[Question] Using qNegIntegratedPosteriorVariance as a value function in KG

[LLipschitz]

Is it possible to use qNegIntegratedPosteriorVariance as a value function in Knowledge Gradient acquisition function? Note that it takes as required arguments model and mc_points.

I tried including mc_points in the __init__ method of KG but get the following error. Didn't mean to be cryptic, but please let me know if you need more information.

value_function(model, objective, sampler, project, EI_target, PI_target, UCB_conf, mc_points_imspe)
    530         base_value_function = UpperConfidenceBound(model=model, beta=UCB_conf)
    531     elif mc_points_imspe is not None:
--> 532         base_value_function = qNegIntegratedPosteriorVariance(model=Model, mc_points=mc_points_imspe, objective=objective)
    533     else:
    534         # print('using posterior mean as value function')

~\AppData\Roaming\Python\Python37\site-packages\botorch\acquisition\active_learning.py in __init__(self, model, mc_points, sampler, objective, X_pending)
     71                 have not yet been evaluated.
     72         """
---> 73         super().__init__(model=model, objective=objective)
     74         if sampler is None:
     75             # If no sampler is provided, we use the following dummy sampler for the

~\AppData\Roaming\Python\Python37\site-packages\botorch\acquisition\analytic.py in __init__(self, model, objective)
     42             objective: A ScalarizedObjective (optional).
     43         """
---> 44         super().__init__(model=model)
     45         if objective is None:
     46             if model.num_outputs != 1:

~\AppData\Roaming\Python\Python37\site-packages\botorch\acquisition\acquisition.py in __init__(self, model)
     31         """
     32         super().__init__()
---> 33         self.add_module("model", model)
     34 
     35     def set_X_pending(self, X_pending: Optional[Tensor] = None) -> None:

~\Anaconda3\lib\site-packages\torch\nn\modules\module.py in add_module(self, name, module)
    371         if not isinstance(module, Module) and module is not None:
    372             raise TypeError("{} is not a Module subclass".format(
--> 373                 torch.typename(module)))
    374         elif not isinstance(name, torch._six.string_classes):
    375             raise TypeError("module name should be a string. Got {}".format(

TypeError: botorch.models.model.Model is not a Module subclass

[Balandat]

Hmm it seems like here you're passing a class not a class instance to the acquisition function?

--> 532         base_value_function = qNegIntegratedPosteriorVariance(model=Model

The Knowledge Gradient essentially uses the posterior mean as the value function. We have qMultiStepLookahead that is quite modular and allows you to specify the value function class (as well as a factory function to construct the arguments for its constructors from the data). I would start there.

In either case, if you can share a full repro it will be easier for us to debug.