Maximizing info about only some parameters

[DavidBrainard]

There can be cases where a psychometric function has numerous parameters, and we want to allow all to vary to account for the data, but we may only care about establishing a few of them to high precision. For example, we might want to know the threshold parameter and not care about the value of the lapse parameter. But still allow the lapse parameter to vary in the fit.

I think this could be handled by marginalizing over the expected posteriors, when computing the expected entropies. It might be possible to figure out a slick way to specify this, and to compute it fast from the discretized posterior (because it should just be summing over some dimensions).

This would be something to think about on a rainy day, or if it becomes pressing for an application.