Potential Principal Component Analysis component

paulk · December 31, 2022, 12:17am

Hi folks,

I am interested in submitting a PR for Principal Component Analysis. What I have in mind is a more generalized Java implementation of the code I use in this Groovy example:

github.com

paulk-asert/groovy-data-science/blob/master/subprojects/Whiskey/src/main/groovy/KMeansPcaCentroidsMedoids_HipparchusJFreeChart.groovy#L77-L102


      
          var points = data*.point as double[][]
          var mean = points.transpose()*.average()
          for (col in 0..<points[0].length) {
              for (row in 0..<points.length) {
                  points[row][col] -= mean[col]
              }
          }
          var realMatrix = MatrixUtils.createRealMatrix(points)
          
          
// calculate PCA by hand: create covariance matrix of points, then find eigen vectors
          // see https://stats.stackexchange.com/questions/2691/making-sense-of-principal-component-analysis-eigenvectors-eigenvalues
          
          
var covariance = new Covariance(realMatrix)
          var covarianceMatrix = covariance.covarianceMatrix
          var ed = new EigenDecomposition(covarianceMatrix)
          double[] eigenValues = ed.realEigenvalues
          var k = clusterer.k
          var principalComponents = MatrixUtils.createRealMatrix(eigenValues.length, k)
          for (int i = 0; i < k; i++) {
              for (int j = 0; j < eigenValues.length; j++) {

This file has been truncated. show original

It uses both the core and statistics modules. Would it make sense to create a PR with the implementation in a new module? Has this topic been covered before?

Cheers, Paul.

luc · December 31, 2022, 7:21pm

Hi @paulk

This sounds very interesting. I don’t know if it would better fit in the existing statistics module or if a dedicated module would be needed. Both are possible.

Please take a look at the Hipparchus Governance and at the Individual Contributor License Agreement it refers to. This paperwork is needed so we can include your work in a next version of Hipparchus. I don’t know if the mail address cited in the governance still works, but you can always send this to me using a private message. Don’t put this on a public message as it will include your name, address and signature!

Thanks for your contribution!

paulk · January 2, 2023, 1:12pm

I’ll fill out the paperwork and submit a PR in the next few days.

paulk · January 3, 2023, 10:59pm

Okay, done.