Currently, DISCO accepts a single dataset and creates the validation dataset using validationSplit.
It would be useful to optionally connect separate training and validation datasets from the beginning. This is important for datasets that require group aware splitting such as lung ultrasound dataset.
This can be also helpful for certain test cases such as non-iid test case.
Currently, DISCO accepts a single dataset and creates the validation dataset using
validationSplit.It would be useful to optionally connect separate training and validation datasets from the beginning. This is important for datasets that require group aware splitting such as lung ultrasound dataset.
This can be also helpful for certain test cases such as non-iid test case.