An implementation of feed forward neural networks using only numpy. Batchnorm, regularisation methods, dropout, various activation and cost functions have been implemented. Optimisation algorithm used is SGD.
For better utility, a feature for saving models and using it again later is also implemeneted.
Written in python 3.6
- Add other optimization techniques like Adam, Adagrad, etc.
- Create visualisation functions (loss or accuracy vs epochs, etc.)