This is a small vanilla neural network I created for the Iris data set. The program allows a variable number of layers and nodes/neurons per layer. The architecture of the program was inspired from the "Neural Networks and Deep Learning" ebook by Michael Nielsen.
- numpy
- sklearn
- matplotlib
To run the program without any hidden layers or any adjustments on hyperparameters, simply execute(*) :
python neuralnet.py
Another way is with flags. I have implemented a few flags that allows one to adjust hyperparameters of the neural network.
| Flag | Details |
|---|---|
-hl |
Hidden layer configurations (Default: N/A) |
-lr |
Learning rate (Default: 0.05) |
-m |
Momentum (Default: 0.05) |
-rg |
L_2 regularization (lambda) (Default: 0.001) |
-ep |
Total number of epochs (Default: 1000) |
For example, to run the neural network with 2 sequential hidden layers with the first having 4 nodes and the next having 6 nodes, the command is:
python neuralnet.py -hl 4 6
(*) This assumes one has Python3 has the only interpreter. Otherwise one would run with python3.