From the scratch implemenation of a NN to predict whether the house price will be above or below the median market price.
Various experimentations have been performed to decide:
- No of hidden layers
- No of nodes in hidden layers
- Various Activation Functions like sigmoid,re-Lu,softmax etc
- Weight Initialisations and learning rate.
- Dropout was also implemented from scratch.
For Theory refer Pattern Recognition and Machine Learning by Bishop.