aiaccel: an HPO library for ABCI

日本語 (Japanese)

A hyperparameter optimization library for AI Bridging Cloud Infrastructure (ABCI). This software solves hyperparameter optimizations related to AI technologies including deep learning and multi-agent simulation. The software currently supports five optimization algorithms: random search, grid search, sobol sequence, nelder-mead method, and TPE.

Installation

The software can be installed using pip.

> pip install git+https://github.com/aistairc/aiaccel.git

Getting started

Running on a local computer

An example for optimizing a simple function (i.e., sphere function) on a local computer.

0. (Optional) Install Virtualenv and create a virtual environment.

   > python3 -m venv optenv
   > source optenv/bin/activate

1. Install aiaccel

   > pip install git+https://github.com/aistairc/aiaccel.git 

2. Create a workspace and copy the sphere example on the repository.

   > mkdir your_workspace_directory
   > cd your_workspace_directory
   > git clone https://github.com/aistairc/aiaccel.git 
   > cp -R ./aiaccel/examples .
   > cd examples
   > ls
   sphere
   
   > cd sphere
   > ls
   config.yaml         user.py

3. Run the parameter optimization

   > aiaccel-start --config config.yaml

   or

   > python -m aiaccel.cli.start --config config.yaml

   Tips: You can start after cleaning the workspace directory using --clean.

   > aiaccel-start --config config.yaml --clean

4. Wait for the program to finish and check the optimization results.

   > ls ./work
   abci_output         alive               hp                  lock
   log                 result              runner              state
   
   > cat ./work/result/final_result.result

5. If you want to change configurations, edit config.yaml.

   > vi config.yaml

Running on ABCI

This tutorial describes how to run examples/sphere on ABCI.

1. First, setup python environment following the ABCI Users Guide:

   > module load python/3.11/3.11.2
   > python3 -m venv optenv
   > source optenv/bin/activate

2. Prepare the workspace by following Steps 1 and 2 in Running on a local computer.

3. Please confirm the configuration file before running master.

   resource:
       type: "abci"
       num_workers: 4

4. Run on an (interactive) job

   > aiaccel-start --config config.yaml

5. If you want to check the running jobs, please refer the ABCI User Guide.

Others

• Check the progress

  > aiaccel-view --config config.yaml

• Display simple graphs

  > aiaccel-plot --config config.yaml

• Output results to workspace/results.csv.

  > aiaccel-report --config config.yaml

Acknowledgement

• Part of this software was developed in a project commissioned by the New Energy and Industrial Technology Development Organization (NEDO).
• aiaccel is built with the help of Optuna.