CS 7641 - Assignment 2: Randomized Optimization

This repository contains the experiments made to analyze and complete assignment 2 of the CS7641 - Machine Learning course from the Georgia Institute of Technology.

The main purpose of this assignment is to analyze 4 randomized optimizations algorithms:

• Randomized Hill Climbing
• Simulated Annealing
• Genetic Algorithm
• Mutual-Information-Maximum Input Clustering (MIMIC)

with 3 optimization problems:

• One Max
• Flip Flop
• Knapsack

as well as finding weights for a neural network using RO algorithms to replace back propagation from gradient descent.

We use the mlrose-hiive module to execute these experiments.

---

Setup Virtual Environment

In the root directory of the project, do one of the following two

Conda

$ conda create --name myenv python=3.8
$ conda activate myenv
$ pip install -r requirements.txt

PipEnv

$ python -m venv /venv
$ source venv/bin/activate
$ pip install -r requirements.txt

mlrose-hiive

If mlrose-hiive does not install automatically with the requirements.txt, then run

$ cd mlrose
$ pip install -e .

Run Scripts

Optimization Problems using Randomized Optimization Algorithms

Open up randomized_optimiztion.ipynb in VS Code (or your preferred editor/jupyter server), select your virtual environment (in this example myenv) as your kernel and run the code cells one by one.

Note that the problems may take a long time to finish computing sometimes (from 30 mins to 2 hours) depending on hardware specs. This is due to the number of experiments we do on each problem (~180 per problem).

Finding Optimal Weights to train a Neural Network using Randomized Optimization Algorithms

Open up neural_network_weight_optimization.ipynb in VS Code (or your preferred editor/jupyter server), select your virtual environment (in this example myenv) as your kernel and run the code cells one by one.

Note that the problems may take a long time to finish computing sometimes (~20 mins max) depending on hardware specs.

---

Note: If you are a GT Student taking CS7641 DO NOT proceed looking at this repository in order to not violate the GT Academic Honor Code