combines several datasets according to weights, dynamically during training of a TF model.
It is composed of two classes:
ScheduleDataset: It encapsulates the same functionality as tf.data.experimental.sample_from_datasets, but weights are internally represented as tensors so that they are part of the computational graph and can be changed during training with ScheduleDataset_Callback.
ScheduleDataset_Callback: A callback class that, when used in .fit, dynamically changes the weights of a ScheduleDataset according to a schedule defined by the dataset_weights_fn function. See associated demo notebook for examples.
See the demo notebook for further details.