grelu.transforms.label_transforms#

grelu.transforms.label_transform contains classes that perform transformations on labels - for example, sequencing coverage values or other values used to train sequence-to-function deep learning models. This allows users to transform the labels at training time in ways that make training easier.

Classes#

LabelTransform

A class to transform sequence labels.

Module Contents#

class grelu.transforms.label_transforms.LabelTransform(min_clip: int | None = None, max_clip: int | None = None, transform_func: str | Callable | None = None)[source]#

A class to transform sequence labels.

Parameters:

  • min_thresh – Minimum allowed value. Elements with value less than this will be clipped to min_thresh.

  • max_thresh – Maximum allowed value. Elements with value greater than this will be clipped to max_thresh

  • transform_func – A function or name of a function that transforms the label values. Allowed names are “log”.

min_clip = None[source]#
max_clip = None[source]#
transform_func = None[source]#
forward(label: numpy.ndarray) → numpy.ndarray[source]#

Apply the transformation.

Parameters:

label – numpy array of shape (B, T, L)

Returns:

Transformed label

__call__(label: numpy.ndarray) → numpy.ndarray[source]#