From 13ede7dc3ad9a07761271d82d9552d3d721168a1 Mon Sep 17 00:00:00 2001
From: Christoph Ruegg <git@cdrnet.ch>
Date: Thu, 28 Nov 2013 09:34:05 +0100
Subject: [PATCH] Linear Regression: use more efficient LA routines (via tibel)

---
 src/Numerics/LinearRegression/MultipleRegression.cs | 6 +++---
 src/Numerics/LinearRegression/WeightedRegression.cs | 8 ++++----
 2 files changed, 7 insertions(+), 7 deletions(-)

diff --git a/src/Numerics/LinearRegression/MultipleRegression.cs b/src/Numerics/LinearRegression/MultipleRegression.cs
index c215f28d..f33419d3 100644
--- a/src/Numerics/LinearRegression/MultipleRegression.cs
+++ b/src/Numerics/LinearRegression/MultipleRegression.cs
@@ -45,7 +45,7 @@ namespace MathNet.Numerics.LinearRegression
         /// <returns>Best fitting vector for model parameters β</returns>
         public static Vector<T> NormalEquations<T>(Matrix<T> x, Vector<T> y) where T : struct, IEquatable<T>, IFormattable
         {
-            return x.TransposeThisAndMultiply(x).Cholesky().Solve(x.Transpose()*y);
+            return x.TransposeThisAndMultiply(x).Cholesky().Solve(x.TransposeThisAndMultiply(y));
         }
 
         /// <summary>
@@ -57,7 +57,7 @@ namespace MathNet.Numerics.LinearRegression
         /// <returns>Best fitting vector for model parameters β</returns>
         public static Matrix<T> NormalEquations<T>(Matrix<T> x, Matrix<T> y) where T : struct, IEquatable<T>, IFormattable
         {
-            return x.TransposeThisAndMultiply(x).Cholesky().Solve(x.Transpose() * y);
+            return x.TransposeThisAndMultiply(x).Cholesky().Solve(x.TransposeThisAndMultiply(y));
         }
 
         /// <summary>
@@ -76,7 +76,7 @@ namespace MathNet.Numerics.LinearRegression
                 predictor = predictor.InsertColumn(0, Vector<T>.Build.Dense(predictor.RowCount, Vector<T>.One));
             }
             var response = Vector<T>.Build.Dense(y);
-            return predictor.TransposeThisAndMultiply(predictor).Cholesky().Solve(predictor.Transpose()*response).ToArray();
+            return predictor.TransposeThisAndMultiply(predictor).Cholesky().Solve(predictor.TransposeThisAndMultiply(response)).ToArray();
         }
 
         /// <summary>
diff --git a/src/Numerics/LinearRegression/WeightedRegression.cs b/src/Numerics/LinearRegression/WeightedRegression.cs
index 15409756..3a6b80a9 100644
--- a/src/Numerics/LinearRegression/WeightedRegression.cs
+++ b/src/Numerics/LinearRegression/WeightedRegression.cs
@@ -42,7 +42,7 @@ namespace MathNet.Numerics.LinearRegression
         /// </summary>
         public static Vector<T> Weighted<T>(Matrix<T> x, Vector<T> y, Matrix<T> w) where T : struct, IEquatable<T>, IFormattable
         {
-            return x.TransposeThisAndMultiply(w*x).Cholesky().Solve(x.Transpose()*(w*y));
+            return x.TransposeThisAndMultiply(w*x).Cholesky().Solve(x.TransposeThisAndMultiply(w*y));
         }
 
         /// <summary>
@@ -50,7 +50,7 @@ namespace MathNet.Numerics.LinearRegression
         /// </summary>
         public static Matrix<T> Weighted<T>(Matrix<T> x, Matrix<T> y, Matrix<T> w) where T : struct, IEquatable<T>, IFormattable
         {
-            return x.TransposeThisAndMultiply(w*x).Cholesky().Solve(x.Transpose()*(w*y));
+            return x.TransposeThisAndMultiply(w*x).Cholesky().Solve(x.TransposeThisAndMultiply(w*y));
         }
 
         /// <summary>
@@ -65,8 +65,8 @@ namespace MathNet.Numerics.LinearRegression
                 predictor = predictor.InsertColumn(0, Vector<T>.Build.Dense(predictor.RowCount, Vector<T>.One));
             }
             var response = Vector<T>.Build.Dense(y);
-            var weights = Matrix<T>.Build.Diagonal(new DiagonalMatrixStorage<T>(predictor.RowCount, predictor.RowCount, w));
-            return predictor.TransposeThisAndMultiply(weights*predictor).Cholesky().Solve(predictor.Transpose()*(weights*response)).ToArray();
+            var weights = Matrix<T>.Build.DenseOfDiagonalArray(w);
+            return predictor.TransposeThisAndMultiply(weights*predictor).Cholesky().Solve(predictor.TransposeThisAndMultiply(weights*response)).ToArray();
         }
 
         /// <summary>