MeanStdScaling is a bad name

[glennmoy]

As noted by @eperim

"Standardised" data is a pretty common nomenclature, while "MeanStdScaling" sounds like it could be anything (and the data is not even scaled by the mean, it is shifted by it, only the std is used for scaling).

[bencottier]

I think it's fine! "Standardised" sounds to me like it could be many things 😅 With MeanStdScaling I know immediately what it means among all the common ways of scaling data in ML (and I thought so before we wrote this Transform).

I do think "Standardised" is better than "Normalised" (which I considered originally) because while that has "Normal" in the name, it can refer to scaling by any kind of norm.

[eperim]

I disagree. "Standardisation" is quite common nomenclature. There's even a special tag for it in Stats Exchange, and you can find the term frequently in glossaries, multiple blog posts, teaching materials, and in other packages (e.g. sklearn). These are all from the first page of a Google search.

With MeanStdScaling I know immediately what it means among all the common ways of scaling data in ML

Naturally you do, because you chose the name, but is it common nomenclature that you expect people in general to be familiar with?

[bencottier]

"Standardisation" is quite common nomenclature.

Thanks for gathering those links, that makes me favour "Standardisation" more.

Naturally you do, because you chose the name, but is it common nomenclature that you expect people in general to be familiar with?

I copied the name from FeatureEngineering, but yes I had the chance to rename in the original PR and decided to keep it. I think more in terms of "how to identify what this literally does with a relatively short name?" than the nomenclature. You might even disagree that "MeanStdScaling" is best for that purpose, but that's my reasoning.

I commented to show there's a difference of opinion, but I don't feel strongly about this. Nomenclature is important and it's very reasonable to rename it.

[molet]

I would also vote for "Standardization" instead of "MeanStdScaling".
Also, would it be possible to add "Normalization" as well to the package?
I.e. ( x - min(x) ) / ( max(x) - min(x) )

[xiaodaigh]

MeanStd is more precise

[juliohm]

Who is maintaining the package? There are many low-hanging fruit issues like this one that could be easily fixed, and many other issues that could be closed. Can I take responsibility for some of these as an external collaborator?

I've finished my first Transform here and as I mentioned in another issue, we could certainly benefit from some refactoring of the API.

[juliohm]

After experimenting with the API a bit more, I realized that it is too rigid and unlikely to change anytime soon due to the internal usage of the package at Invenia. I will start my own package to experiment with some ideas, and hopefully things can be merged in the future.

[mzgubic]

How does StandardScaler/ing sound?