NeuraSearchLib

NAS (Neural Architecture Search) library

Table of Contents

• Description
• Installation
• Usage

Description

This tool is a NAS (Neural Architecture Search) library designed with a modular structure. It is organized into the following components:

• Search space module
• Search strategy module
• Training module
• Evaluation module

The library offers predefined search space templates that are highly customizable:

• Block/Layer-based NAS Search-space model
• Cell-based NAS Search-space model
• Hierarchical NAS Search-space model

The DNN architectures generated from these spaces are represented as flexible recursive Directed Acyclic Graphs (DAG) of neural network (NN) operations.

Installation

Below are the installation instructions for this tool:

1. Get the repository: To obtain this repository, you can either clone it or simply download it directly.
2. Navigate to the project folder
3. Install dependencies:

   pip install -r requirements.txt

If a package fails to install, try installing it manually with: pip install

Usage

Below are the usage instructions for this tool:

1. Edit the configFile.yaml file:

   • In the configFile.yaml, specify your configuration (in terms of search space, search strategy, training strategy, and evaluation strategy) using YAML syntax.
   • You can find examples and scenarios to help structure the file in the test folder.

2. Run the library:

   After configuring the configFile.yaml, run the library using the following command:

   python -m main.py