Getting Started

This section explains the three main workflows of the MLdebugger SDK.

Workflow Overview

MLdebugger supports the following three main use cases in machine learning model development and operation.

1. Tracing + Evaluation

The basic workflow for collecting inference log data and running evaluations.

Supported tasks: Classification / Object Detection / 3D Object Detection

• Collect data with the appropriate Tracer for each task
• Run evaluations with Evaluator, review results with Result

See Tracing + Evaluation for details.

2. DataFiltering

The workflow for selecting and filtering data based on error patterns.

• Use ClassificationDataFilter / ObjectDetectionDataFilter / ObjectDetection3DDataFilter
• Supports batch processing (query) and real-time filtering

3. Logging

The workflow for collecting inference logs from models in production and monitoring.

• Use ClassificationLogger / ObjectDetectionLogger / ObjectDetection3DLogger
• Monitor model inference behavior in the web app

Prerequisites

Before starting, ensure the following are complete:

1. SDK Installation: See Installation
2. Authentication Credentials: See Authentication