Skip to main content

FaissImputer

Warning: FaissImputer 0.1.x has a known neighbor-mapping bug that can produce incorrect imputations, and it is incompatible with scikit-learn 1.8+. Use version 0.2.0 or newer.

PyPI Version License

A scikit-learn-compatible missing-value imputer with Faiss-backed neighbor search.

What's new in 0.3.1

  • Optimizes the donor-array layout for donor_policy="complete" to reduce repeated column-projection overhead.
  • Adds a reproducible missingness-pattern benchmark, raw results, and a report.
  • Leaves the public API, imputation algorithm, and available-donor policy unchanged.

Performance depends on data size and missingness patterns. Small training sets with many different patterns can still be slower than KNNImputer.

See the 0.3.1 release notes and the complete-donor benchmark report.

What's new in 0.3.0

  • Adds donor_policy="available" to use partially observed training rows as donors.
  • Keeps donor_policy="complete" as the default, preserving the existing behavior.
  • Adds batched NaN-aware distances and numerical safeguards for the available-donor policy.
  • Expands regression tests, dependency compatibility checks, and installed-package smoke tests.

See the 0.3.0 release notes.

Installation

FaissImputer requires Python 3.10 or newer.

python -m pip install --upgrade "faiss-imputer>=0.3.1"

Usage

Complete donors (default)

Training data must contain enough fully observed rows to supply n_neighbors donors.

import numpy as np

from faiss_imputer import FaissImputer

X_train = np.array(
    [
        [1.0, 10.0, 100.0],
        [2.0, 20.0, 200.0],
        [3.0, 30.0, 300.0],
    ],
    dtype=np.float32,
)
X_missing = np.array(
    [
        [1.5, np.nan, 150.0],
        [2.5, 25.0, np.nan],
    ],
    dtype=np.float32,
)

imputer = FaissImputer(n_neighbors=2)
X_imputed = imputer.fit(X_train).transform(X_missing)
print(X_imputed)

Expected output:

[[  1.5  15.  150. ]
 [  2.5  25.  250. ]]

Partially observed donors

Set donor_policy="available" to choose donors separately for each missing feature. No fully observed training row is required.

import numpy as np

from faiss_imputer import FaissImputer

# No training row is completely observed.
X_train = np.array(
    [
        [0.0, 10.0, np.nan],
        [2.0, 30.0, np.nan],
        [1.0, np.nan, 20.0],
        [3.0, np.nan, 40.0],
    ],
    dtype=np.float32,
)
X_missing = np.array(
    [[0.1, np.nan, np.nan], [np.nan, np.nan, np.nan]],
    dtype=np.float32,
)

imputer = FaissImputer(n_neighbors=1, donor_policy="available")
X_imputed = imputer.fit(X_train).transform(X_missing)
print(X_imputed)

Expected output:

[[ 0.1 10.  20. ]
 [ 1.5 20.  30. ]]

Both policies return a new NumPy float32 array. The training and query inputs are not modified.

Parameters

  • n_neighbors (default: 3): Positive integer specifying the maximum number of donors used for each missing feature. The complete-donor policy requires at least this many complete training rows.
  • metric (default: "l2"): Supports "l2" and "ip". Raw inner product is not cosine similarity. The available-donor policy requires "l2".
  • strategy (default: "mean"): Supports "mean" and "median" for aggregating donor values and calculating fallback column statistics.
  • index_factory (default: "Flat"): Faiss index description for the complete-donor policy. The available-donor policy accepts only "Flat" and uses the distance backend described below.
  • donor_policy (default: "complete"): Use fully observed training rows with "complete", or allow partially observed training rows with "available".

Important behavior

Shared behavior

  • Inputs must be two-dimensional numeric array-like data, with NaN marking missing values. Values are converted to float32; infinity is not accepted.
  • transform() requires the same number of features as fit().
  • An entirely missing query row uses column means or medians learned during fit().
  • A failed fit(), including a failed refit, clears the fitted state.

Complete-donor policy

  • Only fully observed training rows are used as donors. At least one complete row is required.
  • n_neighbors cannot exceed the number of complete donors.
  • Neighbor search uses only the originally observed columns of each query row.
  • The default index_factory="Flat" performs exact neighbor search.

Available-donor policy

  • Only metric="l2" and index_factory="Flat" are supported. Both "mean" and "median" strategies are supported.
  • A donor must observe the feature being imputed and share at least one originally observed feature with the query row.
  • Donors are ranked by squared L2 distance over shared observed features, scaled by the total feature count divided by the shared feature count.
  • Each missing feature uses up to n_neighbors eligible donors. Fewer eligible donors are allowed.
  • If no eligible donor exists for a feature, its fitted column mean or median is used.
  • Entirely missing training rows are ignored; entirely missing training columns are rejected.

The available-donor policy uses batched NaN-aware distances and Faiss neighbor selection, with float64 safeguards for detected numerical risks. This backend is also used when all training donors are complete.

