Changing output format for preprocessing functions + some minor fixes

src/convenience.jl

@@ -65,21 +65,6 @@ nTrainValTest = [nTrain, nVal, nTest];
 return splits, nTrainValTest
 end
 
-"""
-    createDataSplitInds(X, nSplits, props=[6/10, 2/10, 2/10], rngseed=42)
-
-Creates training/validation/test split for dataset
-
-### Arguments
-    - 'X'       : Int, number of samples
-    - 'props'   : 3x1 Vector of proportions (train/val/test), must sum to 1
-    - 'nSplits' : Int, number of data splits
-    - 'rngseed' : Int, for reproducibility
-
-### Returns
-    - 'splits'        : size(X,1) x nSplits matrix with indices
-    - 'nTrainValTest' : 3x1 vector indicating number of samples in training/validation/testing
-"""
 function createDataSplitInds(X::Int64, nSplits, props=[6/10, 2/10, 2/10], rngseed=42)
 
 n             = X;
@@ -122,9 +107,7 @@ Creates training/validation/test split for dataset
     - 'splitInd'      : Index of split to use
     - 'X' and 'Y'     : Data matrix and response vector/matrix
 
-### Returns
+### Returns dictionary with keys XTrain, XVal, XTest, YTrain, YVal, YTest
-    - 'splits'        : size(X,1) x nSplits matrix with indices
-    - 'nTrainValTest' : 3x1 vector indicating number of samples in training/validation/testing
 """
 function createDataSplit(splits, nTrainValTest, splitInd, X, Y)
 

src/preprocessing.jl

@@ -67,8 +67,7 @@ means = mean(X, dims=2);
 stds  = std(X, dims=2);
 X_SNV = @. (X - means) / stds;
 
-return_values = Dict([("X_Cor", X_SNV), ("means", means), ("stds", stds)]);
+return X_SNV, Dict([("means", means), ("stds", stds)]);
-return return_values
 end
 
 
@@ -81,7 +80,7 @@ end
 
 MSC preprocessing (subtract constant trend, scale based on projection onto ref. spectrum)
 Second argument is mean (default) or svd (first right singular vector), or a vector to be used as reference spectrum
-Returns dictionary with keys X\\_Cor, X\\_Ref, coeffs
+Returns X\\_Cor and dictionary with keys X\\_Ref, coeffs
 """
 function MSC(X, ref::String="mean")
   
@@ -99,8 +98,8 @@ B      = [ones(size(X,2), 1) X_Ref];
 coeffs = B \ X';
 X_MSC  = @. (X - coeffs[1,:]) / coeffs[2,:];
 
-return_values = Dict([("X_Cor", X_MSC), ("X_Ref", X_Ref), ("coeffs", coeffs)])
+return_values = Dict([("X_Ref", X_Ref), ("coeffs", coeffs)])
-return return_values
+return X_MSC, return_values
 end
 
 
@@ -111,7 +110,7 @@ B     = [ones(size(X,2), 1) X_Ref];
 coeffs = B \ X';
 X_MSC = @. (X - coeffs[1,:]) / coeffs[2,:];
 
-return_values = Dict([("X_Cor", X_MSC), ("X_Ref", X_Ref), ("coeffs", coeffs)])
+return_values = X_MSC, Dict([("X_Ref", X_Ref), ("coeffs", coeffs)])
 return return_values
 end
 
@@ -125,7 +124,7 @@ EMSC correction with mean spectrum as reference and polynomial trends of form Li
 First argument is spectra to be corrected, second argument is either degree for polynomial regression
 or a basis to be used for correction (basis used should be output of this function, the reference
 spectrum is assumed to be the first basis vector).
-Returns dictionary with keys X\\_Cor, basis, coeffs
+Returns X\\_Cor and dictionary with keys basis, coeffs
 """
 function EMSCTraditional(X, polDeg::Int64=2)
 
@@ -172,7 +171,7 @@ Inputs:
     der_order - derivative order
     keep_endpoints - boolean, truncates spectra if false. NOTE: For derivatives MUST USE FALSE
 
-Output: Dictionary with keys X\\_Cor, der\\_order, degree, window\\_size, filter\\_coeffs, keep\\_endpoints
+Output:  X\\_Cor, Dictionary with keys der\\_order, degree, window\\_size, filter\\_coeffs, keep\\_endpoints
 """
 function savitzkyGolay(X, d=2, w=2, der_order=0, keep_endpoints=false)
 
@@ -198,7 +197,7 @@ if !keep_endpoints
     X_Cor = X_Cor[:, 1:end-w-1];
 end
 
-return Dict([("X_Cor", X_Cor), ("der_order", der_order), ("degree", d), ("window_size", w), ("filter_coeffs", a), ("keep_endpoints", keep_endpoints)]);
+return X_Cor, Dict([("der_order", der_order), ("degree", d), ("window_size", w), ("filter_coeffs", a), ("keep_endpoints", keep_endpoints)]);
 end
 
 
@@ -272,8 +271,7 @@ for i=1:n
     X_Cor[i,:] = X[i,:] - baseline[i,:];
 end
 
-return_values = Dict([("baseline", baseline), ("X_Cor", X_Cor)]);
+return X_Cor, baseline;
-return return_values
 end
 
 
@@ -287,10 +285,13 @@ end
 
 """
     function EMSC(X, polDeg=2, refType="svd", nRef=1,  baseDeg=-1, intF=0)
+    function EMSCC(X, model::EMSCModel)
 
 Implementation of EMSC pre-processing based on Skogholt et al. (2018).
 intF is 0 (no interferent), vector, or matrix with intereferents as rows.
-Returns dictionary with keys X\\_Cor, model, coeffs
+Returns X\\_Cor and dictionary with keys model, coeffs
+Second function performs EMSC correction on X with given model.
+Returns X\\_Cor and dictionary with keys model, coeffs
 """
 function EMSC(X, polDeg=2, refType="svd", nRef=1,  baseDeg=-1, intF=0)
 
@@ -360,13 +361,7 @@ end
 
 
 
-"""
+function EMSC(X, model::EMSCModel)
- - EMSCCorrection(X, model::EMSCModel)
-
-Performs EMSC correction on X with given model.
-Returns dictionary with keys X\\_Cor, model, coeffs
-"""
-function EMSCCorrection(X, model::EMSCModel)
 
 n, _ = size(X);
     
@@ -389,6 +384,6 @@ X_Cor  = X_Cor ./ mult;
 [X_Cor[i,:] = X_Cor[i,:] - model.baseline for i in 1:n]; # Surely this should be possible with broadcasting instead...
 # Thought it would be + baseline above, but this works...
     
-return_values = Dict([("X_Cor", X_Cor), ("model", model), ("coeffs", coeffs)]);
+return_values = X_Cor, Dict([("model", model), ("coeffs", coeffs)]);
 return return_values
 end