🚦 Zero-false-positive drift detection for analytics pipelines.

Fast: Optimized with Cython - handles 10,000+ unique values in under 2 seconds.
Flexible: Works with both CSV and Parquet files.

⚠️ Early Stage Project: This project is in active development with frequent breaking changes. The API and data formats may change between versions. Please pin your dependencies if using in production.

pip install driftfx

import pandas as pd, driftfx as dr

dr.snapshot(df_baseline, "baseline", cols=["name"])
result = dr.check(df_new, "baseline", cols=["name"])

if not result.is_clean():
    if result.renames:
        print("Renames / typos:", result.renames[:5])     # first 5 examples
    if result.brand_new:
        print("Brand-new names:", result.brand_new[:5])   # first 5 examples

# snapshot baseline (CSV or Parquet)
$ driftfx snapshot --input data.csv --cols name --baseline baseline/
$ Snapshot complete ✓

# check new batch
$ driftfx check --input new.parquet --cols name --baseline baseline/
$ [✖] name: 17 renames / 31 new codes
$ Drift detected: 48 anomalies 

With Cython-optimized Levenshtein distance calculations:

The Cython implementation provides an 8x speedup for snapshot operations compared to pure Python.