Energydl: A tool for energy-aware interactive training of neural network architectures

This repository contains the code implementation corresponding to the following research paper:

Álvaro Domingo, Silverio Martínez-Fernández, Roberto Verdechia. Energy-aware training of neural network architectures: Tradeoff between correctness and energy consumption.

The code demonstrates the algorithms and methods outlined in the paper, providing a practical reference for researchers and developers interested in replicating the results or exploring the proposed techniques further.

Table of Contents

• Installation
• Usage
• Requirements
• Contributing
• License

Installation

1. Clone this repository:

   git clone https://github.com/alvarodr21/energydl.git
   

2. Navigate to the project directory:

   cd energydl
   

3. Create a virtual environment (optional but recommended):

   python -m venv venv
   

4. Activate the virtual environment:

   On Windows:

   venv\Scripts\activate
   

   On macOS and Linux:

   source venv/bin/activate
   

5. Install the required dependencies:

   pip install -r requirements.txt
   

Usage

The module is contained in the file energydl.py. It can be imported to use in any Python script by from energydl import energy_aware_train. This function is aimed to replace TensorFlow's tf.keras.Model.fit on any script, passing the model as the first argument and keeping the rest. The behavior of it can also be customized by some unique parameters not present in Model.fit.

A demo of a simple usage of this function is included in demo_energydl.py. It can be simply called with python demo_energydl.py.