8000 Inference Args and Prediction Confidence? · Issue #286 · gcorso/DiffDock · GitHub
[go: up one dir, main page]
More Web Proxy on the site http://driver.im/
Skip to content
Inference Args and Prediction Confidence? #286
Open
@rjrich

Description

@rjrich
  1. Which inference arguments can have a positive influence on prediction accuracy or confidence? For example, what is the effect of increasing the actual-steps, and/or inference_steps, and/or samples_per_complex?

  2. If changing inference arguments from those in the default_inference_args.yaml file, how is this done on the command line, or could one make a copy of the yaml file and change certain arguments to custom values?

  3. Does increasing the samples_per_complex argument change the nature of the output? For example, is the output always in the form of ligand sdf files or can it be cif of pdb files of the complexes?

  4. When using my local installation of DiffDock in a conda environment, does the code automatically run on one or more GPUs or must the GPU(s) be specified? In my case, I have 2 x Nvidia A6000 GPUs; can I run the code on both of them or does it run on only a single GPU? If only one, can I specify which one? If so, how?

Thank you!

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions

      0