This is a Python implementation of the Perceptron algorithm. I wanted to write my own artificial neural network to gain an understanding of how they work and started with the simplest one. I make use of no external libraries and the code, although easy to understand, is not very efficient. It uses .txt files!

A Perceptron is used to define a hyperplane which divides the input space into two half-spaces and can be used to make binary classification.

• perceptron.py : Python script used to run the algorithm.
• constants.txt : Used to read constant values for the algorithm- number of input nodes, bias factor and learning rate.
• trainingData.txt : Each line contains an input vector + subsequent output, used to train the perceptron.
• testingData.txt : Each line contains an input vector for which the classification has to be made.
• weights.txt : Used to read weights for each node of the input vector. Gets created/overwritten on training.
• output.txt : Contains input vector + predicted output combinations for the test data.

A boolean function is a two-class classification problem- the inputs are binary and the output is 1 if the corresponding function value is true and 0 otherwise.

Functions like AND and OR are linearly separable and are solvable using the perceptron. A perceptron implementing the AND function can be found in the /res folder. The OR can also be implemented by modifying the trainingData.txt file.

Functions like XOR cannot be solved with the simple perceptron because their graphs are not linearly separable. Hence, there is no single line/hyperplane that can divide the input space into two classes. Note that a multilayer perceptron (MLP) can be used to solve problems like the XOR.

Modify the values in constants.txt, trainingData.txt and testingData.txt to your own data.

Then, train the perceptron algorithm using the following command (This will create/overwrite weights.txt):

python perceptron.py

Run the algorithm on the test data using the following command (This will create output.txt):

python perceptron.py test