Torchdatasets: PyTorch Dataset Utilities with TensorFlow-style API

Torchdatasets is a functional utility library for PyTorch created by szymonmaszke and last updated in June 2022. It provides extensions for PyTorch datasets such as mapping, caching, and filtering, mimicking the API of tensorflow.data.

Use Cases

• Applying `map` transformations to PyTorch Dataset objects for lazy evaluation
• Utilizing `cache` to persist processed data to disk or memory
• Filtering dataset samples using the `filter` method based on custom logic
• Combining multiple datasets using the `concatenate` feature

Strengths

• MIT License allows for commercial and private use
• Provides tensorflow.data-like functional API for PyTorch
• Includes built-in support for disk-based caching of dataset states

Limitations

• Repository has not been updated since June 2022
• Functionality is a software wrapper rather than a collection of raw data records

Provenance

Source

GitHub repository by szymonmaszke

Collection Method

Software development

Freshness

Last updated June 2022; no recent maintenance indicated.