mlrose: Machine Learning, Randomized Optimization and SEarch

mlrose is a Python package for applying some of the most common randomized optimization and search algorithms to a range of different optimization problems, over both discrete- and continuous-valued parameter spaces.

Project Background

mlrose was initially developed to support students of Georgia Tech's OMSCS/OMSA offering of CS 7641: Machine Learning.

It includes implementations of all randomized optimization algorithms taught in this course, as well as functionality to apply these algorithms to integer-string optimization problems, such as N-Queens and the Knapsack problem; continuous-valued optimization problems, such as the neural network weight problem; and tour optimization problems, such as the Travelling Salesperson problem. It also has the flexibility to solve user-defined optimization problems.

At the time of development, there did not exist a single Python package that collected all of this functionality together in the one location.

Main Features

Randomized Optimization Algorithms

• Implementations of: hill climbing, randomized hill climbing, simulated annealing, genetic algorithm and (discrete) MIMIC;
• Solve both maximization and minimization problems;
• Define the algorithm's initial state or start from a random state;
• Define your own simulated annealing decay schedule or use one of three pre-defined, customizable decay schedules: geometric decay, arithmetic decay or exponential decay.

Problem Types

• Solve discrete-value (bit-string and integer-string), continuous-value and tour optimization (travelling salesperson) problems;
• Define your own fitness function for optimization or use a pre-defined function.
• Pre-defined fitness functions exist for solving the: One Max, Flip Flop, Four Peaks, Six Peaks, Continuous Peaks, Knapsack, Travelling Salesperson, N-Queens and Max-K Color optimization problems.

Machine Learning Weight Optimization

• Optimize the weights of neural networks, linear regression models and logistic regression models using randomized hill climbing, simulated annealing, the genetic algorithm or gradient descent;
• Supports classification and regression neural networks.

Installation

mlrose was written in Python 3 and requires NumPy, SciPy and Scikit-Learn (sklearn).

The latest version can be installed using pip: pip install mlrose-hiive

Once it is installed, simply import it like so: python import mlrose_hiive

Documentation

The official mlrose documentation can be found here.

A Jupyter notebook containing the examples used in the documentation is also available here.

Licensing, Authors, Acknowledgements

mlrose was written by Genevieve Hayes and is distributed under the 3-Clause BSD license.

You can cite mlrose in research publications and reports as follows: * Rollings, A. (2020). mlrose: Machine Learning, Randomized Optimization and SEarch package for Python, hiive extended remix. https://github.com/hiive/mlrose. Accessed: day month year.

Please also keep the original author's citation: * Hayes, G. (2019). mlrose: Machine Learning, Randomized Optimization and SEarch package for Python. https://github.com/gkhayes/mlrose. Accessed: day month year.

You can cite this fork in a similar way, but please be sure to reference the original work. Thanks to David S. Park for the MIMIC enhancements (from https://github.com/parkds/mlrose).

BibTeX entry: ``` @misc{Hayes19, author = {Hayes, G}, title = {{mlrose: Machine Learning, Randomized Optimization and SEarch package for Python}}, year = 2019, howpublished = {\url{https://github.com/gkhayes/mlrose}}, note = {Accessed: day month year} }

@misc{Rollings20, author = {Rollings, A.}, title = {{mlrose: Machine Learning, Randomized Optimization and SEarch package for Python, hiive extended remix}}, year = 2020, howpublished = {\url{https://github.com/hiive/mlrose}}, note = {Accessed: day month year} } ```