mlops

Shared MLOps platform package — not a machine learning project of its own. It gives every ML project the generic building blocks of a training pipeline (data validation, experiment tracking, quality gates, drift monitoring), so each project imports them instead of re-writing the same tracking/quality-gate/validation/monitoring logic:

• mlops.validation.run_expectations(df, rules) — run Great Expectations rules
• mlops.mlflow_tracking.log_and_register(...) / promote_model_to_production(...) — MLflow experiment tracking and model registry
• mlops.quality_gate.check_threshold(...) — stop a pipeline if a metric regressed
• mlops.monitoring.check_drift(reference_data, current_data) — Evidently drift detection
• mlops.git_info.get_current_commit_hash() — commit hash for run metadata

Everything project-specific (data source, feature engineering, model training, API serving, the orchestration flow that wires these steps together) stays in each consuming project's own repo — mlops only provides the reusable building blocks above.

Install

pip install mlops_abp
# or with uv
uv add mlops_abp

The distribution is named mlops_abp on PyPI; the importable package is mlops:

from mlops import log_and_register, check_threshold, run_expectations, check_drift

For local development against a checkout of this repo instead of the published release:

cd /path/to/your-project
uv add --editable /path/to/mlops

Usage

from mlops import check_threshold, log_and_register, run_expectations

run_expectations(df_train, rules=[
    {"expectation": "expect_column_values_to_not_be_null", "kwargs": {"column": "target"}},
])

run_id = log_and_register(
    model=model,
    run_name="baseline",
    params=params,
    metrics=metrics,
    model_name="your_model_name",
    experiment_name="your_project_experiment",  # distinct per project
)
check_threshold(metrics["accuracy"], threshold=0.85, metric_name="accuracy")

Drift monitoring:

from mlops import check_drift

if check_drift(reference_data=baseline_df, current_data=recent_df):
    print("Drift detected — consider retraining")

Shared platform infrastructure (optional, for self-hosting MLflow/Prefect)

git clone <this-repo>
cd mlops
cp .env.example .env
docker compose up -d   # postgres (MLflow backend store) + mlflow-server + prefect-server

• MLflow UI: http://localhost:5000
• Prefect UI: http://localhost:4200

Development

uv sync
uv run pytest tests/ -v
uv run ruff check mlops/ tests/

License

MIT — see LICENSE.