Internals

The tensor algebra of the rules: multiplying a tensor by a vector or a tensor along some of its axes, and summing those axes out. The rules of DiscreteTransition are built from these two; they are not exported.

DiscreteTransitionMessagePassingRules.multiply_dimensions! — Function

multiply_dimensions!(tensor, dims::NTuple{N, Int}, values) -> tensor

Multiply tensor, elementwise and in place, by values along the axes dims: the k-th axis of values runs along the axis dims[k] of tensor, and dims must be increasing, since values is reshaped, not permuted. When the element types differ, both are promoted first and a new tensor is returned; otherwise tensor itself.

Examples

julia> DiscreteTransitionMessagePassingRules.multiply_dimensions!([1.0 2.0; 3.0 4.0], (2,), [10.0, 100.0])
2×2 Matrix{Float64}:
 10.0  200.0
 30.0  400.0

See also sum_out_dimensions.

source

DiscreteTransitionMessagePassingRules.sum_out_dimensions — Function

sum_out_dimensions(tensor, dims::NTuple{N, Int}, values) -> Array

Multiply tensor by values along the axes dims, as multiply_dimensions! does (and so in place, when the element types agree), and sum those axes out, keeping them as singletons: the inner product of tensor and values over dims.

Examples

julia> DiscreteTransitionMessagePassingRules.sum_out_dimensions([1.0 2.0; 3.0 4.0], (2,), [0.5, 0.5])
2×1 Matrix{Float64}:
 1.5
 3.5

source