model_remediation.min_diff.losses.GaussianKernel

Gaussian kernel class.

Inherits From: MinDiffKernel

View aliases

Main aliases

model_remediation.min_diff.losses.kernels.GaussianKernel

model_remediation.min_diff.losses.GaussianKernel(
    kernel_length: complex = 0.1, **kwargs
)

The Gaussian kernel is a mathematical tool that approximates a given distribution as a sum of gaussian distributions. This is particularly useful when we are trying to determine a distribution from a set of points.

Arguments
`kernel_length`	Length (sometimes also called 'width') of the kernel. Defaults to `0.1`. This parameter essentially describes how far apart points can be and still affect each other. The choice for kernel length should be influenced by the average distance of inputs. The smaller the distance, the smaller the kernel length likely needs to be for best performance. In general, a good first guess is the standard deviation of your predictions. Note: A kernel length that is too large will result in losing most of the kernel's non-linearity making it much less effective. A kernel length that is too small will make the kernel highly sensitive to input noise potentially leading to unstable results.
`**kwargs`	Named parameters that will be passed directly to the base class' `__init__` function.

Arguments

kernel_length

Length (sometimes also called 'width') of the kernel. Defaults to 0.1. This parameter essentially describes how far apart points can be and still affect each other.

The choice for kernel length should be influenced by the average distance of inputs. The smaller the distance, the smaller the kernel length likely needs to be for best performance. In general, a good first guess is the standard deviation of your predictions.

**kwargs Named parameters that will be passed directly to the base class' __init__ function.

See paper for reference on how it can be used in MinDiff.