Fix transformation fit call with default variables

[wildart]

Fix #594

[kleinschmidt]

Thanks, looks like we missed this one before. Did you double check that the other cases are handled in the same way?

[kleinschmidt]

I wonder if we should back port this to the 0.32 series too (as with #565)

[wildart]

I made dims=1 to be the default value for vectors. Using nothing doesn't make sense in such case.

[kleinschmidt]

Actually I think it's better to leave the default value as nothing, unless we want to make the dims= kwarg optional for vectors.

[kleinschmidt]

The logic being that eventually (aka on the next breaking change release) there will be an error if dims is not specified.

[wildart]

there will be an error if dims is not specified.

After all, there is only one dimension in a vector. In future, this will be a default argument with value 1. It is not that, we cannot figure out the parameter type, so the dims is required. We still rely on multiple dispatch. I understand necessity of the dimension selection for a matrix argument, but not for a vector. mean doesn't require dims parameter for the vectors nor many functions. Why should any function operating on vectors require this parameter?

[nalimilan]

For consistency with Statistics, where cov and cor treat columns as variables, it would make sense to switch the default to dims=1 at some point after removing the deprecation. Then the default would also be correct for vectors. I'm hesitant about whether we should make this change for vectors now in anticipation of the general switch or not.

[wildart]

The functions with the matrix parameter have the warnings about dims parameter usage. But is this rely necessary to inform user use dims=1 parameter with the vector input? Will dims be a positional non-default parameter?

[nalimilan]

No, dims will remain a keyword argument AFAICT, that's consistent with Base, Statistics and Distances. I don't have a strong opinion about vectors, so let's hear what others think.

[kleinschmidt]

If dims=1 is going to be the default everywhere I have no problem making it the default for vectors as well; it's the only sensible choice for vectors so there's no sense in requiring users specify it. It's not even clear to me that having it as a kwarg is even necessary except for consistency with the matrix methods

[kleinschmidt]

Also just so I'm clear on the overall plan here: for v0.34 we'll make dims=1 the default and remove the dims=nothing path, correct?

[nalimilan]

Maybe first remove the deprecation, and wait for a release before switching the default?