FairOps SDK

FairOps SDK is the standalone client library for integrating machine learning prediction streams into the FairOps compliance pipeline. It enables real-time transmission of prediction events, probabilities, ground truths, and sensitive demographic cuts seamlessly to Google BigQuery and the FairOps Engine.

Installation

You can install the SDK via pip:

pip install fairops-sdk

Quick Start

Integrating your model takes literally 3 lines of code:

from fairops_sdk import FairOpsClient

# 1. Initialize mapping your GCP tracking configurations
client = FairOpsClient(
    project_id="your-gcp-project-id",
    # topic_id="fairops-predictions" # Uses default if omitted
)

# 2. Extract your features / scores
features = {"income": 45000, "credit": "Good"}
demographics = {"race": "White", "sex": "Female"}

# 3. Stream the prediction to the Auditor Pipeline
client.log_prediction(
    model_id="credit-risk-v1",
    prediction_label=1,
    prediction_score=0.88,
    features=features,
    demographic_tags=demographics,
    ground_truth=1  # Optional: Log if doing delayed ground-truth syncing
)

Why FairOps?

The fairops-sdk is an asynchronous, non-blocking telemetry client. When you call log_prediction, the payload is aggressively validated using PyDantic schemas before being offloaded to a background thread to prevent any latency injection on your actual API inference gateways.

It handles network retries, connection backoffs, and strict typing automatically.

Requirements

• Python 3.11+
• google-cloud-pubsub
• pydantic >= 2.6.0

Authors & License

Maintained by the FairOps Open Source team. Licensed under Apache 2.0.