TorchBrain documentation

TorchBrain is a library for writing training pipelines and deep learning models focused on neural data.

Its goal is to provide PyTorch-compatible interfaces like Dataset, Samplers, nn.Modules, and more to help you write your very own models and training scripts. To achieve efficient training, TorchBrain defines its own light-weight data format. It stores neural recordings temporally and is optimized for lazily loading arbitrary time-slices. Finally, TorchBrain also provides a collection of tools to pre-process existing datasets into this data format.

If you encounter any bugs or have feature requests, please submit them to our GitHub Issues page.

Note

We have merged temporaldata and brainsets into torch_brain. If you are migrating from v0.1.x, please see this migration guide.

---

• Install TorchBrain
• User Guide
  • The Data Format
    • Meet the Data Objects
    • Saving and Loading Data
    • Interval Operations
    • Domains
    • Regular Time Series with Gaps
  • Brainsets: Dataset Pre-Processing Tools
    • Using the brainsets CLI
    • Preparing a new Dataset
    • Creating a BrainsetPipeline
    • Creating an OpenNeuroPipeline
  • Training Pipeline Concepts
    • Data Sampling
  • Examples
    • A Minimal Training Setup on NLB Maze
    • Working with the Neuroprobe Benchmark

• CLI Reference

• API Reference
  • torch_brain.batching
  • torch_brain.data
  • torch_brain.datasets
  • torch_brain.models
  • torch_brain.nn
  • torch_brain.pipeline
  • torch_brain.pipeline.openneuro
  • torch_brain.samplers
  • torch_brain.transforms
  • torch_brain.utils
  • torch_brain.utils.bids
  • torch_brain.utils.dandi
  • torch_brain.utils.mne
  • torch_brain.utils.openneuro
  • torch_brain.utils.s3
  • torch_brain.utils.signal
  • torch_brain.utils.split
  • torch_brain.utils.stitcher