derivation of Tensor.matmul

[houstonhaynes]

derivation of Tensor.matmul

Issue #437 | Created by @psiLearn | 2022-08-31 08:35:33 UTC |

My Example is this short code:

#r "nuget: DiffSharp-cpu"

open DiffSharp
open DiffSharp.Model

let createStiffnessFunction m c d =
    fun (x:Tensor) -> 
        (dsharp.tensor [[0.; 1.];[-c; -d]]).matmul(x)

let f = createStiffnessFunction 0.5 1000. 0.5

let df = dsharp.diff f

f (dsharp.tensor [0.001; 0.])
df (dsharp.tensor [0.001; 0.])

I get this error message. Is it not possible to derive the result of a matmul?

System.Exception: f must be a function of a scalar, encountered f:[|2|]->[|2|]
   at Microsoft.FSharp.Core.PrintfModule.PrintFormatToStringThenFail@1439.Invoke(String message) in D:\a\_work\1\s\src\fsharp\FSharp.Core\printf.fs:line 1439
   at DiffSharp.dsharp.fdiff(FSharpFunc`2 f, Tensor x)
   at DiffSharp.dsharp.diff(FSharpFunc`2 f, Tensor x)
   at <StartupCode$FSI_0096>.$FSI_0096.main@()