Support `botorch_modular.BoTorchModel.feature_importances`

Hello, thanks for the previous help. Now I face a new problem. I am trying to use a new generation strategy. However, the feature importance plot can not be drawn. The following is the code of the generation strategy.

gs = GenerationStrategy(
    steps=[
        # Quasi-random initialization step
        GenerationStep(
            model=Models.SOBOL,
            num_trials=5,  # How many trials should be produced from this generation step
        ),
        # Bayesian optimization step using the custom acquisition function
        GenerationStep(
            model=Models.BOTORCH_MODULAR,
            num_trials=-1,  # No limitation on how many trials should be produced from this step
            # For `BOTORCH_MODULAR`, we pass in kwargs to specify what surrogate or acquisition function to use.
            model_kwargs={
                "surrogate": Surrogate(SingleTaskGP),
                "botorch_acqf_class": qNoisyExpectedImprovement
            },
        ),
    ]
)

The following figure is the error: Here, the singleTaskGp is used as the surrogate model, is it not a Gaussian process?

[Balandat]

Could you trigger the “Show CI” selector on the plot as well? It would be good to also understand how well calibrated these models are.

[yuanquan010]

Actually, I used 100 trials. The following figure shows the code. Based on the Cross-validation plot, it looks okay.

Another thing worth mentioning is that when I come back to use models.GPEI in the generation strategy, as shown in the following figure.

I can get the feature importance, but the GP model can not fit well. In the above two experiments, only the generation strategy is changed. Is there some internal connection in these errors?