Welcome to Fuel’s documentation!¶
- Fuel allows you to easily read different types of data (NumPy binary files, CSV files, HDF5 files, text files) using a single interface which is based on Python’s iterator types.
- Provides a a series of wrappers around frequently used datasets such as MNIST, CIFAR-10 (vision), the One Billion Word Dataset (text corpus), and many more.
- Allows you iterate over data in a variety of ways, e.g. in order, shuffled, sampled, etc.
- Gives you the possibility to process your data on-the-fly through a series of (chained) transformation procedures. This way you can whiten your data, noise, rotate, crop, pad, sort or shuffle, cache it, and much more.
- Is pickle-friendly, allowing you to stop and resume long-running experiments in the middle of a pass over your dataset without losing any training progress.
Fuel is a new project which is still under development. As such, certain
(all) parts of the framework are subject to change. The last stable (but
possibly outdated) release can be found in the
That said, if you are interested in using Fuel and run into any problems, feel free to ask your question on the mailing list. Also, don’t hesitate to file bug reports and feature requests by making a GitHub issue.
The best way to get started with Fuel is to have a look at the overview documentation section.