Handwritten digits recognition using one-against-all classification (oaa) in Vowpal Wabbit

https://ashokharnal.wordpress.com/2015/03/06/handwritten-digits-recognition-using-one-against-all-classification-oaa-in-vowpal-wabbit/

Kaggle recently hosted a machine learning competition to recognize handwritten digits from 0 to 9. Handwritten digits have been taken from MNIST database (Modified National Institute of Standards and Technology). We decided to use Vowpal Wabbit for learning the pattern of handwritten digits in the training file and apply the learning on the ‘test’ dataset to predict what all digits it represented. Our score on Kaggle was 0.97943 i.e. 97.94% accurate prediction.
Vowpal Wabbit is very easy to install. Its installation on CentOS may not take more than 20 minutes. See the instructions here.
Dataset is in two files: train.csv and test.csv. File, train.csv, contains 42,000 images each of a single handwritten digit from 0-9. Each image is 28 X 28 pixels that is in all 784 pixels–all lined up in one long row. First five lines of train.csv appear as follows: