produce detection lists from predictsignauxfaibles
select classification thresholds using one of three methods, with notebook 03-select-thresholds.ipynb as a tutorial + starting point for selecting a threshold right after a prediction file is issued
Minor modifications:
added a CLI option "--explain" to predictsignauxfaibles as computing explanation fields takes a lot of time
added a CLI option "--save_model" which saves model objects (including its weights) using pickle
To be added in subsequent PR:
using validation set distinct from test set in 03-select-thresholds.ipynb
manually adjust for class imbalance on data loaded from a csv file in 03-select-thresholds.ipynb
This PR adds functionalities to:
predictsignauxfaibles03-select-thresholds.ipynbas a tutorial + starting point for selecting a threshold right after a prediction file is issuedMinor modifications:
predictsignauxfaiblesas computing explanation fields takes a lot of timepickleTo be added in subsequent PR:
03-select-thresholds.ipynb03-select-thresholds.ipynb