Different models can have wildly different distributions (see #253). That implies that analysis-pipeline should have an optional step at the end where it generates 100 (configurable, of course) puzzles and then displays a histogram and key stats.
In a perfect world it would be possible to run the pipeline and then do the difficulties with the new model immediately. Currently that requires a new build. Maybe a method in sudoku to load a model at runtime from a JSON file?
Different models can have wildly different distributions (see #253). That implies that analysis-pipeline should have an optional step at the end where it generates 100 (configurable, of course) puzzles and then displays a histogram and key stats.
In a perfect world it would be possible to run the pipeline and then do the difficulties with the new model immediately. Currently that requires a new build. Maybe a method in sudoku to load a model at runtime from a JSON file?