Mercurial > repos > bgruening > sklearn_generalized_linear
diff model_prediction.py @ 37:913bf1c4c7bb draft
"planemo upload for repository https://github.com/bgruening/galaxytools/tree/master/tools/sklearn commit ea12f973df4b97a2691d9e4ce6bf6fae59d57717"
author | bgruening |
---|---|
date | Sat, 01 May 2021 01:18:07 +0000 |
parents | 616a241c5b37 |
children | fe181d613429 |
line wrap: on
line diff
--- a/model_prediction.py Tue Apr 13 22:54:21 2021 +0000 +++ b/model_prediction.py Sat May 01 01:18:07 2021 +0000 @@ -63,7 +63,8 @@ if hasattr(main_est, "config") and hasattr(main_est, "load_weights"): if not infile_weights or infile_weights == "None": raise ValueError( - "The selected model skeleton asks for weights, " "but dataset for weights wan not selected!" + "The selected model skeleton asks for weights, " + "but dataset for weights wan not selected!" ) main_est.load_weights(infile_weights) @@ -72,7 +73,9 @@ # tabular input if input_type == "tabular": header = "infer" if params["input_options"]["header1"] else None - column_option = params["input_options"]["column_selector_options_1"]["selected_column_selector_option"] + column_option = params["input_options"]["column_selector_options_1"][ + "selected_column_selector_option" + ] if column_option in [ "by_index_number", "all_but_by_index_number", @@ -122,9 +125,13 @@ pred_data_generator = klass(fasta_path, seq_length=seq_length) if params["method"] == "predict": - preds = estimator.predict(X, data_generator=pred_data_generator, steps=steps) + preds = estimator.predict( + X, data_generator=pred_data_generator, steps=steps + ) else: - preds = estimator.predict_proba(X, data_generator=pred_data_generator, steps=steps) + preds = estimator.predict_proba( + X, data_generator=pred_data_generator, steps=steps + ) # vcf input elif input_type == "variant_effect": @@ -135,7 +142,9 @@ if options["blacklist_regions"] == "none": options["blacklist_regions"] = None - pred_data_generator = klass(ref_genome_path=ref_seq, vcf_path=vcf_path, **options) + pred_data_generator = klass( + ref_genome_path=ref_seq, vcf_path=vcf_path, **options + ) pred_data_generator.set_processing_attrs()