mojolearn.KNNImputer#

class mojolearn.KNNImputer(*, missing_values=nan, n_neighbors=5, weights='uniform', metric='nan_euclidean', copy=True, add_indicator=False, keep_empty_features=False)[source]#

Bases: _XNeighbors

Imputation of missing values by k nearest neighbors.

Fills each missing value from the n_neighbors nearest training rows that have that feature, using NaN-aware euclidean distances, following scikit-learn’s KNNImputer. Distance ties go to the lower row index. When no donor is at a finite distance, the training column mean is used. Features that are entirely missing at fit are dropped from the output (or set to zero with keep_empty_features=True). Computation is float32.

Parameters:

missing_values (float, default nan) – Value marking a missing entry: NaN or another number. Strings, None and booleans raise NotImplementedError.
n_neighbors (int, default 5) – Number of donor neighbors; at least 1.
weights ({‘uniform’, ‘distance’}, default ‘uniform’) – Donor weighting: equal, or by inverse distance. Callables raise NotImplementedError.
metric ({‘nan_euclidean’}, default ‘nan_euclidean’) – Distance; any other value raises NotImplementedError.
copy (bool, default True) – Accepted for scikit-learn compatibility; the output is always a new array.
add_indicator (bool, default False) – Append a 0/1 indicator column for each feature that had missing values at fit.
keep_empty_features (bool, default False) – Keep features that are entirely missing at fit, imputed as zero.
numeric_mode ({‘identical’, ‘deterministic’, ‘fast’} or None, default None) – Numeric mode for this estimator ('fast' is Apple only). None uses the process default.

Variables:

n_features_in_ (int) – Number of features seen in fit.

Examples

>>> import numpy as np, mojolearn
>>> X = np.random.default_rng(0).random((1000, 8), dtype=np.float32)
>>> X[::7, 3] = np.nan
>>> X_filled = mojolearn.KNNImputer(n_neighbors=5).fit_transform(X)

__init__(*, missing_values=nan, n_neighbors=5, weights='uniform', metric='nan_euclidean', copy=True, add_indicator=False, keep_empty_features=False)[source]#

Methods

`fit`(X[, y])	Store the training data used as donors.
`fit_transform`(X[, y])	Fit to X, then impute X: `fit(X).transform(X)`.
`get_feature_names_out`([input_features])	Output feature names.
`get_params`([deep])
`numeric_mode_used`()	The numeric mode this estimator runs in.
`set_params`(**params)	Set constructor parameters.
`transform`(X)	Impute the missing values of X.
`vendor_used`()	The GPU backend this estimator runs on.

fit(X, y=None)[source]#

Store the training data used as donors.

Parameters:

X (array-like or scipy.sparse matrix of shape (n_samples, n_features)) – Training data, with missing values marked by missing_values.
y (None) – Ignored.

Returns:

self (KNNImputer) – The fitted imputer.

transform(X)[source]#

Impute the missing values of X.

Parameters:: X (array-like or scipy.sparse matrix of shape (n_samples, n_features)) – Data to impute; the number of features must match fit (ValueError otherwise).
Returns:: X_imputed (Array of shape (n_samples, n_features_out), float32) – Imputed data, without the features that were entirely missing at fit (unless keep_empty_features), followed by the indicator columns when add_indicator is True.

fit_transform(X, y=None)[source]#

Fit to X, then impute X: fit(X).transform(X).

Parameters:

X (array-like of shape (n_samples, n_features)) – Data with missing values.
y (None) – Ignored.

Returns:

X_imputed (Array of shape (n_samples, n_features_out), float32)

get_feature_names_out(input_features=None)[source]#

Output feature names.

The kept input names, followed by missingindicator_<name> for each indicator column when add_indicator is True.

Parameters:: input_features (list of str or None, default None) – Input feature names; None means x0, x1, … Its length must equal n_features_in_ (ValueError otherwise).
Returns:: names (list of str)