This is a project with C++11 and Python3.
Also use MINIST dataset to train model and save model to a *.npz file.
The model was used to recognize the number in the images.
prj1 is used for testing that via using python call binary files of c++ compiled;
prj5 is a 4-layer MLP model, the model file was saved in p5_params.npz;
prj6 is a 6-layer CNN model, the model file was saved in p6_params.npz;
For MLP, the test accuracy is greater than 96.1%, the sum of log(loss) is less than 4773.266.
For CNN, the test accuracy is greater than 96.7%, the sum of log(loss) is less than 4749.440.