fix Matrix.map references

fslaborg · Jun 8, 2021 · 3d4f3b8 · 3d4f3b8
1 parent 7d89443
commit 3d4f3b8
Show file tree

Hide file tree

Showing 4 changed files with 23 additions and 28 deletions.
diff --git a/src/FSharp.Stats/ML/Unsupervised/GapStatistics.fs b/src/FSharp.Stats/ML/Unsupervised/GapStatistics.fs
@@ -299,7 +299,7 @@ https://www.datanovia.com/en/lessons/determining-the-optimal-number-of-clusters-
         // Calculate log(sum_i(within-cluster_i sum of squares around cluster_i mean)) of kmeans clustering result.
         let logDispersionKMeansInitCvMax = 
             let aggregator = IterativeClustering.avgCentroid
-            let factory = IterativeClustering.intitCVMAX 
+            let factory = IterativeClustering.initCVMAX 
             let clusterDispersion (cluster:float[][]) =
                 let distances =
                     cluster 

diff --git a/src/FSharp.Stats/ML/Unsupervised/IterativeClustering.fs b/src/FSharp.Stats/ML/Unsupervised/IterativeClustering.fs
@@ -51,17 +51,14 @@ module IterativeClustering =
 
 
 
-    // cvmax - Algorith by Moth’d Belal. Al-Daoud (Ref.: A New Algorithm for Cluster Initialization)
-    let intitCVMAX (sample: float[] array) k =
-        let dmatrix = matrix sample
+    // cvmax - Algorithm by Moth’d Belal. Al-Daoud (Ref.: A New Algorithm for Cluster Initialization)
+    let initCVMAX (sampleRows: float[] []) k =
+        let dmatrix = matrix sampleRows
         let cvmax =
-            dmatrix
-            |> Matrix.Generic.enumerateColumnWise Seq.var
-            |> Seq.zip (Matrix.Generic.enumerateColumnWise id dmatrix)
-            |> Seq.maxBy snd
-            |> fst
-            |> Seq.mapi (fun rowI value -> (rowI,value)) 
-            |> Seq.toArray 
+            sampleRows
+            |> JaggedArray.transpose
+            |> Array.maxBy Seq.var
+            |> Array.indexed
             |> Array.sortBy snd
 
         if cvmax.Length < k then failwithf "Number of data points must be at least %i" k        
@@ -78,7 +75,7 @@ module IterativeClustering =
                 | x                       -> chunkSize * (i - 1) + ((cvmax.Length - chunkSize * (i - 1)) / 2)
             //printfn "Array.lenght = %i and index = %i" cvmax.Length (index-1)
             yield cvmax.[index-1] |> fst]
-        |> Seq.map (fun rowI -> dmatrix.Row(rowI).ToArray())
+        |> Seq.map (fun rowI -> sampleRows.[rowI])
         |> Seq.toArray
 
 //    // cvmax - Algorith by Moth’d Belal. Al-Daoud (Ref.: A New Algorithm for Cluster Initialization)

diff --git a/src/FSharp.Stats/ML/Unsupervised/PrincipalComponentAnalysis.fs b/src/FSharp.Stats/ML/Unsupervised/PrincipalComponentAnalysis.fs
@@ -66,7 +66,7 @@ module PCA =
             data |> Matrix.meanColumnWise                    
         // Atttention: not entierly shure if space of data before
         let colStDev =
-            data |> Matrix.Generic.enumerateColumnWise Seq.stDevPopulation |> Seq.toArray
+            data |> Matrix.Generic.mapCols Seq.stDevPopulation |> Seq.toArray
 
         let adjust (direction:AdjustmentDirection) (aData:Matrix<float>) = 
             match direction with
@@ -88,11 +88,11 @@ module PCA =
     /// Returns an AdjustmentFactory which centers and standardize the data
     let toAdjustCorrelation (data:Matrix<float>) : AdjustmentFactory =                
         let colMeans =
-            data |> Matrix.meanColumnWise                    
-        let sqrtI = sqrt (float data.NumRows)
-        // Atttention: not entierly shure if space of data before
+            data |> Matrix.meanColumnWise 
+        let sqrtI = sqrt (float data.NumRows)                   
+        // Attention: not entirely sure if space of data before
         let colStDev =
-            data |> Matrix.Generic.enumerateColumnWise Seq.stDevPopulation |> Seq.toArray    
+            data |> Matrix.Generic.mapCols Seq.stDevPopulation |> Seq.toArray    
         let adjust (direction:AdjustmentDirection)  (aData:Matrix<float>) = 
             match direction with
             | Obverse ->

diff --git a/src/FSharp.Stats/Signal/Normalization.fs b/src/FSharp.Stats/Signal/Normalization.fs
@@ -23,16 +23,15 @@ module Normalization =
     ///
     /// The additional function is applied on all values of the matrix when calculating the normalization factors. By this, a zero in the original dataset will still remain zero.
     let medianOfRatiosBy (f: float -> float) (data:Matrix<float>) =
-        let sampleWiseCorrectionFactors =
+        let sampleWiseCorrectionFactors =            
             data
-            |> Matrix.mapiRows (fun _ v -> 
-                let v = Seq.map f v
+            |> Matrix.mapiRows (fun _ v ->
+                let v = RowVector.map f v
                 let geometricMean = Seq.meanGeometric v           
-                Seq.map (fun s -> s / geometricMean) v
+                RowVector.map (fun s -> s / geometricMean) v
                 ) 
-            |> matrix
+            |> Matrix.ofRows
             |> Matrix.mapiCols (fun _ v -> Vector.median v)
-            |> vector
         data
         |> Matrix.mapi (fun r c v ->
             v / sampleWiseCorrectionFactors.[c]
@@ -53,13 +52,12 @@ module Normalization =
         let sampleWiseCorrectionFactors =
             data
             |> Matrix.mapiCols (fun _ v -> 
-                let v = Seq.map f v
+                let v = Vector.map f v
                 let geometricMean = Seq.meanGeometric v           
-                Seq.map (fun s -> s / geometricMean) v
+                Vector.map (fun s -> s / geometricMean) v
                 ) 
-            |> matrix
-            |> Matrix.mapiCols (fun _ v -> Seq.median v)
-            |> vector
+            |> Matrix.ofCols
+            |> Matrix.mapiRows (fun _ v -> Seq.median v)
         data
         |> Matrix.mapi (fun r c v ->
             v / sampleWiseCorrectionFactors.[r]