FEATURE - More control over training/validation set during experiment

[KaikeWesleyReis]

1) I think Talos should add this feature:

Implement some more parameters for the Scan function:

• x_val: set different from x, to use as validation only!
• y_val:set different from y, to use as validation only!
• validation_split: if x_val / y_val is None, define x split to validate your model

The idea here is to allow the user more control over the experiment, allowing to choose a database to use as validation or the portion of x / y to use as validation. These options are equivalent to the options in `keras```.

2) Why?

In models development it's necessary to have a higher control over the train/validation set to ensure the reproducibility of the experiment.

For example: let's say I created a model A with certain hyper parameters and considering defined train/validation set (keras CIFAR-10 is a good example). However, I would like to evaluate other hyper parameters for my architecture using talos. To do this, it's necessary to fix, as model A, the original train/validation set for my experiment to give the opportunity to compare and to ensure the reproducibility of the code on other machines.

3) I believe this feature is

• [ ] critically important
• [X] must have
• [ ] nice to have

4) Given the chance, I'd be happy to make a PR for this feature

• [ ] definitely
• [X] possibly
• [ ] unlikely

But to a PR, I need to understand how/where talos does train validation split, if you show me I would glad help this package improvement (it is awesome).

[github-actions[bot]]

Welcome to Talos community! Thanks so much for creating your first issue :)

[mikkokotila]

Scan() already supports this. Have a look at https://autonomio.github.io/talos/#/Scan.

Closing this as resolved. Feel free to open new issue if anything.