Impute#

This module contains some functions related to impute.

Annoy (Approximate Nearest Neighbors Oh Yeah) is a C++ library with Python bindings to search for points in space that are close to a given query point. It also creates large read-only file-based data structures that are mmapped into memory so that many processes may share the same data.

Voyager is optimized for modern nearest-neighbor search.

HNSW (Hierarchical Navigable Small World) provides better accuracy and speed, outperforming Annoy in most use cases, especially when precision is important.

See also

ANNImputer#

This module contains some functions related to ANNImputer.

TL;DR#

  • Purpose: Approximate k-nearest-neighbors (KNN) imputation

  • Import path: from scikitplot.impute import ANNImputer

  • Functionality: Replaces missing values using neighbors retrieved via Annoy

  • Parameters: n_neighbors, n_trees, metric, optional search_k, etc.

Overview#

ANNImputer (from impute) is an approximate nearest-neighbors imputer that uses Annoy to fill missing values in datasets. It replaces missing entries by querying the nearest neighbors of samples with missing values and computing imputations from those neighbors.

Motivation#

Unlike exact KNN imputation KNNImputer, using Annoy allows: - Faster neighbor retrieval in high-dimensional data - Memory-efficient indexing of large datasets - Sharing of prebuilt indexes across processes

Example: Your exact NumPy array example:

import numpy as np
from scikitplot.experimental import enable_ann_imputer
from scikitplot.impute import ANNImputer

X = np.array([[1, 2, np.nan],
              [3, 4, 3],
              [np.nan, 6, 5],
              [8, 8, 7]])

imputer = ANNImputer(n_trees=5, n_neighbors=5)
X_imputed = imputer.fit_transform(X)

print(X_imputed)
# Output:
# array([[1., 2., 5.],
#        [3., 4., 3.],
#        [4., 6., 5.],
#        [8., 8., 7.]])

Mechanism#

  • Builds an Annoy index from complete samples

  • Queries nearest neighbors for incomplete samples

  • Imputes missing values based on neighbor vectors

  • Integer identifiers are used internally; memory allocated to max(id)+1

Notes#

  • Memory-efficient and fast for large datasets

  • Approximate neighbors; exact KNN may differ slightly

  • Shares indexes across processes using mmap

  • Behavior depends on n_trees and search_k parameters

Comparison#

  • Similar in usage to KNNImputer, but faster on large, high-dimensional datasets

  • Provides a trade-off between accuracy and speed via Annoy parameters