Skip to content

Determinism #308

Description

@adamjstewart

To what extent can R-tree be guaranteed to be deterministic?

We're using R-tree in TorchGeo for ML, where reproducibility of experiments and data splits is very important. Currently, we populate an R-tree index with a list of files, then use index.intersection(index.bounds) to iterate over all files. We've had a number of issues in the past with sets/dicts and want to make sure we're using R-tree correctly as well. By determinism, I mean that given the same random seed, all code reproduces the same results exactly.

As far as I can tell, files are returned in insertion order. Is this behavior guaranteed? Are there any other parts of R-tree that may not be deterministic?

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type
    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions