Bees Algorithm for Green Vehicle Routing Problem

Introduction

This repository contains code for Bees Algorithm on Green Vehicle Routing Problem. We have implemented and tested the following variants of Bees Algorithm:

• Basic Bees Algorithm
• Standard Bees Algorithm
• Reduced(2 parameter) Bees Algorithm
• Improved Bees Algorithm(with Simulated Annealing)

We have tested and compared the algorithms on the following datasets:

• Goeke's Dataset(2017)
• Erdogan

Apart from that, I have added a few data collection and quality of life utilities:

• Collect N runs of the algorithm on the same scene (src/Data Collection/CollectRuns.py)
• Perform basic statistical analysis on the runs, like mean, deviation, min, max (src/Data Collection/statistics.py)
• Read statisitcal properties for all settings and construct a table out of it (src/Data Collection/TableStats.py)
• Grid Search to find optimal hyperparameter setting (src/Data Collection/GridSearch.py)
• Plot a graph to visualize the results (todo)

Note that all of the above programs either use Multiprocessing by default (can be enabled by setting

MULTIPROCESSING = True

or an analogous variable) or have a Multiprocessing alternative(src/Data Collection/statistics_multiprocessing.py) in order to help collect the data fast. Please set number of workers according to the number of cores available. You might not want to use up all your cores.

Installation and Requirements

Requriements

Package	Version	Purpose
Numpy	1.22.x	Matrix and Array computations
Open CV	contrib, 4.5.x.x	Visualisation of output
Pandas	1.4.x	Dataset
tqdm	4.64.x	Progress bar for data collection programs

Installation Steps

• Clone the respository

  git clone https://github.com/Aryan-Satpathy/BA-GVRP.git

• Make a virtual environment

  cd BA-GVRP
  virtualenv <env_name>
  source <env_name/bin/activate>

  To deactivate the environment

  deactivate

• Install required libraries

  pip install -r requirements.txt

Running Code

cd src/
python main.py -- dataset 4 --useNFE --n 60000 --basic --gui

If you want to have a look at the available command line arguments

python main.py --help

Collecting Data

• CollectRuns.py: Change NumberOfRuns = 100 (line 22) to the number of runs you want to collect your data for. It will collect 100 runs for all the scenes in the given dataset. You can select algorithm by changing args = ['--basic', '--reduced'] (line 25).
• statistics.py: You can run

  python statistics.py --help

  to understand the various command line arguments available. Most of the command line arguments are simply passed forward as command line arguments while calling main.py.
• GridSearch.py: Change parameter_names = ['ns', 'nb'] # , 'nrb'] # , 'stlim', 'alpha'] (line 24) and

  parameter_space = [
  [5, 10, 20], # ns
  [7, 8, 9, 10, 11], # nb
  ]# [5, 10, 20],
  # ]

  (line 26-30) to perform Grid Search for the parameters. Note that you need to add an argument in statistics.py or statistics_multiprocessing.py to change the Bees Algorithm on which hyperparameter tuning is performed. The program will only give you a table containing the results for all the possible combinations of the parameter space. One can then find out the best setting using any csv editor.
• TableStats.py: After data has been collected, one can use TableStats.py to get a table containing performance on the different scenes, for the different Bees Algorithm variants.

Results

Results on Goeke Dataset(VRP 2017)