RT @KordingLab
So what went wrong? The authors apparently used the test data to select features. Obvious mistake. A reminder for everyone into ML: never use the test set for *anything* but testing. Only practical way to do so in medicine? Lock away the test set till algorithm is registered.