Make PandasStandardScaler for HUGS

[sgoldenCS]

In order to keep track of each step of the process in preparing for HUGS, I am making a issue/branch for the next step (the data prep module). I think standard scaling is probably fine for our example data. This issue intends to complete one step of #34.

[sgoldenCS]

A quick note: I will be developing the scaler here and then proceed to merge it into the branch for #34.

[sgoldenCS]

I have nearly finished the implementation of a standard scaler. It should be quite robust and includes options for inplace scaling, axis arguments, and a tunable epsilon that controls if a feature has enough variance to be scaled by the standard deviation.

I have written unit tests to check the majority of options. I still need to check the reverse scaling.

I have a bunch of extra functions based on the syntax from scikit-learn (fit, transform, inverse_transform) and simply call them using the interface defined in the core. I'm open to having the core interface renamed, but it's not super important.

Some important things to note:

• The implementation is done mostly by hand to avoid issues with scikit-learn and Pandas.
  • Pandas does not allow for axis==None with variance (it produces a deprecation warning and treats it like axis==0)
  • Pandas' implementation of variance assumes sample variance (degrees of freedom == 1, so we divide by N-1). This differs from numpy
  • scikit-learn StandardScaler doesn't allow for any axis parameters
• scikit-learn does some interesting things with output types. They implement a registered set of converters for Pandas and Polars (see https://github.com/scikit-learn/scikit-learn/blob/8721245511de2f225ff5f9aa5f5fadce663cd4a3/sklearn/utils/_set_output.py#L183)
• It's possible that this code works for a few different input datatypes but I haven't tested anything other than Pandas. I believe that the main issue is the current implementation for inplace == False assumes the input is Pandas. Anything that operates like a numpy array should work fine until the output conversion.

[sgoldenCS]

I've written the code for reverse scaling and the associated unit test. I think this can be made into a pull request now.