Vectora Python SDK

Vectora helps you monitor sklearn models in production without adding latency to your prediction path.

The SDK captures:

• prediction metrics like accuracy, F1, precision, and recall
• feature distribution summaries for drift detection
• SHAP-based feature importance when shap is available
• trace IDs that connect predictions back to the Vectora dashboard

Install

pip install vectora

Quickstart

from vectora import VectoraClient
from vectora.models import SklearnConnector

client = VectoraClient(api_key="vct_live_xxx")
connector = SklearnConnector(
    client=client,
    model=your_sklearn_model,
    model_id="11111111-1111-1111-1111-111111111111",
)

predictions = connector.predict(X_test, y_true=y_test)

predict() returns the model's predictions immediately. Vectora sends the monitoring payload in a background thread so your production inference path stays fast.

What gets sent

Each prediction call sends a payload to /api/ingest/metrics with:

• trace_id
• model_id
• metrics
• feature_distributions
• shap_values
• sample_count
• timestamp

If SHAP is not installed, the SDK logs a warning once and continues without SHAP values.

If the network call fails, the SDK logs the error to stderr and never raises it back to your prediction path.

API

VectoraClient

client = VectoraClient(
    api_key="vct_live_xxx",
    base_url="https://vectora.ai",
    timeout=5.0,
    max_retries=2,
)

SklearnConnector

connector = SklearnConnector(client, model_id="11111111-1111-1111-1111-111111111111", model=trained_model)
predictions = connector.predict(X, y_true=y_true)

Coming Soon

vectora.llm and vectora.agent are reserved for future releases and raise ComingSoonError when accessed.

Release

Tagging the repository with v* triggers the GitHub Actions publish workflow in .github/workflows/publish.yml.

git tag v0.1.0
git push origin v0.1.0

The workflow:

• runs the SDK test suite
• builds the source and wheel distributions
• publishes to PyPI using GitHub Actions trusted publishing