Fixed error in KDE calculation

[MaroGol]

In sklearn's implementation of KDE, under site-packages/sklearn/neighbors/_kde.py, we see the following:

if isinstance(self.bandwidth, str):
    if self.bandwidth == "scott":
        self.bandwidth_ = X.shape[0] ** (-1 / (X.shape[1] + 4))
    elif self.bandwidth == "silverman":
        self.bandwidth_ = (X.shape[0] * (X.shape[1] + 2) / 4) ** (
            -1 / (X.shape[1] + 4)
        )

When given a bandwidth of "scott", it calculates:

X.shape[0] ** (-1 / (X.shape[1] + 4))

Which in our case is n ** (-1/5).
But the standard Scott's rule formula is $\sigma \cdot n^{-\frac{1}{5}}$.
So sklearn is assuming $\sigma=1$, in other words, that the data has already been standardized, which isn't the case.

As a result, the KDE estimate overfits or underfits the data, which then results in erroneous ID_OOD splits down the line.

A fix could be to standardize the data before calculating the KDE, or to just pass the Scott's formula bandwidth directly, which is what I opted to do in this fix.

Let me know if you think I missed anything :)