Examples¶

The notebooks are small demonstrations of specific APIs. They are not complete training or evaluation pipelines.

1. Simulator components plots signals from individual compartments.
2. Multi-compartment model combines compartments, masks, fractions, and noise.
3. Using pretrained models loads a checkpoint and calls model sampling methods on synthetic data.
4. Train a small model demonstrates model building and a short in-process training loop. It does not replace dmri train for checkpointed runs.
5. Getting example data downloads an open dataset, converts it into the four-file FSL layout dmri predict reads, and runs a prediction.

To run them locally:

uv pip install -e '.[dev]'
uv run jupyter lab docs/examples

The pretrained example downloads a checkpoint. CPU execution can be slow; a supported GPU is recommended for posterior sampling.

Notebooks 1-4 are re-executed on every documentation build, so their figures always match the code. "Getting example data" is not: it downloads 174 MB and runs a full prediction, so it ships with committed outputs and is meant to be run by hand.