a simple classifier of digts_and_letters based on cnn framework(beta)
data_set_link:https://www.westernsydney.edu.au/bens/home/reproducible_research/emnist(mnist_version)
data_set_intro:EMNIST.pdf in project directory.
what improvement could be done in the future?
maybe tune the hyper_params,change the network,add new regularizer and loss fn,etc.
the average testing acc and loss respectively of classes(62):0.82 0.57