GaussianProcessJax

class lsst.meas.algorithms.GaussianProcessJax(std=1.0, correlation_length=1.0, white_noise=0.0, mean=0.0)

Bases: object

Gaussian Process regression in JAX. Kernel is assumed to be isotropic RBF kernel, and solved using exact Cholesky decomposition. The interpolation solution is obtained by solving the linear system: y_interp = kernel_rect @ (kernel + y_err**2 * I)^-1 @ y_training. See the Rasmussen and Williams book for more details. Each function is decorated with @jit to compile the function. Exist package like tinygp, that is implemented in jax also. This class is a custom implementation of Gaussian Processes, which allows for setting the hyperparameters, fine-tuning the mean function, and other specifications.

Methods Summary

fit(x_train, y_train)
predict(x_predict)

Methods Documentation

fit(x_train, y_train)

predict(x_predict)