Jonas Dehning

Add event_ndims to Transform class to fix shape errors in non-autobatching mode

3

Closes #297 ### Description of the problem: When not using auto_batching, the event_shape of the distributions and potentials must match. The reason is, when the log-likelihood is calculated, it is...

Error in transformed variables with manual model vectorization

4

There seems to be an error when using manual model vectorization for transformed variables when the event shape of the variable is larger than 1. A minimal example would be:...

Add stein variational gradient descent as sampling possibility, for increased performance

6

In PyMC3, stein variational gradient is already implemented, but it has to be tested how well it works/how biased it is for small number of particles. In addition, the optimal...

Add plotting functions for hierarchical models

1

Possible plots: - For one variable: the name of the regions as y axis, x-axis are CI intervals - For each region, table of histograms, one histogram/pdf per variable, like...

In order to parametrize better the hierarchical model, a non-centered parametrical distribution is better when the data is not very informative, see for example: https://docs.pymc.io/notebooks/Diagnosing_biased_Inference_with_Divergences.html or https://pdfs.semanticscholar.org/7b85/fb48a077c679c325433fbe13b87560e12886.pdf I implemented it...

Merge mobility analysis with master

2

@joaopn What is the status there? Otherwise I think that @semohr has also some code to retrieve the mobility data. I just want to avoid diverging development