BUG: gp.Latent.conditional conditional cov needs stabilizing? #8108
Description
Describe the issue:
With big Xnew arrays, the conditional covariance matrix breaks the model I think because it's not stabilized. The jitter argument is never applied to it. Adding the commented bit of code in fixes the error, but probably unideal.
Reproduceable code example:
= depth.reshape(-1, 1)[::-1]
X = np.arange(99, 0, -1).reshape(-1, 1)
depth_grid = np.linspace(100, 1, 500).reshape(-1, 1)
with pm.Model() as model:
cov = pm.gp.cov.ExpQuad(input_dim=1, ls=1)# + pm.gp.cov.WhiteNoise(sigma=1e-2)
gp = pm.gp.Latent(cov_func=cov)
f = gp.prior("f", X=X)
f_grid = gp.conditional("f_grid", Xnew=depth_grid, jitter=1e-4)Error message:
LinAlgError: Matrix is not positive definitePyMC version information:
5.21.1
Context for the issue:
No response