Matrix multiplication gives incorrect results for matrix got from DenseMatrix.horzcat function

[tutysara]

I had to add bias terms to a matrix before multiplying, I used DenseMatrix.horzcat function to append the column 1's to the original matrix. However there is vast difference between the result got by multiplying a matrix hard coded with the column of 1's and the one which is got as a result of applying the function DenseMatrix.horzcat. I am using a DenseMatrix of Double, I know there could be some round off errors, but I think the difference is considerable. Guys can you please try running it on your machines and see if it works fine or it is really a bug that needs a fix.

code

package org.tutysara
import scalala.scalar._;
import scalala.tensor.::;
import scalala.tensor.mutable._;
import scalala.tensor.dense._;
import scalala.tensor.sparse._;
import scalala.library.Library._;
import scalala.library.LinearAlgebra._;
import scalala.library.Statistics._;
import scalala.library.Plotting._;
import scalala.operators.Implicits._;
import scalala.generic.math.Library._
object test {

  def main(args: Array[String]): Unit = {
   val X_hc=DenseMatrix(
       (1.000000 ,  0.084147,  -0.027942,  -0.099999),
       (1.000000  , 0.090930,   0.065699,  -0.053657),
       (1.000000  , 0.014112,   0.098936,   0.042017),
       (1.000000  ,-0.075680,   0.041212,   0.099061),
       (1.000000  ,-0.095892,  -0.054402,   0.065029)
       )

  val X=DenseMatrix(
       (0.841471,  -0.279415,  -0.999990 ), 
       (0.909297,   0.656987,  -0.536573),  
       (0.141120,   0.989358,   0.420167),
       (-0.756802,   0.412118,   0.990607),  
       (-0.958924,  -0.544021,   0.650288 )
      )
  val X_new = DenseMatrix.horzcat( DenseMatrix.ones[Double](X.numRows,1),X); //add bias terms
  val Theta1 =DenseMatrix(
          (0.084147 , -0.027942,  -0.099999,  -0.028790),
          (0.090930 ,  0.065699 , -0.053657,  -0.096140),
          (0.014112 ,  0.098936  , 0.042017,  -0.075099),
          (-0.075680 ,  0.041212 ,  0.099061,   0.014988),
          (-0.095892 , -0.054402 ,  0.065029,   0.091295)
   )

  println("X = \n"+X)
  println("Theta1 = \n"+Theta1)
  println("X_new = \n"+X_new)
  println("X_hc = \n"+X_hc)
  val res=Theta1*X_new.t
  println("res = \n"+res)
  val rec_x_hc=Theta1*X_hc.t
  println("res_x_hc = \n"+rec_x_hc)
   /*
    * res should be
   8.7469e-02   7.6581e-02   7.2650e-02   7.9289e-02   9.0394e-02
   1.0757e-01   9.8537e-02   8.2509e-02   7.4223e-02   8.1297e-02
   2.8773e-02   2.9898e-02   1.6510e-02   9.1676e-04  -2.5446e-03
  -7.6479e-02  -6.6229e-02  -6.4668e-02  -7.3232e-02  -8.4047e-02
  -1.1142e-01  -1.0147e-01  -8.6391e-02  -8.0052e-02  -8.8277e-02

    */
  /*
  Output
  ================================================================
  X = 
 0.841471  -0.279415  -0.999990  
 0.909297   0.656987  -0.536573  
 0.141120   0.989358   0.420167  
-0.756802   0.412118   0.990607  
-0.958924  -0.544021   0.650288  
Theta1 = 
 0.0841470  -0.0279420  -0.0999990  -0.0287900  
 0.0909300   0.0656990  -0.0536570  -0.0961400  
 0.0141120   0.0989360   0.0420170  -0.0750990  
-0.0756800   0.0412120   0.0990610   0.0149880  
-0.0958920  -0.0544020   0.0650290   0.0912950  
X_new = 
 1.00000   0.841471  -0.279415  -0.999990  
 1.00000   0.909297   0.656987  -0.536573  
 1.00000   0.141120   0.989358   0.420167  
 1.00000  -0.756802   0.412118   0.990607  
 1.00000  -0.958924  -0.544021   0.650288  
X_hc = 
 1.00000   0.0841470  -0.0279420  -0.0999990  
 1.00000   0.0909300   0.0656990  -0.0536570  
 1.00000   0.0141120   0.0989360   0.0420170  
 1.00000  -0.0756800   0.0412120   0.0990610  
 1.00000  -0.0958920  -0.0544020   0.0650290  
res = 
 0.117366    0.00848932  -0.0308276     0.0355626   0.146621    
 0.257345    0.167004     0.00672061   -0.0761411  -0.00539850  
 0.160722    0.171975     0.0380896    -0.117841   -0.152454    
-0.0836683   0.0188336    0.0344401    -0.0511973  -0.159344    
-0.251134   -0.151623    -0.000873103   0.0625166  -0.0197337   
res_x_hc = 
 0.0874689   0.0765812   0.0726495   0.0792885     0.0903944   
 0.107572    0.0985374   0.0825090   0.0742229     0.0812972   
 0.0287730   0.0298983   0.0165097   0.000916746  -0.00254459  
-0.0764789  -0.0662286  -0.0646680  -0.0732317    -0.0840464   
-0.111416   -0.101465   -0.0863901  -0.0800511    -0.0882762   

   */
  }

}

[dlwh]

Your X and X_hc's columns (except the first) differ by a factor of 10. Fixing that fixes the issue.

[tutysara]

thx, haven't noticed that. You guys are lightning fast.

[tutysara]

I had send a mail to Scalala's group as well, You can trash that mail when it comes to moderation.