Numpy-Neural-Network

The repository contains the code for creating a simple neural network from python and numpy

The initial idea is a NN using numpy :

• the project uses a spiral data generated by code
• there can be multiple classes so to calculate the loss , we have used categorical cross entropy

NOTE : We have further implement MNIST dataset to classify the 28x28 digits using NN build from numpy and maths. Have Fun !!

---

Below is the progess in the round 1 of training with a simple code and less optimization

---

For next round we added more layers and made the learning rate half each time after 4 epochs : to achieve accuracy of 95.9%

Here is the link to MNIST repo -> MNIST github