bertrandbev / eigen-js Goto Github PK

Hey there, I tried forking this repo and building without OSQP as instructed in the readme using the latest version of emscripten:

emcc -D NO_OSQP -I lib/eigen -Isrc -s DISABLE_EXCEPTION_CATCHING=0 -s ASSERTIONS=0 -O3 -s ALLOW_MEMORY_GROWTH=1 -s MODULARIZE=1 --bind -o build/eigen_gen.js src/cpp/embind.cc

Then I indeed did a yarn build, pushed the new index.js that got generated, and tried to install the fork.

Upon trying to use it, I get greeted by the following error:

index.js:1  Uncaught (in promise) TypeError: Cannot read properties of undefined (reading 'prototype')
    at D.initClass (index.js:1:935979)
    at index.js:1:936495
    at Set.forEach (<anonymous>)
    at index.js:1:936476
initClass @ index.js:1
(anonymous) @ index.js:1
(anonymous) @ index.js:1
Promise.then (async)
(anonymous) @ HSWFC.js:36

I suspected that it was because of the following piece of code at the end of eigen.mjs that adds the GC to the classes, since it still has the QuadProgSolver in there:

eig.ready = Module.then(module => {
  const classes = new Set([
    "Vector",
    "Vector2d",
    "Complex",
    "Matrix",
    "SparseMatrix",
    "TripletVector",
    "ComplexDenseMatrix",
    "Solvers",
    "Decompositions",
    "QuadProgSolver",
    "Random",
  ]);
  classes.forEach(className => {
    eig[className] = GC.initClass(classes, module[className])
  })
  addHelpers(eig);
})

After removing "QuadProgSolver" it indeed started working. I'm not savvy enough in web build tools to come up with a nicer solution for this, but maybe it is good to put a disclaimer/warning in the readme for now about it, so people can still perform a functioning yarn build without OSQP in there.

Thanks for making this library!

It would be nice to have a .d.ts file, to make Eigen.js easier to use in TypeScript. Would you accept a pull request adding such a file, or would you prefer for those types to be added to DefinitelyTyped instead?

As you can see below I've tried to inverse this matrix and the answer isn't compatible with Matlab answer.
Do you know why the NaN (Not a number) appear in response?

EigenJS

Matlab

I'm trying to run eigen-js in Node.js and I'm getting this error when instantiating a new Matrix:

TypeError: eigen_1.default.Matrix is not a constructor

Is this supposed to work on Node.js?

Hello, I have this code for an Extended Kalman filter in TypeScript.

import eigen from 'eigen'

export type NonLinearFunction = (x: eigen.Matrix) => eigen.Matrix
export type JacobianFunction = (x: eigen.Matrix) => eigen.Matrix

export class ExtendedKalmanFilter {
  private x: eigen.Matrix
  private P: eigen.Matrix
  private Q: eigen.Matrix
  private R: eigen.Matrix
  private f: NonLinearFunction
  private h: NonLinearFunction
  private Fj: JacobianFunction
  private Hj: JacobianFunction

  constructor(
    initialState: eigen.Matrix,
    initialCovariance: eigen.Matrix,
    processCovariance: eigen.Matrix,
    measurementCovariance: eigen.Matrix,
    stateTransition: NonLinearFunction,
    observation: NonLinearFunction,
    stateTransitionJacobian: JacobianFunction,
    observationJacobian: JacobianFunction,
  ) {
    this.x = initialState
    this.P = initialCovariance
    this.Q = processCovariance
    this.R = measurementCovariance
    this.f = stateTransition
    this.h = observation
    this.Fj = stateTransitionJacobian
    this.Hj = observationJacobian
  }

  predict(): void {
    this.x = this.f(this.x)
    const F = this.Fj(this.x)
    this.P = F.matMul(this.P).matMulSelf(F.transpose()).matAddSelf(this.Q)
  }

