Summary
We currently refer to parameter estimates calculated from the z-scored mixing matrix and the un-normalized optimally combined data as "betas" and parameter estimates calculated from the z-scored mixing matrix and the z-scored optimally combined data (followed by numpy.arctanh and variance rescaling) as "weights". In general, I believe most people use the word "beta" to refer to standardized parameter estimates instead of unstandardized ones (originally brought up in #448).
To be honest, I think we should also just simplify computefeats2 to just z-score the data and mixing matrix, without the weird extra steps (cropping values >0.999 or <-0.999, then numpy.arctanh, then variance rescaling). It's not clear to me if we want standardized parameter estimates (betas) or z-statistics from a significance test. If we want the latter, I think we can simplify the code.
Related to #178.
Summary
We currently refer to parameter estimates calculated from the z-scored mixing matrix and the un-normalized optimally combined data as "betas" and parameter estimates calculated from the z-scored mixing matrix and the z-scored optimally combined data (followed by
numpy.arctanhand variance rescaling) as "weights". In general, I believe most people use the word "beta" to refer to standardized parameter estimates instead of unstandardized ones (originally brought up in #448).To be honest, I think we should also just simplify
computefeats2to just z-score the data and mixing matrix, without the weird extra steps (cropping values >0.999 or <-0.999, thennumpy.arctanh, then variance rescaling). It's not clear to me if we want standardized parameter estimates (betas) or z-statistics from a significance test. If we want the latter, I think we can simplify the code.Related to #178.