Sklearn GaussianProcessRegressor fixing kernel hyperparameters?

Question

I'm trying to use sklearn's gaussian process for timeseries decomposition.

kernel = ConstantKernel() * 
RBF() *
ExpSineSquared(periodicity=7)

Is there a way to fix the parameters other then periodicity_bounds=(7, 7)

If i do kernel.hyperparameters i can see they have a attribute fixed=False

How do i set this to true?

score 7 · Accepted Answer · answered Dec 11 '18 at 08:50

7

Its not documented on the Kernels them self. But the hyper parameters can fixed by the following.

ExpSineSquared(periodicity=7, periodicity_bounds='fixed')

answered Dec 11 '18 at 08:50

CodeMonkey

You can find the documentation of the Hyperparameter class here: https://scikit-learn.org/stable/modules/generated/sklearn.gaussian_process.kernels.Hyperparameter.html Indeed, the bound attributes can take "fixed" if you want the hyperparameter's value cannot be changed. – Pierre Dec 11 '18 at 09:22

1 Answers1