Skip to content
CI

Add option for fixed (not learnable) random field parameters in BatchGaussianRandomFieldModule. These parameters will often be (effectively) constant in space/time, which means there are initialized with huge (correlation) values. As such, keeping things stable also required using jnp.expm1 to implement (1 - self.phi**2) in the GaussianRandomField. #292

Sign in to view logs

Add option for fixed (not learnable) random field parameters in BatchGaussianRandomFieldModule. These parameters will often be (effectively) constant in space/time, which means there are initialized with huge (correlation) values. As such, keeping things stable also required using jnp.expm1 to implement (1 - self.phi**2) in the GaussianRandomField.

Add option for fixed (not learnable) random field parameters in BatchGaussianRandomFieldModule. These parameters will often be (effectively) constant in space/time, which means there are initialized with huge (correlation) values. As such, keeping things stable also required using jnp.expm1 to implement (1 - self.phi**2) in the GaussianRandomField. #292

python 3.11

python 3.11
failed Jun 19, 2024 in 6m 32s