Energy, Emissions and Performance: Cross-Language and Cross-Algorithm Analysis in Machine Learning

This repository is a companion page for the following research, submitted for revision at the 9th International Workshop on Green and Sustainable Software (GREENS’25)

Authors Blinded for Review. 2025. Energy, Emissions and Performance: Cross-Language and Cross-Algorithm Analysis in Machine Learning.Submitted for revision at the 9th International Workshop on Green and Sustainable Software (GREENS’25)

---

Quick Start

Prerequisites:

1. Conda (for managing environments)
2. MATLAB (with Statistics and Machine Learning Toolbox)
3. Java SDK
4. R
   1. pandoc (brew install pandoc, to handle conversion from .Rmd to html)
5. g++ for C++ or(gcc or whatever you need to run cpp on your system)

Steps:

0. Make sure you installed the Prerequisites
1. Clone the repository:

   git clone https://github.com/Pampaj7/SWAM.git
   cd SWAM

Usage

0. You must change the python paths on all the languages:
   1. run which python on the terminal and copy that path
      • rRunner.R put your path in use_python()
      • runAlgorithm.m put your path in pyenv()
      • PythonHandler.java put your path in the first parameter of new ProcessBuilder()
      • pythonLinker.cpp put your path in add_to_sys_path_py()
1. Ensure you have created conda environments named cpp and sw; see allRunner.sh for more info
2. Navigate to the src directory of the project using your terminal:
   1. cd ../SWAM/src
3. Run the execution script:
   1. ./allRunner.sh
   • note: this step requires a long time because it runs 30 measures for all the combinations. If you want to speed up the process you can change the number of measures, or if you want to skip this step you can copy our measurements.
4. Ensure that all prerequisites are met. If so, the program will begin executing. If not cd into specific language folder and make sure you have the dependencies installed, some languages like cpp require system level libraries
5. Upon completion, a file named raw_merged_emissions.csv will be generated in the data folder, containing:
   1. 30 rows for each unique combination of dataset, algorithm, programming language, and phase (training or testing).

Result

After the program completes its execution, you will find all the generated plots in the graphics folder. Additionally, the data folder will contain all the datasets used to create those plots.

Repository Structure

This is the root directory of the repository. The directory is structured as follows:

.
├── README.md
├── data
│   └── mean_emissions.csv          Final experimental results data
├── plots                           Folder Containing plots
├── requirements.txt
└── src
    ├── cpp                         Folder containing cpp files
    ├── java                        Folder containing java files
    ├── matlab                      Folder containing matlab files
    ├── processedDataset            Folder containing processedDataset files
    ├── python                      Folder containing python files
    ├── R                           Folder containing R files
    └── Utils                       Folder containing a .py used to preprocess the datasets

Compilers Versions

Java version "20.0.1" 2023-04-18
Gcc version Apple clang version 16.0.0 (clang-1600.0.26.4)
Maven version Apache Maven 3.9.9
Python version 3.10.14
R version 4.3.2
Matlab Version '24.1.0.2653294 (R2024a)