What would be a good solution for you? - subsample data - remove duplicity (py27, py35) of the fmri_spm_nested examples and exercise them only in py27 (versus the other examples which are run in py35). - remove examples / reconsider what we want to include