Roadmap

[ffelten]

This issue gives an overview of the next development efforts. We're planning to align with the academic calendar to do brainstorming and planning sessions and fill this. The issues are listed by priority (highest priority first).

Summer 2026

• Additional Usage Examples
  • Multi-Fidelity (e.g., Beams) -- @arthurdrake1
  • Dataset quality/estimation -- what is a smart way to downsample the dataset
• Additional Problems
  • Generating more difficult variants of EngiBench problems
  • MTO 2D Maybe
• Contributor Experience Improvements
• Refactoring in EngiOpt: try to unify inference code for different models?
• New Features/Capabilities
  • EngiOpt handling non-vector input (2D ThermoElastic and similar)

Spring 2026 (DL end of May)

General release cycle focuses primarily on usability and improvements. Reducing barriers to entry and friction for users and contributors. We may also add in some additional features as bandwidth allows.

• Dissemination
  • Prepare a Workshop [@SoheylM @mkeeler43] -- Prepare DCC Workshop Proposal #222
• Documentation Improvements
  • Constraints API is not super well-explained yet
    • We need a documentation page on how to use the Constraint API and what it is for and what it is not for, e.g., checking valid inputs for simulator vs. satisfying actual numerical constraints in the optimization. - Add documentation for the Constraint API #223
• Examples/tutorials Improvements
  • Extract Constraints/Violations and compute % violations -Add documentation for the Constraint API #223
  • Use of SLURM / Data Generation capabilities - @fgvangessel-umd
  • How to do simple model benchmark -- @mkeeler43
    • Show how to load up a pre-trained EngiOpt model and explore it using sliders or interactively
• New Features/capabilities (possible if time permitting, but maybe moved to Summer '26)
  • Transforms ability (@xarthurx )
    • Make Matrix of EngiBench/EngiOpt Input and Output types
    • Prioritize which conversions are highest value / can we jsut convert to/from a single one?
    • How do we parametrize any conversation parameters?
    • How do we tell that it was successful/validated
  • Integrate Aero3D problem (@cashend )
  • Beams3D problem (@arthurdrake1 )
  • Field Output (e.g., outputting pressure fields on wing or voltage curves in Power Electronics) Should we provide a mechanism to output performance fields/information? #71 -- @fgvangessel-umd
    • Frank could use help in identifying where in the problems (simulate and optimize) in each problem.
    • Geri will need to comment on any changes to Core API
  • Return the gradient information from solvers -- @gapaza @cashend

Fall 2025 (DL end of December)

This one focuses on opening to external contributions and prepares for broader adoption within the scientific community (workshops, hackathon, competition).

• Testing
  • Optimization/Simulation with Slurm for all problems tested (5 groups of 15 should be okay). Report all the encountered bugs in issues. [@g-braeunlich]
  • Fix remaining bugs in Airfoil [@fgvangessel-umd @g-braeunlich]
  • Fix Photonics Units in HF Dataset Column for better CoG calcs [@markfuge @mkeeler43] -- Double check HF Photonics units in the column #217
• Release the new version with Slurm API, new constraints, bug fixes. [@ffelten]
  • Update EngiOpt to latest version
  • Support Python 3.13 #200
• Contributor guide and PR template [@ffelten]
• Step-by-step guide to add a new algorithm in EngiOpt - This is an issue now in EngiOpt so tracking it there.
• Profit.

Summer (DL end of August)

This "release" focuses on improving documentation and easing setups, preparing the class this Fall semester.

• Onboarding documents: [@g-braeunlich, @SoheylM, @ffelten]
  • [internal] local dev env. setup (VSCode + copilot, python venv, ruff, pre-commit, mypy) [@ffelten, @SoheylM] https://github.com/IDEALLab/FirstSteps | https://github.com/IDEALLab/lab-handbook
  • [internal] what is a linter, gitflow explanation, github CI actions [@SoheylM, @ffelten] https://github.com/IDEALLab/FirstSteps | https://github.com/IDEALLab/lab-handbook
  • "hello world" -> launch an engiopt training and eval locally @ffelten https://github.com/IDEALLab/EngiOpt/blob/main/example_easy_model.ipynb
  • [internal] what is an HPC, how to ssh to Euler, how to ssh with vscode @SoheylM https://github.com/IDEALLab/lab-handbook
  • [internal] "hello HPC" -> use HPC euler scripts to launch engiopt jobs
• Colab tutorials:
  • full pipeline for creating a dataset, slurm Slurm Example #137
  • simulate/optimize (https://colab.research.google.com/github/ideallab/engibench/blob/main/tutorial.ipynb ?)
  • train a model @ffelten https://github.com/IDEALLab/EngiOpt/blob/main/example_easy_model.ipynb
  • evaluate a model @ffelten https://github.com/IDEALLab/EngiOpt/blob/main/example_easy_model.ipynb
• Cross architecture tests: GitHub CI Mac, Linux, Windows GitHub CI on different OSes #150
• Wanted algorithms Wanted algorithms EngiOpt#39