Better comments on examples.

Author: joaopauloschuler

examples/SimpleFashionMNIST/SimpleFashionMNIST.lpr

@@ -2,6 +2,12 @@
 (*
  Coded by Joao Paulo Schwarz Schuler.
  https://github.com/joaopauloschuler/neural-api
+ -----------------------------------------------
+ The code shows an example of training and fitting a convolutional
+ neural network (CNN) using the Fashion MNIST dataset. It creates a neural
+ network with specific layers and configurations, loads the fashion MNIST data,
+ and then trains the network using the provided data. The code also sets
+ various parameters for training, such as learning rate, decay, and batch size.
 *)
 {$mode objfpc}{$H+}
 
@@ -15,34 +21,42 @@   TTestCNNAlgo = class(TCustomApplication)
     procedure DoRun; override;
   end;
 
+  // Implementation of the TTestCNNAlgo class
   procedure TTestCNNAlgo.DoRun;
   var
-    NN: TNNet;
-    NeuralFit: TNeuralImageFit;
-    ImgTrainingVolumes, ImgValidationVolumes, ImgTestVolumes: TNNetVolumeList;
+    NN: TNNet;  // Neural network object
+    NeuralFit: TNeuralImageFit;  // Object for training and fitting the neural network
+    ImgTrainingVolumes, ImgValidationVolumes, ImgTestVolumes: TNNetVolumeList;  // Lists of training, validation, and test image volumes
   begin
+    // Checking if the MNIST files exist and loading the data
     if Not(CheckMNISTFile('train', {IsFashion=}true)) or
       Not(CheckMNISTFile('t10k', {IsFashion=}true)) then
     begin
       Terminate;
       exit;
     end;
+
     WriteLn('Creating Neural Network...');
-    NN := THistoricalNets.Create();
+
+    // Creating the neural network with specific layers and configurations
+    NN := TNNet.Create();
     NN.AddLayer([
-      TNNetInput.Create(28, 28, 1),
-      TNNetConvolutionLinear.Create(64, 5, 2, 1, 1),
-      TNNetMaxPool.Create(4),
-      TNNetConvolutionReLU.Create(64, 3, 1, 1, 1),
-      TNNetConvolutionReLU.Create(64, 3, 1, 1, 1),
-      TNNetFullConnectReLU.Create(32),
-      TNNetFullConnectReLU.Create(32),
-      TNNetFullConnectLinear.Create(10),
-      TNNetSoftMax.Create()
+      TNNetInput.Create(28, 28, 1),  // Input layer for 28x28 grayscale images
+      TNNetConvolutionLinear.Create(64, 5, 2, 1, 1),  // Convolutional layer with linear activation
+      TNNetMaxPool.Create(4),  // Max pooling layer
+      TNNetConvolutionReLU.Create(64, 3, 1, 1, 1),  // Convolutional layer with ReLU activation
+      TNNetConvolutionReLU.Create(64, 3, 1, 1, 1),  // Convolutional layer with ReLU activation
+      TNNetFullConnectReLU.Create(32),  // Fully connected layer with ReLU activation
+      TNNetFullConnectReLU.Create(32),  // Fully connected layer with ReLU activation
+      TNNetFullConnectLinear.Create(10),  // Fully connected layer with linear activation
+      TNNetSoftMax.Create()  // Softmax layer for classification
     ]);
+
+    // Creating the training, validation, and test image volumes from the fashion MNIST files
     CreateMNISTVolumes(ImgTrainingVolumes, ImgValidationVolumes, ImgTestVolumes,
       'train', 't10k', {Verbose=}true, {IsFashion=}true);
 
+    // Creating and configuring the NeuralFit object for training the neural network
     NeuralFit := TNeuralImageFit.Create;
     NeuralFit.FileNameBase := 'SimpleFashionMNIST';
     NeuralFit.InitialLearningRate := 0.001;
@@ -53,21 +67,25 @@   TTestCNNAlgo = class(TCustomApplication)
     NeuralFit.HasFlipX := true;
     NeuralFit.HasFlipY := false;
     NeuralFit.MaxCropSize := 4;
+
+    // Training and fitting the neural network using the provided data
     NeuralFit.Fit(NN, ImgTrainingVolumes, ImgValidationVolumes, ImgTestVolumes, {NumClasses=}10, {batchsize=}128, {epochs=}50);
-    NeuralFit.Free;
 
