Creating container (Docker, etc) for elastica about pyelastica HOT 4 CLOSED

Comment by nmnaughton
Saturday Oct 03, 2020 at 00:18 GMT

@kshivvy @armantekinalp

Should we revisit this in light of Keshav's progress? It didn't work on Blue Waters, but his Docker file might be a good template to get started on this.

from pyelastica.

@mstoelzle I realized that you may have some experience in this context (a brief look into your group repositories revealed docker images of elastica-related packages). Do you wish to maybe take a stab at this issue?

from pyelastica.

@mstoelzle I realized that you may have some experience in this context (a brief look into your group repositories revealed docker images of elastica-related packages). Do you wish to maybe take a stab at this issue?

@bhosale2 Thanks for your message. Yes, I have some experience with Docker containers, but not with Singularity containers.

To me, the purpose and motivation for this feature is not clearly defined enough yet. In most cases, the package is easy enough to install. Have you confirmed that 'numba' cannot be installed without admin rights even with the pip --user flag?

• If the purpose is to run long simulations on a cluster without any additional (complicated) dependencies, would there maybe a more basic / simple solution around to install numba?
• If the purpose is to do RL / machine learning on a cluster, I would rather want to build on a PyTorch / TensorFlow / Cuda Docker images, as it is pretty straight forward to install PyElastica on these containers instead of the alternative of adding Cuda / PyTorch to an existing, custom PyElastica image, which is a pain.
• Are there any other applications invisioned for this container? For example an interplay with ROS? Then it would make more sense to build on a ROS image, which we are doing in our lab.

from pyelastica.

At this point we package using pypi. In future if we replace backhand code with c++ or cython then we can reopen this issue.

from pyelastica.