Dear Alex,
Thanks you for posting your message!
Information about the features is currently only documented within comments in the source code since it isn't intended for users to read. The values in the features.txt are only intended for internal use and the way some of these features are calculated from the data are quite quick and dirty (e.g. t50 and auc)... but this works fine for it's purpose as a feature set for the machine learning. Nonetheless, you can find the information in the source code file eventer.m, which is here:
| % Description of features |
| % 2) t50: duration of event above half-maximum amplitude (approximates FWHM) |
| % 3) tdecay: duration of event post-peak above half-maximum amplitude (approximates decay half-life) |
| % 4) auc: area-under-curve of the event |
| % 5) skew: skewness metric for the distribution of the event sample points around the baseline |
| % 6) ampl: event amplitude calculated from template fitting |
| % 7) r: correlation coefficient of the template fitting |
| % 8) rstdev: standard deviation of residuals from the fit |
| % 9) before: time from preceding event (truncated) |
% 10) after: time to next event (truncated)
I personally would not report this info or use this information for anything, partly because of the quick and dirty calculations, and partly because the features will include all 'events' (whether or not they were classified as events by the machine learning). If I wanted to get the above features to document your results, I would get them from the averages documented in the summary.txt file in the ALL_events output folder. If I wanted properties of individual events (like amplitude or interevent interval of individual events), I would get them from the text files in the txt directory from that folder. For features other than those (e.g. rise, decay etc), I would load the event_data data file (I think the default is abf format at the moment) in your favourite general purpose electrophysiology data analysis software (my favourite is Stimfit).
Hope this helps. Was your question more out of curiosity or did you have specific purposes for the feature set?
Best
Andy