alibi-detect

What it is and what it does

Alibi Detect is a detection library for identifying anomalies, distribution shifts, and adversarial examples in machine learning pipelines. It implements multiple algorithms—Isolation Forest, VAE, Seq2Seq, Prophet, and others—each suited to different data types (tabular, images, time series, text) and detection scenarios (online vs. offline). The library returns predictions as dictionaries containing metadata and detection scores, allowing you to flag suspicious instances or monitor for concept drift in production.

The package depends on standard scientific Python tools (numpy, pandas, scikit-learn, transformers) and optionally supports TensorFlow or PyTorch for neural network–based detectors. It is designed for both research and production monitoring, though its Business Source License 1.1 requires careful review for commercial deployments.

Use it for:

• Detect outliers in tabular data using Isolation Forest or Mahalanobis Distance for real-time anomaly flagging.
• Monitor for concept drift in production models by comparing training and live data distributions.
• Identify adversarial examples in image classifiers using Adversarial AE or model distillation.
• Detect anomalies in time series data with Seq2Seq or Prophet for sensor or financial data streams.
• Flag unusual text inputs in NLP pipelines using model distillation–based adversarial detection.

Worth the install?

AI-flagged interpretation of the facts on this page — verify before relying

Alibi Detect provides outlier, adversarial, and drift detection algorithms for tabular data, images, text, and time series, supporting both online and offline detection with TensorFlow and PyTorch backends.

Yes, with conditions. Alibi Detect is well-maintained (active repository, recent release) and covers a broad range of detection tasks across data types. However, the Business Source License 1.1 is proprietary and may restrict commercial use—verify licensing terms before deploying. The aging maintenance window (246 days) is not a blocker for stable use but suggests slower feature updates. Install if your use case aligns with the license and you need production-grade drift or outlier detection.

Install

alibi-detect on PyPI

pip install alibi-detect

uv add alibi-detect

poetry add alibi-detect

Installing alibi-detect

Before you install

Low install friction with a pure-Python wheel. Maintenance status is aging—last release 246 days ago—but the repository remains active with 2546 stars and no archived status. The 17 runtime dependencies are all common scientific Python packages.

License in practice

Licensed under Business Source License 1.1 with unclear treatment. This is a proprietary license that may restrict commercial use or require a separate commercial agreement; review the license terms carefully before deploying in production.

Quickstart

pip install alibi-detect

from alibi_detect.od import OutlierVAE
from alibi_detect.saving import save_detector, load_detector

od = OutlierVAE(threshold=0.1, encoder_net=encoder_net, decoder_net=decoder_net, latent_dim=1024)
od.fit(x_train)
preds = od.predict(x_test)
save_detector(od, './my_detector/')
od = load_detector('./my_detector/')

Requires Python >=3.9. Optional TensorFlow or PyTorch backends must be installed separately for drift detection with those frameworks.

Verify before relying

• Whether Business Source License 1.1 permits your intended use case (commercial, internal, or research only).
• Performance characteristics and scalability limits for large-scale tabular or image datasets.
• Whether aging maintenance (246 days since last release) affects stability for your production timeline.

Package facts

Evidence: alibi_detect-0.13.0-py3-none-any.whl

Tags

More Scientific/Engineering packages