MLflow Project Setup | Sarun Nuntaviriyakul

MLflow Overview
Usage
- Import mlflow library
- Experiment Tracking
  - Logging Parameters, Metrics, and Artifacts
  - Bulk logging of metrics

MLflow Overview

MLflow is an open-source platform designed to manage the machine learning lifecycle, including experimentation, reproducibility, deployment.

Usage

Import mlflow library

import mlflow

Set the remote tracking URI to log experiments to mlflow server

remote_server_uri = "http://your-mlflow-server:5001"
mlflow.set_tracking_uri(remote_server_uri)

Experiment Tracking

Set the experiment name to organize runs

mlflow.set_experiment("jetson_orin_exp")

Logging Parameters, Metrics, and Artifacts

with mlflow.start_run(run_name=model_name + "_" + video_name_short + "_" + str(round)) as active_run:
    # Log parameters
    mlflow.log_param("model_path", model_name)

    ...
    result = model.predict(frame,device=0,conf=0.25,verbose=False)
    ...

    # Log metrics
    mlflow.log_metric("fps", fps)

Bulk logging of metrics

from mlflow.tracking import MlflowClient
from mlflow.entities import Metric


with mlflow.start_run() as active_run:
    mlflow_client = MlflowClient()
    all_metrics = []
    for metric_name in history.history:
        for i in history.epoch:
            metric = Metric(
                key="track_"+metric_name,
                value=history.history[metric_name][i],
                timestamp=0,
                step=i,
            )
            all_metrics.append(metric)

    mlflow_client.log_batch(run_id=active_run.info.run_id, metrics=all_metrics)