Benchmark

Partial donors: development snapshot

The partial-donor benchmark report includes timings, tolerance-based agreement checks against KNNImputer, raw results, and limitations.

It measures development commit 0a3cc077, not the final 0.3.0 implementation. Subsequent changes include the backend used for "available" with fully observed donors.

Performance depends on the donor policy, data size, and missingness pattern. Neither faster execution nor lower memory use than KNNImputer is guaranteed. Agreement on the tested cases is not a guarantee of identical results on every dataset.

Version 0.2.0: historical results

In a controlled synthetic benchmark with complete training donors, FaissImputer 0.2.0 and scikit-learn's KNNImputer produced exactly matching imputed values. With 20,000 training rows, FaissImputer transformed repeated missingness patterns up to 19.62 times faster, while performance was effectively tied when almost every test row had a different pattern.

When the training data itself contained missing values, FaissImputer was less accurate because version 0.2.0 uses only fully observed rows as donors. These single-thread results are specific to the tested synthetic data and hardware, not a general performance guarantee.

See the full 0.2.0 benchmark report, reproduction script, and raw results.

What's new in 0.2.2

  • Clears fitted state after a failed fit(), including failed refits.
  • Corrects scikit-learn transformer, NaN-support, and float32 dtype-preservation tags.
  • Adds regression tests for failed fits, estimator tags, and pipeline use.

See the 0.2.2 release notes.

What's new in 0.2.1

Version 0.2.1 is a packaging and documentation hotfix. The imputation behavior is unchanged from 0.2.0.

  • Corrects the PyPI description so it no longer presents 0.2.0 as unreleased.
  • Uses portable warning markup and absolute documentation links that work on PyPI.
  • Verifies built package metadata against the release tag before publication.

See the 0.2.1 release notes.

What's new in 0.2.0

Version 0.2.0 corrected the imputation behavior of 0.1.x.

  • Corrects the neighbor-to-donor mapping that could return values from the wrong training row.
  • Uses the donor rows learned during fit() when new data is passed to transform().
  • Searches for neighbors using only the columns observed in each query row.
  • Preserves the input array and reuses fallback statistics learned during fit().
  • Supports modern scikit-learn validation APIs and adds regression tests and automatic CI.

See the 0.2.0 release notes.

Example notebook

See Imputing Missing Values with Faiss Imputer for a complete-donor example.

Contributing

Contributions are welcome! Please open an issue or create a pull request. Further work is tracked in the roadmap.

Author

License

This project is licensed under the MIT License.

Third-Party Licenses

FaissImputer depends on Meta's Faiss, which is distributed under the MIT License.

FaissImputer is not affiliated with or endorsed by Meta or the Faiss maintainers.

Metadata

Release files for faiss-imputer 0.3.1

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for faiss-imputer 0.3.1
File Size Uploaded Provenance Actions
faiss_imputer-0.3.1.tar.gz 16.8 kB With provenance Details for faiss_imputer-0.3.1.tar.gz

Built distribution (wheel)

Table of built distributions (wheels) for faiss-imputer 0.3.1
File Interpreter ABI Platform Actions
faiss_imputer-0.3.1-py3-none-any.whl Python 3 none any Details for faiss_imputer-0.3.1-py3-none-any.whl

Total release size: 27.0 kB

Release files / faiss_imputer-0.3.1.tar.gz

Download URL faiss_imputer-0.3.1.tar.gz
Size 16.8 kB
Tags Source
SHA-256 checksum
How to use checksums
7d28785d1441eff701078819bcbcbe7830d0d0526ba5b8c612b46666804b4b29
BLAKE2b-256 checksum
How to use checksums
a0de8d72bd16ea952af08f18be93950e798249fc7599ad4c3e031aa4fa61d46b
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.14

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Sep 3, 2026.

Transparency log

Release files / faiss_imputer-0.3.1-py3-none-any.whl

Download URL faiss_imputer-0.3.1-py3-none-any.whl
Size 10.2 kB
Tags Python 3
SHA-256 checksum
How to use checksums
c221349c8f8f04871208a9f4081681f8f3af6b97f2aeb10ee08a506cf97a93ea
BLAKE2b-256 checksum
How to use checksums
112a0db1eea0256ad57ef89fb7a541352c7befcd97edff3b64a2b449894bc0b8
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.14

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Sep 3, 2026.

Transparency log

Security for faiss-imputer 0.3.1

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

Full provenance

Both files in this release have been uploaded with provenance metadata.

Select from 2 files

Attestation for faiss_imputer-0.3.1.tar.gz

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Sep 3, 2026.

Transparency log

A note on package security

PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.

To learn more about protecting yourself from malicious actors, visit our package security documentation.

Release history Release notifications | RSS feed

This release

0.3.1 This release

2 release files Full provenance

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page