In this project, we compare machine learning algorithms to predict Farsi hand-written digits. The simplest way is to employ a K-Nearest Neighbors model. However, a Deep Neural Network with fully connected layers increases accuracy. Finally, fitting a Convolutional Neural Network reaches us to 99 percent accuracy after a few epochs.
It is just like MNIST dataset, but with Persian digits. the samples are all hand-written, grayscale, at the center and without rotation. That is why I called this repository "pretty-digits". I also used HodaDatasetReader to get .cdb files and read them.
This project is licensed under the MIT License - see the LICENSE file for details