Skip to content

HPO for supported models on original data

Pre-release
Pre-release
Compare
Choose a tag to compare
@wilke wilke released this 29 Sep 19:12
· 3 commits to release since this release
v0.2.0-beta
6fd39d7
This commit was created on GitHub.com and signed with GitHub’s verified signature. The key has expired.
GPG key ID: 4AEE18F83AFDEB23
Expired
Learn about vigilant mode.

IMPROVE: Early Adopter Release v0.2.0 (2023-09-29)

This release allows optimizing hyperparameters for machine learning using data provided or described by the original model code creator. It includes instructions for setting up the infrastructure and performing initial HPO.

To get started please use our guides in the documentation.

Included in this release are:

  • Hyper Parameter Optimization (HPO) workflow version v0.6
  • IMPROVE models, build and testing code version v0.6
  • IMPROVE documentation version v0.6.0
  • CANDLE library v0.8

Documentation:

Repository: https://github.com/JDACS4C-IMPROVE/docs
Documentation: https://jdacs4c-improve.github.io/docs/index.html

Release:

New documentation guides for:

  • Setting up the initial environment for building support model container images and executing HPO workflow
  • Instructions for making container images for supported models
  • Instructions for setting up HPO for a supported model

Build and test framework

Repo: https://github.com/JDACS4C-IMPROVE/Singularity

Release:

This release includes:

  • build instructions for supported model containers.
  • test scripts for models and containers

Workflows

This release of the supervisor supports the Hyper Parameter Optimization Workflow, which was used to optimize multiple IMPROVE models via the DEAP algorithm. We developed a new front-end command-line tool called supervisor that eases the management of configuration files for the system. We also improved the interfaces between Supervisor and IMPROVE containers, and clarified output from the system, including statistics from the models and the overall HPO progress.

Repo: Supervisor

Release:

HPO Examples

Repo: HPO

Release:

This repository contains minimal testing config for HPO and a collection of best HPO results for given HPO and GA parameter space.

CANDLE library

Vendorized version of candle_lib. This version is used by all models and provides a standardized interface for model execution.

Release:

New supported models:

This models have been modified to provide a standardized interface for model training.

Assets 2
Loading