MNIST_classifier

A fully connected neural network written using only matrix and numerical operations.

Install required packages

pip install -r reuqirements.txt

Experiment with different neural network architectures

Run ID	Architecture(number of neurons at each layer)	Performance on test set (out of 10000 images)	Performance on train set(out of 60000 images)
1	784, 300, 10	9024	53341
2	784, 2500, 2000, 1500, 1000, 500, 10	9415	55931
3	784, 5, 5, 7, 10	2583	15310
4	784, 1, 10	2365	13803
5	784, 2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2, 10	1135	6555
6	784,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5, 10	1135	6555
7	784, 10	8743	51599
8	784, 1, 300, 10	1135	6555
9	Decision Tree	9061	53534