jirryzhang / digitrecognizer Goto Github PK

Read the full article at http://primaryobjects.com/CMS/Article159.aspx

This is a machine learning project for recognizing hand-written digits of 0-9 as their numeric values. The data comes from the [MNIST data-set] (http://yann.lecun.com/exdb/mnist), as part of a Kaggle competition.

The code is developed in C# .NET, using the CsvHelper library for parsing the csv file of images, and Accord .NET for the gaussian SVM. The images are provided as 28x28 gray-scale pixel values (0-255), resulting in 784 pixels per image (ie., 784 columns in each csv row). An additional column is included for the label identifier.

For example (for the digit '7'): 7,0,0,0,0,255,127,0,0...

Kory Becker http://www.primaryobjects.com/kory-becker.aspx