Using Python to build a backpropogating neural network from scratch for CS303. It uses the Numpy and Matplotlib libraries. Make sure you have installed that before using my code.
The activation function we use is the Sigmoid function, hence the training and the testing data needs to be normalised.
The parameters of the neural network can be edited all at once in the following section of the code -
#Parameters for neural network
hidden = 6 #No of Hidden neurons!
output = 1 #Output neurons
input = 1 #Input neurons
eta = 0.01 #Learning rate
errors = []
bias = 1 #Bias
epochs = 2500 #Number of epochs
The dataset that this neural network works on is normalised. If you're using it on another dataset, make sure it is normailsed.
The legacy folder has all the files that were basically my (failed) attempts at making this algorithm work.
.
├── legacy # Defunct files. Tinkered around on them.
├── arewethere.py # Main BBP algorithm implmentation
├── data # Testing and training data
-
Run the command
pip3 install -r install.txtto install all the dependencies (Python 3). -
The script, arewethere.py, is the final script. You will need to have some testing data as well as some training data for it to work.
-
Specify where you training and testing data is.
-
Change your neural network parameters and then just run it. Should work. It will also automically plot the error rate.