WhiteKernel
Defined in: ml/kernels/white.js:32
Abstract base class for GP kernels
Extends
Constructors
Constructor
new WhiteKernel(
noiseLevelOrOpts?):WhiteKernel
Defined in: ml/kernels/white.js:45
Parameters
noiseLevelOrOpts?
any = 1.0
Noise variance, or an options
object { noiseLevel, noiseLevelBounds } (aliases: noise_level,
variance, noise_level_bounds). noiseLevelBounds is [low, high],
honoured by hyperparameter optimization; a floor is the usual reason to
set it, since marginal likelihood with many ARD length scales can drive
the noise to zero and explain everything through the kernel.
Returns
WhiteKernel
Example
new WhiteKernel(0.1)new WhiteKernel({ noiseLevel: 0.1, noiseLevelBounds: [0.05, 2] })Overrides
Properties
noiseLevel
noiseLevel:
any
Defined in: ml/kernels/white.js:49
noiseLevelBounds
noiseLevelBounds:
any
Defined in: ml/kernels/white.js:54
Methods
call()
call(
X1,X2?):Matrix
Defined in: ml/kernels/white.js:76
noiseLevel·I for K(X, X), all zeros for a cross-covariance K(X1, X2). Overridden rather than left to the base pointwise loop so the distinction rests on which matrix is being built, not on row identity.
Parameters
X1
any
X2?
any = null
Returns
Overrides
clone()
clone():
Kernel
Defined in: ml/kernels/base.js:87
Clone the kernel with the same parameters
Returns
New kernel instance
Inherited from
compute()
compute(
x1,x2):any
Defined in: ml/kernels/white.js:67
Covariance between two observations. noiseLevel only when they are the
same observation — identified by reference, not by value, since the whole
point of white noise is that two readings of the same input are still
independent. Callers that mean “the variance at this point” pass the same
row twice (compute(x, x)), which is exactly the diagonal case.
Parameters
x1
any
x2
any
Returns
any
Overrides
getParams()
getParams():
object
Defined in: ml/kernels/white.js:90
Get kernel hyperparameters
Returns
object
Hyperparameters
noiseLevel
noiseLevel:
any
Overrides
setParams()
setParams(
params):void
Defined in: ml/kernels/white.js:96
Set kernel hyperparameters
Parameters
params
New parameters
noise_level
any
noiseLevel
any
variance
any
Returns
void