logistic-regression-from-scratch

Designing a simple logistic regression algorithm from scratch, without the use of external ML libraries.
(current version: 0.1)

Author: Siddharth Yadav (syntax-surgeon)

main.ipynb

• Contains version-0.1 of the main program which tries to implement a logistic regression algorithm
• The core of the algorithm is based on an intuitive implementation of basic gradient descent combined with a non-linear function (sigmoid)
• The current version of the codebase is functional, however, future versions may be OOP-based
• Simple plotting functionality is included to visualize the results of the training

Figure showing the dataset with the final classifier (left) and the mean log loss error per epoch (right)

Features planned for release 1.0

• Converting the codebase to an OOP architecture
• Inclusion of additional classifier training features such as 'strict_reduce' and 'descent_type'
• Advancing the graphing/plotting functionality to include intermediate perceptrons
• Increasing the verbosity of the in-code documentation