Pertaining to #46
An interesting suggestion from a commenter at my blog is that some of the sub-monthly periods identified by machine learning is due to the non-uniform sampling rate, as monthly data violates the Nyquist criteria because of changing month lengths. He further suggested that non-uniform sampling can routinely extract signal at higher resolutions than uniform sampling, for example in variable star analysis.
This is a typical research article
"Variable Stars: which Nyquist Frequency ?"
L. Eyer and P. Bartholdi
And there are implications for machine learning in this chapter
"Data Mining and Machine-Learning in Time-Domain Discovery & Classification"
J.S.Bloom and J.W. Richards