Normally when doing a fit to some time series data (e.g., a polynomial fit), functions will return an associated error with each fitted point. I'm now trying out scikit-learn's support vector regression (SVR) fitting instead, which doesn't have any such return. There is a handy function in scikit-learn called validation_score that can tell me the accuracy score of various fits, from which I select the best one. So maybe that's good enough? But as a scientist, I'm wondering if this will get slapped down in peer review of a publication.
What's the right way to propagate the errors of my time series data through the SVR fit?