Cifar10_Bayesian_Classifier.

In this we use Bayesian Statistical principles to classify images present in 10 different clases such as airplane, automobile, bird, cat, deer, dog, frog, horse, ship and truck. For more information on Cifar 10 dataset visit: https://www.cs.toronto.edu/~kriz/cifar.html

Data Preprocessing:

There are total of 50,000 training images of dimensions 32X32X3 divided into 10 different classes. Testing set consisits of seperate 10,000 images. The only important feature of Cifar-10 images is their average color. That means that we calculate the mean color of 32X32 images and each of the i=1,...,50000 CIFAR-10 images is then represented by only three values xi=(ri,gi,bi).

Implementation:

To use the above rule, we computed the mean and variance of the three color channels for each class. The function def cifar10_naivebayes_learn(Xf,Y) computes the normal distribution parameters (mu,sigma,p) for all ten classes(mu and sigma are 10X3 and prior p is 10X1). Finally, the function def cifar10_classifier_naivebayes(x,mu,sigma,p) returns the Bayesian optimal class c for the sample x.

Code file:

cifar10_bayesian.py

Dataset:

Dataset_Link.txt