[BUG] Can't mix and match BiphasicAxonMapModel

[mbeyeler]

I have a stimulus made exclusively of well-behaved BiphasicPulseTrain objects and the following model:

model = p2p.models.Model(
    spatial=p2p.models.AxonMapSpatial(n_axons=2000, yrange=(-8, 8), xrange=(-12, 12)),
    temporal=p2p.models.FadingTemporal(tau=200)
).build()
model.predict_percept(implant).play()

It works just fine. But switch out AxonMapSpatial for BiphasicAxonMapSpatial:

model = p2p.models.Model(
    spatial=p2p.models.BiphasicAxonMapSpatial(n_axons=2000, yrange=(-8, 8), xrange=(-12, 12)),
    temporal=p2p.models.FadingTemporal(tau=200)
).build()
model.predict_percept(implant).play()

results in ValueError: Cannot calculate temporal response, because stimulus/percept does not have a time component, which is clearly not true (/models/base.py, line 585)

[jgranley]

This is currently by design, but is something that we could consider changing (and that error certainly is not the most helpful).

Right now it doesn't completely make sense to use it with a temporal model, because the biphasic model is already accounting for lots of temporal information (pdur, freq) in its calculations, and its not clear how this should interact with another model that also uses those features.

This is similar to the issue with the Dynaphos model (it really isnt possible to completely separate out temporal and spatial components in either of these).

From the BiphasicAxonMapSpatial docstring:
.. note:: Using this model in combination with a temporal model is not currently supported and will give unexpected results

Should we consider allowing mix and match behavior? This might allow for the desired qualitative effects (e.g. phosphenes that scale with stim params and fade), but doesn't really make sense looking at the theoretic motivation / assumptions behind each model.

[jgranley]

Also wanted to add that enabling this behavior would not be hard. The bug arises because the BiphasicAxonMapModel is designed to output a singular brightest frame, which then does not work with the temporal model. A quick fix would be to just copy this frame at every point and then use the temporal model.