  update(z: eigen.Matrix): void {
    const y = z.matSub(this.h(this.x))
    const H = this.Hj(this.x)
    const S = H.matMul(this.P).matMulSelf(H.transpose()).matAddSelf(this.R)
    const K = this.P.matMul(H.transpose()).matMulSelf(S.inverse())

    this.x = this.x.matAdd(K.matMul(y))
    this.P = eigen.Matrix.identity(this.P.rows(), this.P.cols())
      .matSub(K.matMul(H))
      .matMul(this.P)
  }

  getState(): eigen.Matrix {
    return this.x
  }

  getCovariance(): eigen.Matrix {
    return this.P
  }
}

I'm having memory issues running this in Node.js.

What's the proper way to use GC in this case?

Thanks!

Problem

eigen.Decompositions.lu() returns wrong values which don't satisfy LU = A.

Steps to Reproduce

Run the following code in demo page...

const M = new eig.Matrix([[2, 1], [1, 4]]);
const lu = eig.Decompositions.lu(M);
return {...lu, "L*U": lu.L.matMul(lu.U)}

...and you will get:

While the expected values are:

L = [[1, 0], [0.5, 1]],
U = [[2, 1], [0, 3.5]]

I used Emscripten compiled eigen_gen.js and eigen_gen.wasm to build the index.js of this project.

But when I run test.mjs I get an error of "TypeError: Cannot read properties of undefined (reading 'prototype')at Function.initClass(....\dist\index.js:1:1001402)"

The code snippet in index.js is:
var Q=g(1),B=g(2),C=g(3);function E(A){return Object.getOwnPropertyNames(A).filter(I=>"constructor"!==I&&"function"==typeof A[I])}class D{static add(...A){A.flat(1/0).forEach(A=>{D.objects.add(A)})}static pushException(...A){A.flat(1/0).forEach(A=>{const I=D.whitelist.get(A)||0;D.whitelist.set(A,I+1)})}static popException(...A){A.flat(1/0).forEach(A=>{const I=D.whitelist.get(A)||0;D.whitelist.set(A,I-1),D.whitelist.get(A)<=0&&D.whitelist.remove(A)})}static flush(){const A=[...D.objects].filter(A=>!D.whitelist.has(A));return A.forEach(A=>{A.delete(),D.objects.delete(A)}),A.length}static set(A,I,g){A[I]&&D.popException(A[I]),D.pushException(g),A[I]=g}static initClass(A,I){const g=function(...A){const g=new I(...A);return D.add(g),g},Q=[I,I.prototype];

This is driving me crazy. Can you give me an index.js that you compiled。

Hi @BertrandBev,

There's a badge in the repo that mentions that the license is MIT, but points to Naereen's copyright/license.

Could you please confirm the license of this repo?

Thanks!

Hey this looks like a great project! I have had a hard time finding something as powerful as Eigen for Javascript.

Do you have a list of features or roadmap you are working on?

It is hoped that LU's method for solving linear systems can be added

Hey there 👋

First of all, thanks for this awesome project!
Are there any plans to incorporate the solve methods for solving linear least squares systems yet?

I'm particularly interested in the solve methods from both the QR, and SVD classes.
If you're interested I'd try and create a PR, but I have to admit that C++ is not my strong suit 😅

P.S.: Here's a link to another Eigen port which could be used as inspiration: https://observablehq.com/@rreusser/eigen
(without the memory management implementation though, as far as I can tell).

Hi, firstly this is a AWESOME project!

I'm searching for some library like this one for a project of a desktop aplication built in Electron for my graduation in Engineering.

I need something fast and powerful to operations with matrices that sometimes reaches 1000 rows and columns. I've been thinkg about create a Addon for NodeJS in C++ using Eigen Library but I have some problems because it's not very simple.

Did you think that eigen-js could help me in this situation?
These matrices that I need to solve are generated from electrical circuits, so, most of them are with float numbers.

Why didn't you include the eigen js you built in the repo?