gpflow.optimizers.natgrad¶

gpflow.optimizers.natgrad.expectation_to_meanvarsqrt¶

gpflow.optimizers.natgrad.expectation_to_meanvarsqrt(eta1, eta2)[source]¶

Parameters

eta1 (Tensor) –
eta2 (Tensor) –

Return type

Tuple[Tensor, Tensor]

gpflow.optimizers.natgrad.expectation_to_natural¶

gpflow.optimizers.natgrad.expectation_to_natural(eta1, eta2)[source]¶

Parameters

eta1 (Tensor) –
eta2 (Tensor) –

Return type

Tuple[Tensor, Tensor]

gpflow.optimizers.natgrad.meanvarsqrt_to_expectation¶

gpflow.optimizers.natgrad.meanvarsqrt_to_expectation(m, v_sqrt)[source]¶

Parameters

m (Tensor) –
v_sqrt (Tensor) –

Return type

Tuple[Tensor, Tensor]

gpflow.optimizers.natgrad.meanvarsqrt_to_natural¶

gpflow.optimizers.natgrad.meanvarsqrt_to_natural(mu, s_sqrt)[source]¶

Parameters

mu (Tensor) –
s_sqrt (Tensor) –

Return type

Tuple[Tensor, Tensor]

gpflow.optimizers.natgrad.natural_to_expectation¶

gpflow.optimizers.natgrad.natural_to_expectation(nat1, nat2)[source]¶

Parameters

nat1 (Tensor) –
nat2 (Tensor) –

Return type

Tuple[Tensor, Tensor]

gpflow.optimizers.natgrad.natural_to_meanvarsqrt¶

gpflow.optimizers.natgrad.natural_to_meanvarsqrt(nat1, nat2)[source]¶

Parameters

nat1 (Tensor) –
nat2 (Tensor) –

Return type

Tuple[Tensor, Tensor]

gpflow.optimizers.natgrad.swap_dimensions¶

gpflow.optimizers.natgrad.swap_dimensions(method)[source]¶

Converts between GPflow indexing and tensorflow indexing method is a function that broadcasts over the first dimension (i.e. like all tensorflow matrix ops):

method inputs [D, N, 1], [D, N, N] method outputs [D, N, 1], [D, N, N]

Return type: Callable[..., Tuple[Tensor, Tensor]]
Returns: Function that broadcasts over the final dimension (i.e. compatible with GPflow): inputs: [N, D], [D, N, N] outputs: [N, D], [D, N, N]

Parameters: method (Callable[[Tensor, Tensor], Tuple[Tensor, Tensor]]) –