Kaggle Digit Recognizer using Softmax regression

By | December 2, 2017

Kaggle Digit Recognizer using Softmax regression

As a machine learning enthusiast, one would definitely want to solve some practical problems. Kaggle is one of the best platform to do so. It provides list of data science problems, including paid competitions. Kaggle digit recognizer is one of those problems.

MNIST dataset

One of the tutorial problem is called digit recognizer. We are given the MNIST dataset, which contains images of handwritten digits. Our task is to build a model, train the model with MNINST data. After training, we have to use the test data to predict scores. The final task is to submit predicted scores/labels to Kaggle. The Kaggle is having the ground truth labels for the test dataset. Comparing the ground truth labels with our submitted prediction labels, Kaggle gives us accuracy score. For my submission using simple softmax regression with tensorflow, I received an accuracy score of 91.40%.

You can find the code on GitHub here! or alternatively check the following code: