pysiglib.branched_sig_combine_backprop

pysiglib.branched_sig_combine_backprop#

branched_sig_combine_backprop(derivs, bsig1, bsig2, dimension, degree, *, planar=False, n_jobs=1)[source]#

Backpropagates through the branched signature combine (Butcher product).

Given out = branched_sig_combine(bsig1, bsig2, dimension, degree) and upstream derivatives derivs = dF/d(out), computes (dF/d(bsig1), dF/d(bsig2)).

Parameters:

  • derivs (numpy.ndarray | torch.tensor) – Upstream derivatives, same shape as combine output.

  • bsig1 (numpy.ndarray | torch.tensor) – First branched signature input to the forward combine.

  • bsig2 (numpy.ndarray | torch.tensor) – Second branched signature input to the forward combine.

  • dimension (int) – Dimension of the underlying path.

  • degree (int) – Maximum order.

  • planar (bool) – If True, backpropagate through planar branched sig combine.

  • n_jobs (int) – Number of parallel threads for batch processing.

Returns:

Tuple (dF/d(bsig1), dF/d(bsig2)), in the same scalar-term format as the inputs.


Citation#

If you found this library useful in your research, please consider citing the paper:

@article{shmelev2025pysiglib,
  title={pySigLib-Fast Signature-Based Computations on CPU and GPU},
  author={Shmelev, Daniil and Salvi, Cristopher},
  journal={arXiv preprint arXiv:2509.10613},
  year={2025}
}
Contents