mixture mc model

[belledon]

working as of 0dc523e however the temporal dynamics of belief change might need to be tweaked. Currently, the ramp object can change from congruent to incongruent at every time point. The change is dramatic at the moment of collision when the gt is incongruent. However there is a small diffusion towards the congruent prior shortly after due switch probability.

This has the interpretation that the observer has horrible short-term memory loss. This impact is pernicious since change the belief of mass after impact does not have much effect on likelihood suggesting that we might have to add more thank markov-chain structure to the change in belief. Equally as interesting attention to take hold here, with the logic that since there is not collision in the forseeablle future, the belief distribution for mass should not be rejuvinated.

yes will finally get to plotting density in log space.

[iyildirim]

I like the point with rejuvination.

Another possibility is to make transition probabilities asymmetric in the prior. congruent -> incongruent 0.1, incongruent -> congruent 0.001.

[belledon]

awesome suggestion!

After also re-adding the log-uniform proposal function we have some nice inference traces (note density is in log space now)

[iyildirim]

Looks good. I'm used to looking at the previous plots -- it'd help if you can align me to these new log plots when we talk tomorrow.

What is the variance in the prior's material mixtures now? I remember they used to be 4 with truncated Gaussians. What are they now? Thanks!