Back to models
Transformer

SignatureTransformer

API referenceSource
MultivariateUnequal length

Transformation class from the signature method.

Quickstart

python
from sktime.transformations.signature_based import SignatureTransformer

estimator = SignatureTransformer(augmentation_list=('basepoint', 'addtime'), window_name='dyadic', window_depth=3, window_length=None, window_step=None, rescaling=None, sig_tfm='signature', depth=4, backend='esig')

Parameters(9)

augmentation_list: list or tuple of strings,

possible strings are ['leadlag', 'ir', 'addtime', 'cumsum', 'basepoint'] Augmentations to apply to the data before computing the signature. The order of the augmentations is the order in which they are applied. default: (‘basepoint’, ‘addtime’)

window_name: str, one of ``[‘global’, ‘sliding’, ‘expanding’, ‘dyadic’]``
default: ‘dyadic’ Type of the window to use for the signature transform.
window_depth: int, default=3

The depth of the dyadic window. Ignored unless window_name is 'dyadic'.

window_length: None (default) or int

The length of the sliding/expanding window. (Active Ignored unless window_name is one of ['sliding, 'expanding'].

window_step: None (default) or int

The step of the sliding/expanding window. Ignored unless window_name is one of ['sliding, 'expanding'].

rescaling: None (default) or str, “pre” or “post”,

  • None: No rescaling is applied.

  • “pre”: rescale the path last signature term should be roughly O(1)

  • “post”: Rescales the output signature by multiplying the depth-d term by d!. Aim is that every term becomes ~O(1).

sig_tfm: str, one of ``[‘signature’, ‘logsignature’]``. default: ``’signature’``
The type of signature transform to use, plain or logarithmic.
depth: int, default=4
Signature truncation depth.
backend: str, one of: ``’esig’`` (default), or ``’iisignature’``.
The backend to use for signature computation.