This repo contains modules for educational visualizations of ML techniques. Linear regression with gradient descent, softmax & logistic regression for double and multiclass classification, and full batch gradient descent, SGD, z-scoring, and other techniques for optimization are hand implemented and displayed through use of NumPy and Matplotlib. Modules contain classes and functions with varying levels of reusability, documented within module docstrings and beneath.
Contains SoftmaxRegression class for training and classification of general n-dimensional data. Performs z-scoring normalization and full batch gradient descent across all passed training examples. Online, single example classification is available following training. Running module as main generates, plots, trains, and classifies 3d data. See module for documentation and example of class usage. Data may be plotted with various graphical options and print outs, as shown beneath.
Contains functions for training and classification through logistic regression using full SGD. Running module as main generates 3d visualizations of logistic classifier, including decision boundary graphing. See main for further options and examples of function use.
Contains two generators for creation of uniformly distributed 3d and 2d data, generated from underlying planar or linear functions. Noise, placement, and distribution of data can be modulated via parameters. Optional printing to stdout with generator instantiation details specifications of the generated data. Further documentation in module.
Running module displays simple visualization of gradient descent for linear regression, dynamically fitting a line to noisy 2d data generated from an underlying linear distribution with an animation.
