4 ms·
Machine learning model helps characterize compounds for drug discovery
- blackbear_ 6y agoThese results aren't very reliable. Shows potential, but it definitely needs better methodology and perhaps more data. They trained a decision tree on 36 examples with only 8 positives and tested on 13. Without cross-validation or, even better, nested resampling. All leaves of the decision tree but one contain one or two examples.
- barumi 6y agoI don't understand why the "machine learning" marketing label is relevant in this discussion. Plain old classification problems are very common in data analysis, and so is parameter identification. It could be argued that these statistical techniques are the bread and butter of science. So, what's new or noteworthy on scientists using a method to analyse data? It sounds like someone was hell-bent on capitalizing on the "machine learning" marketing tag without paying much attention to it.