CatBoost Estimators should be able to handle float categories

[tamargrey]

• As a user, I wish I could pass categorical data whose categories are floats into the CatBoostRegressor and CatBoostClassifier estimators. They currently raise Invalid type for cat_feature category for [feature_idx=312]=1.0 : cat_features must be integer or string, real number values and NaN values should be converted to string. with the code snippet below:

Code Example

    import woodwork as ww
    import pandas as pd

    X = pd.DataFrame({"double_cats": pd.Series([1.0, 2.0, 3.0, 4.0, 5.0]*20)})
    y = pd.Series(range(100))
    y.ww.init()
    X.ww.init(logical_types={"double_cats": "Categorical"})

    clf = CatBoostRegressor(
        n_estimators=1,
        max_depth=1,
    )
    clf.fit(X, y)

In the case above the floats can be converted to ints, but we should also handle the case where we cannot convert the floats to ints without truncating the data (we can consider converting to the object dtype in that case)

[tamargrey]

See this comment for more info, but for now we will not be handling the case where actual floats (like 1.1, 2.5, etc) are used, as there's not a clean way to convert the values in a way CatBoost will take, and according to their FAQ, we might not want to use stringified floats anyway.

If either catboost/catboost#1965 or pandas-dev/pandas#51074 is completed and we have updated to those CatBoost or Pandas versions, we could update handle_float_categories_for_catboost to convert floats as well. Hopefully by the time a need for that arises, one of those is fixed.