Hi Pieter, Thanks for pointing this PR out. That certainly fixes the immediate problem with the inconsistent print statements that I highlighted in my original message.
It doesn't address the more fundamental problem, though, which is that the default behavior is to represent the polynomial in this rescaled form, which unnecessarily privileges numerical accuracy over ease of use and consistency with standard usage. I realize that it has been this way for a while, but multiple GitHub issues indicate that it causes confusion, which suggests that the issue should be addressed more meaningfully. (As an aside, the whole module uses a nonstandard definition of weights that also causes confusion.) I would expect the confusion to be compounded if Polynomial.fit (with its cousins) adopts the option to return the covariance matrix (which I recommend), since this will also depend on the scaling. I think it's great to provide the *option* to scale the domain, especially for things like Chebyshev polynomials, where the domain typically needs rescaling, anyway. But a user who wants to fit data with y = a[0] + a[1] * x + a[2] * x**2 should, IMO, get back the best-fit coefficients for the equation as originally formulated by default, not in the form that is most convenient for the numerical analyst. Cheers, Steve _______________________________________________ NumPy-Discussion mailing list -- numpy-discussion@python.org To unsubscribe send an email to numpy-discussion-le...@python.org https://mail.python.org/mailman3/lists/numpy-discussion.python.org/ Member address: arch...@mail-archive.com
