mlops-monitor 0.1.0

Add your description here

mlops-monitor

Lightweight MLOps monitoring toolkit with a web UI, metric logging, drift detection, and run management (inspired by Weights & Biases).

Features

• Log metrics, hyperparameters, and feature distributions from any Python ML script
• Automatic run metadata capture (host, platform, Python version, command, etc.)
• Console output capture (stdout/stderr) into run logs
• Data drift and concept drift detection (C‑backed helpers)
• Web UI: Projects → Runs → Metrics graphs (live and historical)
• SQLite storage, FastAPI + HTMX + Alpine.js + Chart.js frontend

Installation

As a dependency in your project (pip / uv)

# using uv
uv add mlops-monitor

# using pip
pip install mlops-monitor

Editable install for local development / demo

From the repo root:

uv build
# then in your project:
uv add --path ../path/to/mlops-monitor --editable

Or inside the demo/ folder (already configured):

cd demo
uv sync  # uses the editable path defined in pyproject.toml

Quick start

0. Configure project (first time)

uv run monitor edit-config

This will ask:

• Whether to isolate the project (use local DB instead of global)
• Whether to enable email alerts
• Your email address(es) for alerts

1. In your training script

from mlops_monitor import init, log_metrics, capture_output, end

async def main():
    monitor = init("my-project", "run-1")
    capture_output()          # capture prints into run logs
    run_id = await monitor.start()

    for epoch in range(20):
        # train your model ...
        await log_metrics({
            "accuracy": acc,
            "loss": loss,
        }, step=epoch)
        # ...

    await end()                # marks run as completed

if __name__ == "__main__":
    import asyncio
    asyncio.run(main())

2. Launch the UI

uv run monitor ui
# opens at http://localhost:8000

3. Open the UI

• / – Projects list
• /project/{id} – Project overview, run list, and multi‑run metric charts
• /project/{id}/runs/{run_id} – Run details, summary, per‑run charts, and captured logs

Development

git clone ...
cd Cproject
uv sync
uv build                # builds sdist + wheel
uv run monitor ui        # start local UI

The demo/ folder contains a full example using scikit‑learn.