Perceptron Repository

KNN Repository

Lowdown

I developed in my own time a simple perceptron model that could classify an image ‘bright’ or ‘dark’ with a small number of pixels as input. I also developed a KNN classifier that predicted the species of plant an input was based off of attributes that were given.

Key Technologies

• Python

More Information

The perceptron model was very simple, however I implemented it completely from scratch using the mathematics behind the model. This included writing functions for sigmoid, returning the weighted sum (of weights of inputs), and the perceptron training function that updates the weights based on the output from the perceptron. This was all done in around 100 lines of code, and although simple something I am really quite happy with and something that works really well and shows of some of the simplicity and elegance of basic machine learning.

The KNN classifier was also developed from scratch and is an implementation of the K-Nearest-Neighbour algorithm. This works by classifing a new element based of a plurality vote of the elements nearest neighbours. In my case I used the euclidean distance to figure out who the ‘nearest’ neighbours were. This is also a short program with only roughly 100 lines of code and also works very well returning high accuracies on its predictions.