poplar.nn.rescaling

Rescaling objects to be passed to LinearModel to handle the rescaling of input/output data.

These are some standard rescalers to provide something easy to use out of the box; for more complex rescaling behaviour, create your own rescaling class using these examples as a template.

Classes

IdentityRescaler	A placeholder rescaler that leaves input/output data unchanged. Functions may still be applied to the targets by passing them to yfunctions.
ZScoreRescaler	Rescales data to the unit normal distribution.
UniformRescaler	Rescales data to the uniform distribution with bounds [-1, 1].

Module Contents

class poplar.nn.rescaling.IdentityRescaler(yfunctions=None)

A placeholder rescaler that leaves input/output data unchanged. Functions may still be applied to the targets by passing them to yfunctions.

Parameters:

yfunctionslist, optional

A list containing a function and its inverse to apply to the labels prior to rescaling, by default None (i.e. no function is applied)

class poplar.nn.rescaling.ZScoreRescaler(xdata: torch.Tensor, ydata, yfunctions=None)

Rescales data to the unit normal distribution.

Parameters:

xdatatorch.Tensor

Input data.

ydatatorch.Tensor

Input labels corresponding to xdata.

yfunctionslist, optional

A list containing a function and its inverse to apply to the labels prior to rescaling, by default None (i.e. no function is applied)

class poplar.nn.rescaling.UniformRescaler(xdata, ydata, yfunctions=None)

Rescales data to the uniform distribution with bounds [-1, 1].

Parameters:

xdatatorch.Tensor

Input data.

ydatatorch.Tensor

Input labels corresponding to xdata.

yfunctionslist, optional

A list containing a function and its inverse to apply to the labels prior to rescaling, by default None (i.e. no function is applied)