fix: address PR review comments for GAN implementations

- Fix unused variable assignments (WGANGP, ProgressiveGAN, InceptionScore)
- Add readonly modifier to fields in ACGAN
- Fix potential integer overflow in InfoGAN
- Fix precision loss in ProgressiveGAN

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Author: franklinic

src/Metrics/FrechetInceptionDistance.cs

@@ -1,6 +1,5 @@
 using System;
 using System.Linq;
-using AiDotNet.Mathematics;
 using AiDotNet.NeuralNetworks;
 
 namespace AiDotNet.Metrics
@@ -68,7 +67,7 @@ public FrechetInceptionDistance(
             ConvolutionalNeuralNetwork<T>? inceptionNetwork = null,
             int featureDimension = 2048)
         {
-            NumOps = NumericOperations<T>.Instance;
+            NumOps = MathHelper.GetNumericOperations<T>();
             InceptionNetwork = inceptionNetwork;
             FeatureDimension = featureDimension;
             FeatureLayer = -2; // Second to last layer (before classification)
@@ -103,18 +102,18 @@ public double ComputeFID(Tensor<T> realImages, Tensor<T> generatedImages)
         /// <returns>Feature matrix (num_images × feature_dim)</returns>
         private Matrix<T> ExtractFeatures(Tensor<T> images)
         {
+            var numImages = images.Shape[0];
+
             if (InceptionNetwork == null)
             {
                 // If no Inception network provided, return dummy features
                 // In a real implementation, you would load a pre-trained InceptionV3
-                var numImages = images.Shape[0];
                 return CreateDummyFeatures(numImages);
             }
 
             // Set to inference mode
             InceptionNetwork.SetTrainingMode(false);
 
-            var numImages = images.Shape[0];
             var features = new Matrix<T>(numImages, FeatureDimension);
 
             // Process each image
@@ -123,7 +122,12 @@ private Matrix<T> ExtractFeatures(Tensor<T> images)
                 // Extract single image
                 var imageSize = images.Length / numImages;
                 var singleImage = new Tensor<T>(new[] { 1, images.Shape[1], images.Shape[2], images.Shape[3] });
-                Array.Copy(images.Data, i * imageSize, singleImage.Data, 0, imageSize);
+
+                // Copy data from source tensor to single image tensor
+                for (int k = 0; k < imageSize; k++)
+                {
+                    singleImage.SetFlat(k, images.GetFlat(i * imageSize + k));
+                }
 
                 // Forward pass through Inception network
                 var output = InceptionNetwork.Predict(singleImage);
@@ -133,7 +137,7 @@ private Matrix<T> ExtractFeatures(Tensor<T> images)
                 // For now, use output
                 for (int j = 0; j < Math.Min(output.Length, FeatureDimension); j++)
                 {
-                    features[i, j] = output.Data[j];
+                    features[i, j] = output.GetFlat(j);
                 }
             }
 

src/Metrics/InceptionScore.cs

@@ -1,6 +1,5 @@
 using System;
 using System.Linq;
-using AiDotNet.Mathematics;
 using AiDotNet.NeuralNetworks;
 
 namespace AiDotNet.Metrics
@@ -78,7 +77,7 @@ public InceptionScore(
             int numClasses = 1000,
             int numSplits = 10)
         {
-            NumOps = NumericOperations<T>.Instance;
+            NumOps = MathHelper.GetNumericOperations<T>();
             InceptionNetwork = inceptionNetwork;
             NumClasses = numClasses;
             NumSplits = numSplits;
@@ -91,7 +90,7 @@ public InceptionScore(
         /// <returns>Inception Score (higher is better, typical range 1-15+)</returns>
         public double ComputeIS(Tensor<T> generatedImages)
         {
-            var (mean, std) = ComputeISWithUncertainty(generatedImages);
+            var (mean, _) = ComputeISWithUncertainty(generatedImages);
             return mean;
         }
 
@@ -184,7 +183,10 @@ private Matrix<T> GetPredictions(Tensor<T> images)
                 // Extract single image
                 var imageSize = images.Length / numImages;
                 var singleImage = new Tensor<T>(new[] { 1, images.Shape[1], images.Shape[2], images.Shape[3] });
-                Array.Copy(images.Data, i * imageSize, singleImage.Data, 0, imageSize);
+                for (int idx = 0; idx < imageSize; idx++)
+                {
+                    singleImage.SetFlat(idx, images.GetFlat(i * imageSize + idx));
+                }
 
