dummy-link

Einsum

Einstein summation notation in Julia

Readme

Einsum.jl

Einstein summation notation similar to numpy's einsum function (but more flexible!).

PackageEvaluator Package Build Package Status
Einsum Build Status License
Einsum Project Status: Inactive - The project has reached a stable, usable state but is no longer being actively developed; support/maintenance will be provided as time allows. - help wanted!

To install: Pkg.add("Einsum").

Documentation:

If destination is preallocated use =

using Einsum
A = zeros(5,6,7); # need to preallocate destination
X = randn(5,2)
Y = randn(6,2)
Z = randn(7,2)
@einsum A[i,j,k] = X[i,r]*Y[j,r]*Z[k,r]

If destination is not preallocated use :=

using Einsum
X = randn(5,2)
Y = randn(6,2)
Z = randn(7,2)
@einsum A[i,j,k] := X[i,r]*Y[j,r]*Z[k,r] # creates new array A with appropriate dimensions

What happens under the hood:

The @einsum macro automatically generates code that looks much like the following (note that we "sum out" over the index r, since it only occurs on the right hand side of the equation):

for k = 1:size(A,3)
    for j = 1:size(A,2)
        for i = 1:size(A,1)
            s = 0
            for r = 1:size(X,2)
                s += X[i,r] * Y[j,r] * Z[k,r]
            end
            A[i,j,k] = s
        end
    end
end

To see exactly what is generated, use macroexpand:

macroexpand(:(@einsum A[i,j,k] = X[i,r]*Y[j,r]*Z[k,r]))

Other functionality

In principle, the @einsum macro can flexibly implement function calls within the nested for loop structure. For example, consider transposing a block matrix:

z = Any[ rand(2,2) for i=1:2, j=1:2]
@einsum t[i,j] := transpose(z[j,i])

This produces a for loop structure with a transpose function call in the middle. Approximately:

for j = 1:size(z,1)
    for i = 1:size(z,2)
        t[i,j] = transpose(z[j,i])
    end
end

Again, you can use macroexpand to see the exact code that is generated.

Related Packages:

First Commit

04/16/2016

Last Touched

9 days ago

Commits

71 commits

Requires:

Used By: