Skip to content

santidrj/DL-energetic-impact-analysis

Repository files navigation

Analysis on the Energetic Impact of Deep Learning Training

This repository contains the data and code used to generate the results described in the master thesis "An analysis of modeling and training decisions for greener computer vision systems".

Datasets

In the data folder you will find two folders. In the raw folder, you will find a file containing the energetic data of all the experiment runs and a file containing the model metrics (i.e., FLOPs, accuracy, precision, recall, and AUC). In the processed folder, you will find the data obtained after integrating the files in the raw folder, and extracting revelant features.

All the files inside the data folder have been saved using Panda's to_parquet function using the GZIP format.

DL Training Profiling Dataset

The DL Training Profiling Dataset contains all the energy-related measurements taken during the training of the models tested in the study.

You will find a detailed description of the dataset in its dataset card.

DL Training Energy Consumption Dataset

The DL Training Energy Consumption Dataset contains all the energy-related measurements taken during the training of the models tested in the study together with their prediction quality.

You will find a detailed description of the dataset in its dataset card.

Data analysis

The repository contains a Jupyter Notebook with the code used to clean and analyze the data.

License

The software in this work is licensed under the Apache 2.0 License.

The files under the data folder are licensed under the CC-By Attribution 4.0 International.

About

Replication package for the data analysis of the master thesis "An analysis of modeling and training decisions for greener computer vision systems"

Topics

Resources

License

Stars

Watchers

Forks