CarbonCostCode

Repository for testing the carbon cost of codeing related to code used within the MRC-IEU. One Drive folder: https://uob.sharepoint.com/:f:/r/teams/grp-ieu-dmer_prog/Shared%20Documents/Social/2023-06-15-prog3-away-day/away-day-hackathon/materials/GreenAlgorthims?csf=1&web=1&e=gyEdZy This repository has been setup and developed as part of the IEU-Program3-AwayDay. Participants: Winfred Gatua, Maria Sobczyk-Barad, Emma Hazelwood, Olalekan Awoniran, and Hayley Wragg

Requirements

Packages Used

Green Algorithms The initial goal was to use the GreenAlgorithms package from https://www.green-algorithms.org/ to calculate the carbon cost from coding. However this has installation issues and does not allow for comparing domestic computers. Code Carbon The code carbon package is much more straightforward to use however this is called within a python program. We have therefore written a script for calling an R script within python. https://github.com/mlco2/codecarbon

Programs to compare:

• Record – Number of SNPS/ Statistical Power vs Energy Usage and Carbon cost.
• 2 different GWAS exposure – GWAS for BMI (Locke ver Yengo). Both two-sample univariable and multivariable analysis.
• Running MR with/without proxy look-up? Or using pre-clumped instruments from OpenGWAS versus clumping SNPs locally.
• Colocalisation with/without Susie finemapping
• Reading in data – packages/functions (data.table, tidyverse, etc), data formats, non-standard formats (comma separated, tab-separated, having to re-do, etc – carbon cost of a bad readme, also importance of reading in first few rows first versus entire file), transposing (R versus Python packages)
• Chunk size in pandas

Setup

To calculate the cost of a single program populate a csv named: method_id.csv. The name for your csv should be added to the INSERT_FILENAME file. If the method is already in the list match the method_id otherwise add a new row.

Inside the setup csv populate:

• method_id (should all be unique)
• n_snps (for MRs)
• n_rows (for reading data)
• n_chunk_size
• package (for reading data)
• carbon_calculator

Calculating your cost.

Using the offline version of the emission tracker from codecarbon track the power usage and run time.

import codecarbon as cc
with cc.OfflineEmissionsTracker() as tracker():
  # Your Code

Save the cost

import panda as pd
carbon_dict={method_id:your_id,
              n_snps:your_nsnps,
              n_rows:your_n_rows,
              n_chunk_size: your_chunk
              package:your_package
              carbon_calculator:your_calculator
              cost:power_usage,
              run_time:your_time}
  res_df=pd.Dataframe(carbon_dict)
  res_df.to_csv(results_file_name)