Help! Can't get started with BoTorch models about ax

[Jgmedina95]

Thanks! ive edited the issue for better formatting.

[bernardbeckerman]

Hi there, I believe by following the tutorial (https://botorch.org/tutorials/custom_botorch_model_in_ax) more closely you can achieve a successful get_next_trial call. E.g., if you replace your third cell with:

class ExactGPModel(gpytorch.models.ExactGP, GPyTorchModel):
    
    _num_outputs = 1 

    def __init__(self, train_X, train_Y,**kwargs):
        super().__init__(train_X, train_Y.squeeze(-1), GaussianLikelihood(), **kwargs)
        self.mean_module = gpytorch.means.ConstantMean()
        self.covar_module = gpytorch.kernels.ScaleKernel(gpytorch.kernels.RBFKernel())
        self.to(train_X)
        
    def forward(self, x):
        mean_x = self.mean_module(x)
        covar_x = self.covar_module(x)
        return gpytorch.distributions.MultivariateNormal(mean_x, covar_x)

gs = GenerationStrategy(
    steps=[
        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(ExactGPModel),
                
                "botorch_acqf_class": qExpectedImprovement,
                },
        ),
    ]
)

the rest of your notebook should execute as expected. HTH!

[Jgmedina95]

Thanks a lot! this is working now :) Asking in the Botorch repo, apparently the .from_botorch method is very faulty and could be the reason it wasn't working. Thanks for your time!

[bernardbeckerman]

Thanks for reporting this - taking a look!

[bernardbeckerman]

@Jgmedina95 Thanks for the clarification! Though, I'm having a little trouble reproducing your issue. A brief issue with your code is that ExactGPModel.__init__ takes args train_X and train_Y but its contents refer to train_x and train_y (lower-case x and y), which may be causing the issue you're running into. If that doesn't fix, please repost your code with all imports and intermediate quantities (e.g., train_features and train_final_label) defined so that I can rerun with your precise settings. HTH!

[Jgmedina95]

Thanks for looking into this. I missed that typo while writing the issue. I just reproduced the same errors. The code gets a little big quickly so I'm uploading a notebook and the dataset. The notebook is in JSON format, as github doesn't let me upload it in .ipynb format.
modified_features.csv
reproduction_error_botorch.ipynb.json