Allow using GPU-accelarated CurveCurator through the package #405
Description
Activity
I totally support this, also because CurveCurator is a bit slow with maintaining dependencies and like this, we could finally move on to python 3.14 haha :D Does your version also run on CPUs as a fallback, though? Not every device has GPUs.
Yes, it's basically an alternative backend for the fitting, but the old one is still there
Reacted by Judith Bernett- linked a pull request that will close this issueIntegrate forked CurveCurator GPU API into drevalpy.curation #422
on Jun 8, 2026
As I mentioned in our last meeting, I created a GPU-accelerated implementation of CurveCurator. As you probably know, CurveCurator first groups experiments that have the same min and max concentrations and groups them in batches. GPU acceleration only really provides a speed benefit for larger batches. So some routing logic will be needed, especially if we want to enable nextflow to assign GPU-accelerated curveCurator tasks to GPU nodes and the smaller ones to CPU nodes.
I also added a python API for curveCurator, so that we do not have to call it via CLI and observe the subprocess.