Classification

Single-feature classification algorithm using gradient descent.

Using python libraries: Matplotlib, csv, random, and math.

Initalised to use testing data. Datapoints are plotted in blue, with the orange line being the predictions of the algorithm for the given x-value. The cost (mean squared error between predictions and given data) through interations is also graphed to track the improvement of the algorithm (the lower the better).

Adjustable Parameters

• order: user may set the order of the polynomial used to fit the datapoints
• batch: the number of datapoints fit by the algorithm per iteration (setting it to size works well for small datasets. Consider smaller values for large datasets)
• lrate: learning rate. Larger values mean each iteration affects the hypothesis function to a greater extent. Values too small can result in slow convergence, values too large can overshoot the optimal hypothesis.
• iters: the number of interations
• detail: used for plotting. Determines the number of sample points of which the orange line is constructed. More points result in a smoother curve.

Acknowlegements

DATA.csv is a section of this breast cancer dataset.

Thanks to Andrew Ng's course on Coursera, where I learnt these techniques