This digit recognition program is written by C++, and it is a simple multilayer neural network classification system that uses mini-batch gradient descent to execute learning process during training stage. Please follow the instructions to compile the program and execute the program:

1. Compile The Program:
   before running the digit classification program, we have to compile the program correctly. Please use the following command to compile this program:
   

g++ -std=c++11 main.cpp NeuralNetwork.cpp -o file_name

For example:

g++ -std=c++11 main.cpp NeuralNetwork.cpp -o output

2. Execute The Program:
   after compiling the program, we are able to apply two different commands to execute different functions. In the folder, there are two pre-trained weights, so users can directly utilize these two weight’s models to test classification accuracy. Alternatively, users can run training function in the program to generate new weights and apply new weights to do testing stage.
   

2.1: The command for training neural network:

./output run [N_neurons LR max_iter size_mini-batch]

For example:
./output run 128 0.3 30 10

2.2: The command for testing neural network:

./output testing [N_neurons]

Ex:
./output testing 128

where N_neurons: the number of neurons in the hidden layer.
LR: learning rate.
max_iter: maximum iteration
size_mini-batch: the size of mini-batch