Rationalise outputs

[brenmous]

Output in UncoverML is very adhoc. From looking at the documentation it's not known what outputs are produced and where they will be stored. Some are related to config parameters, others are implicit. These need to be fixed and documented.

• cross validation outputs a 'scores.json' file and a 'results.hdf5' file - find the difference
• a scatter plot gets generated for cross validation implicitly (if not classifier)
• cross validation output is included in metadata but not other validation outputs (permutation importance, feature ranking)
• mixture of file formats e.g. crossval scores and feature ranks are JSON but permutation importance is CSV
• feature ranks get implicitly plotted
• geoio module is overloaded with outputting things not related to images (e.g. all of the above, exporting model files). File saving functions for these objects don't need to be in a separate module, just put them where the calculation happens.

[brenmous]

Output is now reduced to:

Always generated *_rawcovariates.csv - intersected targets and covariates *_rawcovariates_mask.csv - mask array for the above *_transformed_targets.csv - list of target values pre and post scaling/transformation *.model - the model, produced by the learn command and consumed by the predict command

Generated if cross validation enabled *_crossval_results.csv - target value with corresponding prediction from cross validation step *_crossval_scores.json - metrics from cross validation e.g. r^2 score

Generated if feature ranking enabled *_featureranks.json - covariates ordered by cross validation scoring, grouped by metric

Generated if covariate/target pickling turned on: features.pk (can be given any name in the config) targets.pk (can be given any name in the config)

With the exception of pickle files and the .model file (these can be given specific paths because they need to be read as well as written), all files are written to the output directory specified in the config