Open Edreva opened 4 years ago
Attempted to train a digit recogniser using SVM and the same MNIST data set however on first test this seemed to have an even lower success rate. Will try to optimise somewhat and see if results improve but not hopeful
Even after manually removing noise and centering one of the number squares containing the number 7 it was still misidentified by the knn digit recogniser. Suspect this is a consequence of using MNIST handwritten digits as the training data rather than printed digits. Must try to find a reasonably large sample of printed digits to test.