Classification using Naive Bayes

Classification using Naive Bayes on the MNIST dataset. For project explanation see [1]

Run

You can run the code using python run_naive_bayes.py --trainpath PATH_TO_TRAIN --testpath PATH_TO_TEST --algorithm ALGORITHM_CODE. There are four options for algorithm:

1. B - Image is binarized and features are modeled as Bernoulli random variables.
2. G_gray - Original grayscale values of the image are used as features and modeled as jointly Gaussian random variables.
3. G_pca - PCA is used for dimensionality reduction of images before classification.
4. G_hog - Histogram of oriented gradients are computed from the image and used as features for classification.

Dataset

You'll require access to the MNIST dataset which can be obtained from [2]. For ease of use, the code has been written to load the data in .csv format which can be obtained from [3]. After downloading store the datasets in ./data/.

References

[1] - Deepankar C. - A Primer to Bayes' Classifier
[2] - Y. LeCun, C. Cortes, C. Burges - The MNIST Database
[3] - MNIST in CSV