ENH avoid checking columns where training data is all nan in KNNImputer

[xuefeng-xu]

Reference Issues/PRs

What does this implement/fix? Explain your changes.

In KNNImputer, columns where training data is all nan will be removed or impute with 0.
Therefore, we only need to check data with valid columns using valid_mask.
This can avoid computing pairwise distance when data with valid columns has no missing values.

Any other comments?

This could potentially save some memory.

[github-actions]

✔️ Linting Passed

All linting checks passed. Your pull request is in excellent shape! ☀️

_{Generated for commit: 4f6bed6. Link to the linter CI: here}

[xuefeng-xu]

This code script shows efficiency improvement if one column of data is all nan.

import numpy as np
from sklearn.datasets import fetch_california_housing
from sklearn.impute import KNNImputer
import pandas as pd

calhousing = fetch_california_housing()

X = pd.DataFrame(calhousing.data, columns=calhousing.feature_names)
y = pd.Series(calhousing.target, name='house_value')

rng = np.random.RandomState(42)
density = 10
mask = rng.randint(density, size=X.shape) == 0

X_na = X.copy()
X_na.values[mask] = np.nan

X_na.values[:, 0] = np.nan # one column is all nan

This PR

%timeit KNNImputer().fit_transform(X_na)
6.72 s ± 59.5 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

Main

%timeit KNNImputer().fit_transform(X_na)
9.24 s ± 109 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)