rogerni / simple_cpp_fcnn_class

This a simple C++ fully connected neural network class for Course COMP3046.

Simple_CPP_FCNN_Class

This a simple C++ fully connected neural network class for Course COMP3046.

Codes

• ANN.h : header file for ANN class
• ANN.cpp : implementation of ANN class
• ANN_Run.cpp : sample driving program for ANN class

Main Functions Supported

• Defining any configuration of FCNN
• Parrellel computation on CPU
• Save and Load trained model
• Output Loss and Testing accuracy
• Output Epoch time spent

Explanation on ANN_Run.cpp

Set Configuration of the Neural Network

vector<int> cfg = { 28 * 28,100,10 };

The numbers in the vector indicates the numbers of neurons on each layer. In the sample file, there are (28 * 28 =) 784 neurons on input layer, 100 neurons on first (the only) hidden layers, and 10 neurons on output layers.

Construct an ANN instance:

ANN net(cfg);

Set Testing data

net.setTestData(test_X, test_Y);

test_X and test_Y are vector<vector<float>> that store the testing X data and testing Y data.

Train

net.train(X_train, y_train, 0.01, 100, 64);

Parameter	Explanation
X_train	vector< vector<float> > containing training X data
y_train	vector< vector<float> > containing training y data
0.01	learning rate
100	Epochs to train
64	Batch size

Write Model to File

net.writeTo("Final_Model");

Load Model from File

ANN net("Final_Model");

Test Accuracy

net.testAccuracy()

Returns accuracy

Acknowledgement

This is a course project for COMP3046 in HKBU. Thanks to course instructor and teaching assistant.

Author

NI Ronghao ( RogerNi - GitHub )