View source on GitHub |
Make an eigenvalue decomposition-based marginal_fn
.
tfp.experimental.distributions.marginal_fns.make_eigh_marginal_fn(
tol=1e-06, name='EigHMarginalFn'
)
For use with GaussianProcess
classes.
A matrix square root is produced using an eigendecomposition. Eigenvalues are forced to be above a tolerance, to ensure positive-definiteness.
Args | |
---|---|
tol
|
Scalar float Tensor . Eigenvalues below tol are raised to tol .
|
name
|
Python str name prefixed to Ops created by this function.
Default value: 'EigHMarginalFn'.
|
Returns | |
---|---|
marginal_function
|
A function that can be used with the marginal_fn
argument to GaussianProcess .
|