Pass additional kwargs to model call function

[jvierstra]

Allow passing of additional keyword arguments to model forward (call) function in deep_lift_shap. This allows one to selectively modify outputs for interpretation (such as applying an exponential transform on prediction logins) and properly compute contributions considering non-linearities.

[jmschrei]

I'm not sure I understand -- can't you use the args param for that?

If you want to modify the outputs of a model before running deep_lift_shap on it I'd recommend creating a short wrapper. It's a few extra lines at the beginning but way cleaner and more flexible, e.g.

from bpnet.chrombpnet import _Exp


class Wrapper(torch.nn.Module):
    def __init__(self, model):
        super().__init__()
        self.model = model
        self.exp = _Exp()

    def forward(self, X):
        return self.exp(self.model(X))

where you use _Exp from https://github.com/jmschrei/bpnet-lite/blob/master/bpnetlite/chrombpnet.py#L20 and then also register it when you use DeepLiftShap.