Using variable amount of hyperparameters based on conditions

[Anselmoo]

In AutoKeras, various types of models with variable-length switching on/off of ImageAugmentation can be automatically applied.

@haifeng-jin can you please provide some details what has to be done to working with variable hyperparameters based on conditions?

 def run_trial(self, trial, x_data, y_data, **kwargs):
        self.trial = trial
        hp = trial.hyperparameters
        model_choice = hp.Choice("model", ["model_one", "model_two"])

        if model_choice == "model_one":
            a = hp.Float("a", min_value=-2, max_value=+2, default=1.0)
        elif model_choice == "model_two":
            b = hp.Float("b", min_value=-2, max_value=+2, default=1.0)
            c = hp.Float("c", min_value=-2, max_value=+2, default=1.0)

In the suggested code, the export in the json because sooner or later wrong. It is a little bit hard to follow up in AutoKeras how that was actually implemented in the block and engine sections.

• Or would it be particular possible to import AutoKeras back into Keras-Tuner for this task?

Similiar discussion are already started on: https://stackoverflow.com/questions/68424234/dependent-hyperparameters-with-keras-tuner