tlc Python Package

The tlc Python package serves as the primary integration point between your machine learning code and the 3LC ecosystem. Integration into an existing codebase is straightforward and can be accomplished by inserting a minimal number of lines into your existing training script.

tlc vs. 3LC

The installable package on PyPI is called 3lc. The installed Python package itself is called tlc due to the limitations of Python package naming conventions. In all other contexts, including the command line interface, 3LC or 3lc is used.

This following sections provide a detailed overview of the package’s capabilities and how to integrate it with your current codebase. See also the Python API documentation for a full breakdown of all the modules, classes and functions in 3LC.

• Core Concepts
  • Objects, URLs, and Schemas
  • Tables
  • Runs
• Register Datasets
  • Importer Tables
  • From PyTorch Dataset
  • From Folders of Images
  • YOLO Format
  • COCO Format with Detectron2
  • Hugging Face Datasets
• Collect Metrics
  • Overview
  • Metrics Collectors
  • Log On-the-Fly Metrics
  • Examples
• 3LC Project Structure
  • Overview
  • Project Folder Structure
• SampleType and structure
  • Schema
  • Sample vs Row
  • SampleType
  • structure
• Sample Weights
  • When to use sample weights
  • Modifying weights in the 3LC dashboard
  • Using weights during training
  • Using weights during metrics collection
• URL Details
  • Supported URL Schemes
  • File Paths
  • Amazon S3
  • Google Cloud Storage (GCS)
  • Azure Blob Storage
• Embeddings
  • What are Neural Network Embeddings?
  • Importance of Embeddings in Machine Learning
  • Collecting Embeddings in 3LC
  • Dimensionality Reduction in 3LC
• Combining Data and Models in 3LC
  • Standard PyTorch Data Pipeline
  • Data Pipeline Using 3LC
  • Examples
  • Intermediate activations and embeddings
  • Conclusion