Multiple types of neural networks hardcoded in Python from scratch without using any framework to understand their fundamental functioning.
Reference - https://towardsdatascience.com/neural-networks-from-scratch-easy-vs-hard-b26ddc2e89c7
-
The dataset used is the sklearn digits datasets which consists of 1797 number images of resolution 8x8.
-
The Neural Network model used can be seen below.
- Results
After 1000 epochs
Loss | Train Accuracy | Test Accuracy |
---|---|---|
0.00213 | 99.87 | 98.33 |
- CNNs