Reinforcement learning - help building a model

[ricardoV94]

I would suggest you check if the logp of your model matches a pure python/numpy implementation.

Is the normal likelihood appropriate for your data?

Perhaps 30 trials is way too little to inform your posterior?