Machine Learning Stuff
Based on the linear regression task from the Coursera Machine Learning course. I wanted to see whether I could implement machine learning algorithms in C++.
Technologies used here:
- C++ 11 (could be a higher version but sticking to this until I need higher)./
- Eigen - C++ matrix library.
- dlib - an alternative C++ matrix library, with added machine learning algorithms.
- matplotlib - a Python graphing library.
- matplotlib_cpp - a C++ wrapper for matploglib.
- Python 3.
Exercise 1a - simple linear regression
Given X and Y points from input data generate a line of best fit using linear regression.
Original points with line of best fit generated using linear regression:
Cost against values of Theta:
Exercise 1b - linear regression with multiple variables
Graph showing how learning rates differ based on different alpha increments during the learning process:
Exercise 2a - simple logistic regression
Given data that includes two exam results plus whether or not the candidate was admitted into the college setting the exams, work out how to predict a candidates likelihood of success based on a pair of results. Plot a line showing the 0.5 boundary between admission and non-admission.
Exercise 2b - multi-variable logistic regression
Perform logistic regression using more many more variables, allowing the generation of much more complex boundaries between two conditions. In the graph below the boundaries marks the point at which a microchip has a 50% change of being successfully accepted. Different values of the lambda correction factor alter the extent to which the results fit the data. Lambda = 0 has a messy, complex boundary, lambda = 1 seems more reasonable, lambda = 100 seems to be rather offset. (the boundaries are made using scatter plots as the graphing library can't mathematically plot such complex shapes).