                 // Forward pass
                 var output = InceptionNetwork.Predict(singleImage);
@@ -195,7 +197,7 @@ private Matrix<T> GetPredictions(Tensor<T> images)
                 // Store predictions
                 for (int j = 0; j < Math.Min(probs.Length, NumClasses); j++)
                 {
-                    predictions[i, j] = probs.Data[j];
+                    predictions[i, j] = probs.GetFlat(j);
                 }
             }
 
@@ -213,26 +215,26 @@ private Tensor<T> Softmax(Tensor<T> logits)
             // Find max for numerical stability
             for (int i = 0; i < logits.Length; i++)
             {
-                if (NumOps.Compare(logits.Data[i], maxLogit) > 0)
+                if (NumOps.GreaterThan(logits.GetFlat(i), maxLogit))
                 {
-                    maxLogit = logits.Data[i];
+                    maxLogit = logits.GetFlat(i);
                 }
             }
 
             // Compute exp(x - max) and sum
             var sum = NumOps.Zero;
             for (int i = 0; i < logits.Length; i++)
             {
-                var shifted = NumOps.Subtract(logits.Data[i], maxLogit);
+                var shifted = NumOps.Subtract(logits.GetFlat(i), maxLogit);
                 var expVal = NumOps.Exp(shifted);
-                result.Data[i] = expVal;
+                result.SetFlat(i, expVal);
                 sum = NumOps.Add(sum, expVal);
             }
 
             // Normalize
             for (int i = 0; i < result.Length; i++)
             {
-                result.Data[i] = NumOps.Divide(result.Data[i], sum);
+                result.SetFlat(i, NumOps.Divide(result.GetFlat(i), sum));
             }
 
             return result;
@@ -325,7 +327,10 @@ private Tensor<T> ExtractImageSubset(Tensor<T> images, int startIdx, int count)
             }
 
             var subset = new Tensor<T>(subsetShape);
-            Array.Copy(images.Data, startIdx * imageSize, subset.Data, 0, count * imageSize);
+            for (int idx = 0; idx < count * imageSize; idx++)
+            {
+                subset.SetFlat(idx, images.GetFlat(startIdx * imageSize + idx));
+            }
 
             return subset;
         }

src/NeuralNetworks/ACGAN.cs

@@ -37,10 +37,10 @@ public class ACGAN<T> : NeuralNetworkBase<T>
     private T _beta1Power;
     private T _beta2Power;
     private double _currentLearningRate;
-    private double _initialLearningRate;
-    private double _learningRateDecay;
-    private List<T> _generatorLosses = [];
-    private List<T> _discriminatorLosses = [];
+    private readonly double _initialLearningRate;
+    private readonly double _learningRateDecay;
+    private readonly List<T> _generatorLosses = [];
+    private readonly List<T> _discriminatorLosses = [];
 
     /// <summary>
     /// The number of classes for classification.
@@ -72,7 +72,7 @@ public class ACGAN<T> : NeuralNetworkBase<T>
     /// </remarks>
     public ConvolutionalNeuralNetwork<T> Discriminator { get; private set; }
 
-    private ILossFunction<T> _lossFunction;
+    private readonly ILossFunction<T> _lossFunction;
 
     /// <summary>
     /// Initializes a new instance of the <see cref="ACGAN{T}"/> class.
@@ -536,4 +536,26 @@ protected override IFullModel<T, Tensor<T>, Tensor<T>> CreateNewInstance()
             _lossFunction,
             _initialLearningRate);
     }
+
+    /// <summary>
+    /// Updates the parameters of all networks in the ACGAN.
+    /// </summary>
+    /// <param name="parameters">The new parameters vector containing parameters for all networks.</param>
+    public override void UpdateParameters(Vector<T> parameters)
+    {
+        int generatorCount = Generator.GetParameterCount();
+        int discriminatorCount = Discriminator.GetParameterCount();
+
+        // Update Generator parameters
+        var generatorParams = new Vector<T>(generatorCount);
+        for (int i = 0; i < generatorCount; i++)
+            generatorParams[i] = parameters[i];
+        Generator.UpdateParameters(generatorParams);
+
+        // Update Discriminator parameters
+        var discriminatorParams = new Vector<T>(discriminatorCount);
+        for (int i = 0; i < discriminatorCount; i++)
+            discriminatorParams[i] = parameters[generatorCount + i];
+        Discriminator.UpdateParameters(discriminatorParams);
+    }
 }