For the holdout cancer type analyses (e.g. #66), we wanted to look at which cancer types tend to perform well/poorly as the holdout set.
In the box plot, positive values mean the classifiers performed better on the training data than on the holdout cancer type, and vice-versa. The results largely make sense: TGCT and SARC are non-carcinomas so it's not surprising that generalization was poor, THCA only has classifiers for 2 genes and one is very undersampled, etc.
For the holdout cancer type analyses (e.g. #66), we wanted to look at which cancer types tend to perform well/poorly as the holdout set.
In the box plot, positive values mean the classifiers performed better on the training data than on the holdout cancer type, and vice-versa. The results largely make sense: TGCT and SARC are non-carcinomas so it's not surprising that generalization was poor, THCA only has classifiers for 2 genes and one is very undersampled, etc.