Supplementary files for generalization manuscript

Figures and raw data for manuscript: "Smaller models do not exhibit superior generalization performance". Includes the following files:
best_vs_smallest.zip: TSV files containing results of "best" vs. "smallest good" experiments (LASSO parameters and performance measurements), for each gene in Vogelstein et al. 2013 datasettcga_ccle_all_genes_figures.zip: Figures (performance curves, model sparsity, best vs. smallest good results) for TCGA <-> CCLE generalization, for each gene in Vogelstein et al. 2013 datasetcancer_type_all_genes_figures.zip: Figures for cancer type generalization experiments, for each gene/cancer type combination in the datasettcga_ccle_raw_results.zip: TSV files containing performance results (AUROC, AUPR) for TCGA -> CCLE generalization, for each gene on each of 8 cross-validation folds (4-fold CV x 2 random seeds), for each LASSO parameterccle_tcga_raw_results.zip: Same as above, for CCLE -> TCGA generalizationcancer_type_raw_results.zip: Same as above, for cancer type holdout generalization experiments on TCGAtcga_ccle_nn_raw_results.zip: Same as above, for TCGA -> CCLE generalization using a fully-connected neural network, with hidden layer size (rather than LASSO parameter) as the regularization axis

本数据集包含手稿《Smaller models do not exhibit superior generalization performance》的配套图表与原始实验数据。包含以下文件：
best_vs_smallest.zip：内含制表符分隔值（TSV）格式文件，记录了Vogelstein等人2013年数据集中每个基因的「最优模型」与「最优小模型」对比实验结果（包含套索回归（LASSO）参数与性能测量指标）。
tcga_ccle_all_genes_figures.zip：内含TCGA与CCLE之间泛化实验的相关图表，涵盖性能曲线、模型稀疏性、「最优模型」与「最优小模型」对比结果，对应Vogelstein等人2013年数据集中的每一个基因。
tcancer_type_all_genes_figures.zip：内含癌症类型泛化实验的相关图表，对应数据集中每一组基因-癌症类型组合。
tcga_ccle_raw_results.zip：内含TSV格式文件，记录了TCGA向CCLE泛化实验的性能结果（包含受试者工作特征曲线下面积（AUROC）、精确率-召回率曲线下面积（AUPR））：针对Vogelstein等人2013年数据集中的每个基因，在8折交叉验证（4折交叉验证×2次随机种子）下，对应每一个LASSO参数的实验结果。
ccle_tcga_raw_results.zip：与上述文件类似，对应CCLE向TCGA的泛化实验。
cancer_type_raw_results.zip：与上述文件类似，对应TCGA数据集上的癌症类型留存泛化实验。
tcga_ccle_nn_raw_results.zip：与上述文件类似，对应使用全连接神经网络开展的TCGA向CCLE泛化实验，此时以隐藏层大小（而非LASSO参数）作为正则化轴。