Multi Layer Perceptron neural network with no ML libraries. Uses Genetic Algorithms for optimisation. Written in Python.
- Model has 1 hidden layer with arbitrary neurons.
- Model output layer has 1 neuron.
- Trains with back propagation using gradient descent of the error function.
- Model optimisation is done via a Genetic Evolutionary Algorithm
pip install numpy pandas copy matplotlib warnings deap scipy bitstring
The best results for the river basin's pan evaporation prediction (Fresno, CA):
| Hyperparameter | Value |
|---|---|
| Hidden neurons | 8 |
| Learning rate | 0.1 |
| Data split | 60/20/20 |
| Momentum | True (alpha = 1.1) |
| Bold Driver | False |
| Random seed | 10 |
Validation RMSE: 0.010292
Test RMSE: 0.009849
Denormalised RMSE: 0.158563
Comparison with multiple linear regression model:
