Adding other optimization algorithm to the Dafoam

[qwefhajk]

I want to run the topology optimization case but I do not have the acess to SNOPT. Therefore, I wander how to add other optimization algorithm like the MMA/GCMMA in the pyopt to the Dafoam? I am just getting started on Dafoam by using the Docker image form. Should I compile it form source instead or is there a way to add pyOpt into the Dafoam docker image.

[friedenhe]

Unfortunately, the MMA/CGMMA optimizers were not added in pyOptSparse. You may need to create an issue on the pyOptSparse repo and request this feature. Once added, you need to compile everything from scratch, or you can compile the new pyOptSparse in the DAFoam Docker image and save it as a new image. Note that our framework does not support pyOpt. We support only pyOptSparse.

[qwefhajk]

Thanks you for the tips. I will look into it with further tests. Really appreciate your help!

[hosseinnegahban]

Unfortunately, the MMA/CGMMA optimizers were not added in pyOptSparse. You may need to create an issue on the pyOptSparse repo and request this feature. Once added, you need to compile everything from scratch, or you can compile the new pyOptSparse in the DAFoam Docker image and save it as a new image. Note that our framework does not support pyOpt. We support only pyOptSparse.

Could you please be more specific on how to compile a new pyOptSparse in the DAFoam Docker image?

[friedenhe]

https://dafoam.github.io/mydoc_installation_docker.html

[Ananth-Narayan-IITM]

@friedenhe, I was infact wanted MMA just to ensure how well it behaves than SNOPT and managed to make this work with pyOptSparse. I did install new version of pyOptSparse and PyOpt (which has MMA). Please refer to instructions which worked for me.

Edit1:
I will try to release few examples and a solver for cases using MMA

• Installing ParOpt (for MMA)- Topology Optimization in Docker

  • ParOpt

    • Clone the repository and extract

      cd ~/dafoam/repos
      wget https://github.com/smdogroup/paropt/archive/refs/tags/v2.1.5.tar.gz
      tar -xvf v2.1.5.tar.gz
      cd paropt-2.1.5
      cp Makefile.in.info Makefile.in
      export METIS_DIR="/home/dafoamuser/dafoam/repos/tacs/extern/metis"

    • Make sure that Makefile.in file is edited such that PAROPT_DIR=/home/dafoamuser/dafoam/repos/paropt-2.1.5 (or other location pointing paropt)
    • Make sure packages and modules are installed as given in INSTALL.txt,

      sudo apt-get install libblas-dev -y
      sudo apt-get install liblapack-dev -y
      sudo apt-get install libopenmpi-dev
      sudo apt-get install openmpi-bin
      pip3 install --upgrade pip
      pip3 install numpy
      pip3 install Cython
      pip3 install mpi4py
      pip3 install matplotlib

      and export PYTHONPATH and PAROPT_DIR as given in INSTALL.txt
    • Now compile them,

      make && pip install -e .[all]

    • Verify the ParOpt installation in Python as

      import paropt 
      print(paropt.__file__)

      Expected output is /home/dafoamuser/dafoam/repos/paropt-2.1.5/paropt/__init__.py

  • pyOptSparse

    • Uninstall current pyOptSparse v2.10.1,

      pip uninstall pyoptsparse

    • Verify if uninstall is success by python -c "import pyoptsparse"
    • Clone the pyOptSparse v2.15.0, which has Python Compatibility of >=3.9 thus suitable for the DAFoam docker (having Python 3.10.8).

      cd ~/dafoam/repos
      git clone https://github.com/mdolab/pyoptsparse.git
      cd pyoptsparse
      git checkout v2.15.0
      pip install .

  • Compatibility check between pyOptSparse and ParOpt versions

    • It was noticed that from paropt/paropt_pyoptsparse.py had issue with ImportError: cannot import name 'Error'. However if you are using latest version of ParOpt make sure to check the next step and act accordingly
    • From paropt/paropt_pyoptsparse.py replace line 13 from from pyoptsparse.pyOpt_error import Error to

      try: 
      	from pyoptsparse.pyOpt_error import Error 
      except ImportError: 
          pass

    • Make sure to replace if this line exists. I noticed that in recent version this has been removed
    • To verify the integration, launch Python

      from pyoptsparse import OPT 
      opt = OPT("ParOpt") 
      print(opt) 

      which outputs similar to <paropt.paropt_pyoptsparse.ParOptSparse object at ...>
  • Upon verification, make to compile the DAFoam code using ./Allmake under respective directory
  • Further verify this by running the cases by modifying runScript.py as

    parser.add_argument( "-optimizer", type=str, default="ParOpt")
    # Under args.optimizer
    elif args.optimizer == "ParOpt":
        prob.driver.opt_settings = {
            "algorithm": "mma",
        }

[friedenhe]

@Ananth-Narayan-IITM Thanks very much for sharing the installation guidelines. We will incorporate this into our new DAFoam version and update the website.