Robert Kern

Sorry, I misphrased. I did not mean to suggest that there _ought_ to be an `abs()` within the power mean implementation. Just that it is not necessarily the case that...

> I don't understand this. It's exactly what's in the Wiki page for p=2, isn't it? Well, not for complexes. The wiki article does not address complexes. The popular convention...

The MATLAB `rms()` definitely does mention complex values and uses the absolute value. See the "More About" section on that page. @endolith also gave several other literature citations using the...

[I was on the other end of that one](https://mail.python.org/archives/list/[email protected]/thread/TIBFUMZJJAPGFPFFY6GX4Z3DM4BEO3H6/#TOWY74Q4R3NS3C2KPNX7VOX3GSF6YOF3) but I was eventually convinced. C.f. numpy/numpy#1236

> It would be possible to add an `abs=True` option that, if it is False, returns the "pseudo-rmse" in analogy to the definition of the pseudo-variance for complex random variables....

Yeah, it's a known (to some) design flaw. I usually have a `_zero = Constant(0.0)` trait or whatever for such cases. It's working as designed; it's just a bad design,...

Unfortunately, it's the last point that makes this difficult for Traits to do internally, too. The dependency chain (well, graph) is generally distributed between several objects, and the graph mutates...

Known. `List` traits are the only ones where the `*_items` is also synced. So more of a feature request than a bug.

This is probably not a good starter issue. Changing the type promotion behavior is a policy decision rather than a technical code fix.

It's working as designed. `copy=True` was never intended to imply a deepcopy of `dtype=object` arrays, just that you would get a new `ndarray` object with the data copied over, as...