m-v-nikolov / karibacalibrationdemo Goto Github PK

Currently both collections utilize the information from fit_entry as specified in kariba_model.py; however, the fit_entry info is written in different formats in all_fits and best_fits; furthermore some of the fit_entries attributes are written weighted/unweighted when viewed as a best_fit or respectively all_fits members. Introduce a fit_entry object that would handle the interaction between all_fits, best_fits and each model properties.

pop_bubbles_map_display takes too many parameters that can be programmatically specified at config time. Need to refactor correspondingly.

All paths can be specified programmatically at config time (e.g. input json); no need for hard-coding.

Convert various traces and heat maps currently displayed as pre-formatted png figures to d3js json based objects

When objective functions terms are computed (as opposed to loaded from files and only re-weighted), a new model is always created in gazetteer.json, even though that model may already have existed.

by weights value (e.g. increasing on clinical cases weight).

Remove dead code and abstract functionality in separate object based modules.

Possible culprit is in kariba_model.py where as a premature optimization a penalty is not computed and set directly to 0 if its weight in the objective function is zero; the logic carries over the resulting 0 in the fit_terms collection that is written as a json after calibration.

Premature optimization is the root of all evil (in programming).