Open anupriyatripathi opened 4 years ago
@anupriyatripathi Note that most GNPS libraries and NIST were used for the training of CSIFingerID. so it should be identical in all cases, but it doesn't prove anything on the accuracy, just that the pipeline is consistent.
@lfnothias Note that this issue will not test the accuracy of CSI:FingerID on library spectra but on real data
I think that's fine as a sanity check. But if an experimental spectrum matches a library spectrum then are very similar to each other, so this is still an overfitting problem, so one has to interpret that with care.
..to compare classyfire output for features with both ms2 and csi smiles