Feature Request: tf.linalg.solve

[antoniotorres]

Are there plans to include tf.linalg.solve as part of tensflow.js core?
Also known as: tf.matrix_solve

TF Docs: https://www.tensorflow.org/api_docs/python/tf/matrix_solve

[nsthorat]

No plans right now, but happy to take contributions!

[caisq]

@antoniotorres Just curious, what is your use case for this function?

[antoniotorres]

@caisq I'm implementing an RBF interpolation.

[kedevked]

I have just implemented a method to get the determinant of a matrix using the Jordan-Gauss method here. It can be used to solve a linear problem of A * X = B. The method tf.linalg.solve that needs to be implemented will take an array of [A, ...] and [B, ...] matrixes and then will compute the array [S, ...] where each A * S = B. I will work on it if no one has done it yet. Could you please tell here if you're working on it already? Thanks !

[antoniotorres]

Hey @kedevked I haven't worked on it.

[kedevked]

Thanks @antoniotorres for letting me know. @nsthorat and @dsmilkov , could you please let us know if you have already worked on this issue ?

[nsthorat]

We have not, @kedevked! The only thing I would say is it would be good to try to mimic with TensorFlow python does in terms of API and functionality.

[kedevked]

@nsthorat thanks for your reply. I have started to implement the Jordan Gauss method. As you suggested, we need to be compliant with the python version. There is an adjoint parameter in the signature of the method. I only need to implement that. Everything is here. I will greatly appreciate your feedback as I am implementing this feature. Thanks!

[Gurpreetsingh9465]

can i work on this issue 😃

[bileschi]

Please, feel welcome to contribute!

[nsthorat]

I think we should hold off on this issue -- we're going to thinking about what a linear algebra package in JS would look like holistically.

[Gurpreetsingh9465]

@nsthorat sir, i am trying to implement linalgebra module in typescript.

these are some of the functions i decided to put in this module. sir, can you suggest some additional functions or these are enough.

1. linalg.cholesky(a)
2. linalg.svd(a[,full_matrices,compute_uv])
3. linalg.eig(a)
4. linalg.eigvals(a)
5. linalg.norm(x[,ord,axis,keepdims])
6. linalg.cond(x[,p])
7. linalg.det(a)
8. linalg.matrix_rank(M[,tol,hermitian])
9. linalg.slogdet(a)
10. trace(a[,offset,axis1,axis2,dtype,out])
11. linalg.pca(a,n_components)
12. linalg.solve(a,b)
13. linalg.tensorsolve(a,b[,axes])
14. linalg,lstsq(a,b[,rcond])
15. linalg.inv(a)
16. linalg.pinv(a[,rcond])

out of these i had implemented

1. solve
2. pca
3. inv
4. eigvals
5. det
6. slog det
7. trace
8. det

should i do a PR regarding this. 😄