Mutation prediction outstanding issues/updates

[jjc2718]

• use_pancancer should be false for all_other_cancers in process_data_for_gene() (i.e. there shouldn't be unused dummy/one-hot variables)
• Label filtering should happen after we take the intersection of samples between gene expression and mutation (this will make the proportions in 08_cell_line_prediction/download_data.ipynb match what we actually see when the scripts run)
• tcga_utilities should probably be renamed to something more general, or split
• CNV data for cell lines, in ccle_data_model _generate_labels()
• remove unknown/non-cancerous samples in load_sample_info()
• maybe try sklearn LogisticRegression with elastic net penalty rather than SGDClassifier
• save label proportions to plot AUPR baseline: CCLE drug response classification, stratified by cancer type #56 (comment)