[Feature request]: EMCEE Inference Proposal Out Of Bounds Enhancements

[TimothyWillard]

Label

enhancement, gempyor, inference

Priority Label

low priority

Is your feature request related to a problem? Please describe.

Specifically spurred by this comment:

flepiMoP/flepimop/gempyor_pkg/src/gempyor/inference_parameter.py

Line 205 in ad62f2d

# TODO: write a more granular method the return for a single parameter and correct the proposal like we did

In particular, when using bounded priors EMCEE seems to reject a large number of proposals for being out of bounds. An example of what this output looks like when running flepimop-calibrate is:

llik is '-48226937.65884703'
llik is '-88930698.4183363'
llik is '-106834157.6086256'
llik is '-34666500.54410425'
  0%|                                                                                                                                                                                        | 0/10 [00:00<?, ?it/s]OUT OF BOUND!!
llik is '-94323243.8932343'
OUT OF BOUND!!
OUT OF BOUND!!

Is your feature request related to a new application, scenario round, pathogen? Please describe.

No response

Describe the solution you'd like

There are two options for how to handle this (not sure of which one is more sound from a modeling perspective):

1. Implement a custom move that is aware of the modifier upper/lower bounds. See EMCEE's Moves documentation for the API to implement.
2. Transform the proposals from EMCEE before injecting them them into the modifiers in the InferenceParameters.inject_proposal. Would result in some challenges in doing the backwards accounting as the parameters that EMCEE records would not directly correspond to what was used by the model.