-    NN.Free;
-    ImgTestVolumes.Free;
-    ImgValidationVolumes.Free;
-    ImgTrainingVolumes.Free;
+    NeuralFit.Free;  // Freeing the NeuralFit object
+
+    NN.Free;  // Freeing the neural network object
+    ImgTestVolumes.Free;  // Freeing the test data volumes
+    ImgValidationVolumes.Free;  // Freeing the validation data volumes
+    ImgTrainingVolumes.Free;  // Freeing the training data volumes
     Terminate;
   end;
 
 var
   Application: TTestCNNAlgo;
 begin
-  Application := TTestCNNAlgo.Create(nil);
-  Application.Title:='Simple Fashion MNIST Classification Example';
-  Application.Run;
-  Application.Free;
+  Application := TTestCNNAlgo.Create(nil);  // Creating an instance of the TTestCNNAlgo class
+  Application.Title:='Simple Fashion MNIST Classification Example';  // Setting the application title
+  Application.Run;  // Running the application
+  Application.Free;  // Freeing the application instance
 end.
+

examples/SimpleMNist/SimpleMNist.lpr

@@ -7,7 +7,8 @@
 
 uses {$IFDEF UNIX} {$IFDEF UseCThreads}
   cthreads, {$ENDIF} {$ENDIF}
-  Classes, SysUtils, CustApp, neuralnetwork, neuralvolume, Math, neuraldatasets, neuralfit;
+  Classes, SysUtils, CustApp, neuralnetwork, neuralvolume, Math,
+  neuraldatasets, neuralfit;
 
 type
   TTestCNNAlgo = class(TCustomApplication)
@@ -17,17 +18,21 @@   TTestCNNAlgo = class(TCustomApplication)
 
   procedure TTestCNNAlgo.DoRun;
   var
-    NN: THistoricalNets;
-    NeuralFit: TNeuralImageFit;
-    ImgTrainingVolumes, ImgValidationVolumes, ImgTestVolumes: TNNetVolumeList;
+    NN: TNNet; // Neural network object
+    NeuralFit: TNeuralImageFit; // Object for neural network fitting
+    ImgTrainingVolumes, ImgValidationVolumes, ImgTestVolumes: TNNetVolumeList; // Volumes for training, validation, and testing
   begin
-    if Not(CheckMNISTFile('train')) or Not(CheckMNISTFile('t10k')) then
+    // Check if MNIST files exist
+    if not (CheckMNISTFile('train')) or not (CheckMNISTFile('t10k')) then
     begin
       Terminate;
-      exit;
+      Exit; // Exit the procedure if MNIST files are not found
     end;
+
     WriteLn('Creating Neural Network...');
-    NN := THistoricalNets.Create();
+    NN := TNNet.Create(); // Create an instance of the neural network
+
+    // Define the layers of the neural network
     NN.AddLayer([
       TNNetInput.Create(28, 28, 1),
       TNNetConvolutionLinear.Create(32, 5, 2, 1, 1),
@@ -41,21 +46,27 @@   TTestCNNAlgo = class(TCustomApplication)
       TNNetFullConnectLinear.Create(10),
       TNNetSoftMax.Create()
     ]);
+
+    // Create MNIST volumes for training, validation, and testing
     CreateMNISTVolumes(ImgTrainingVolumes, ImgValidationVolumes, ImgTestVolumes, 'train', 't10k');
 
+    // Configure the neural network fitting
     NeuralFit := TNeuralImageFit.Create;
     NeuralFit.FileNameBase := 'SimpleMNist';
     NeuralFit.InitialLearningRate := 0.001;
     NeuralFit.LearningRateDecay := 0.01;
     NeuralFit.StaircaseEpochs := 10;
     NeuralFit.Inertia := 0.9;
     NeuralFit.L2Decay := 0.00001;
-    NeuralFit.HasFlipX := false;
-    NeuralFit.HasFlipY := false;
+    NeuralFit.HasFlipX := False;
+    NeuralFit.HasFlipY := False;
     NeuralFit.MaxCropSize := 4;
+
+    // Fit the neural network using the training, validation, and testing volumes
     NeuralFit.Fit(NN, ImgTrainingVolumes, ImgValidationVolumes, ImgTestVolumes, {NumClasses=}10, {batchsize=}128, {epochs=}20);
     NeuralFit.Free;
 
+    // Clean up resources
     NN.Free;
     ImgTestVolumes.Free;
     ImgValidationVolumes.Free;
@@ -66,8 +77,8 @@   TTestCNNAlgo = class(TCustomApplication)
 var
   Application: TTestCNNAlgo;
 begin
-  Application := TTestCNNAlgo.Create(nil);
-  Application.Title:='MNist Classification Example';
-  Application.Run;
-  Application.Free;
+  Application := TTestCNNAlgo.Create(nil); // Create an instance of TTestCNNAlgo
+  Application.Title := 'MNist Classification Example'; // Set the application title
+  Application.Run; // Run the application
+  Application.Free; // Free the application instance
 end.