Please describe the inefficiency that you would like addressed. Is it causing a bottleneck?core/analysis.py has multiple functions that fit the model in order to do the analysis. It causes the model to be fit many times (unnecessarily) which is expensive for large models. It creates a post-tuning bottleneck.
multipredict() and replicate_model() both fit the model n times. This causes the model to be fit 2*n times, instead of the necessary n.
Describe the solution you'd like
Optimize the functions in analysis.py so that the model is fit a minimum of times but still retrieve the necessary information to create our desired analyses. Combine multipredict() and replicate_model() so that they both access the information produced from the n fits.
Describe alternatives you've considered
Haven't considered many. I think it's a clear path.
Additional context
When creating the fix, consider additional analyses we may perform (i.e importance graphs).
Please describe the inefficiency that you would like addressed. Is it causing a bottleneck?
core/analysis.py
has multiple functions that fit the model in order to do the analysis. It causes the model to be fit many times (unnecessarily) which is expensive for large models. It creates a post-tuning bottleneck.multipredict()
andreplicate_model()
both fit the modeln
times. This causes the model to be fit2*n
times, instead of the necessaryn
.Describe the solution you'd like Optimize the functions in
analysis.py
so that the model is fit a minimum of times but still retrieve the necessary information to create our desired analyses. Combinemultipredict()
andreplicate_model()
so that they both access the information produced from then
fits.Describe alternatives you've considered Haven't considered many. I think it's a clear path.
Additional context When creating the fix, consider additional analyses we may perform (i.e importance graphs).