterator.make_initializer(train_dataset)
val_init_op = iterator.make_initializer(val_dataset)

inputs, outputs = iterator.get_next()

model = MyModel(inputs)
loss = MyLoss(model.outputs, outputs)

train_step = ...
val_step = ...

with tf.Session() as sess:
    sess.graph.finalize()
    while True:
        sess.run(train_init_op)
        sess.run(train_step)

        sess.run(val_init_op)
        sess.run(val_step)

my_np_array = np.array(...)
def gen():
    while True: yield my_np_array

data_1 = tf.data.Dataset.from_generator(gen)
data_2 = tf.data.Dataset.from_csv(...)
data = tf.data.Dataset.zip((data_1, data_2))