Pierre Marcenac

Training and validation LMDB folders ('/data/train_lmdb' and '/data/val_lmdb') should not be forced to the user in the preliminary checks, for they're used neither in spearmint nor in run.py. In addition,...

Set of evaluation measures for multi-class multi-label classification

2

This includes: - predictions from inferred scores, - example-based measures, - label-based measures, - all measures (both).

RAI fields in mlcroissant

2

It seems there's a discrepancy between the [RAI specs](https://github.com/mlcommons/croissant/blob/main/docs/croissant-rai-spec.md#rai-property-information) and [their implementation in mlcroissant](https://github.com/mlcommons/croissant/blob/a1a413cf1fb376ecc3bc4d3d6d7858f4f2b3bc61/python/mlcroissant/mlcroissant/_src/structure_graph/nodes/metadata.py#L160-L274). @JoanGi Could you please have a look at the differences (as you initially implemented the RAI...

There is no such thing as `nx.MultiDiGraph[Node]`. As a consequence, pytype always expects Hashables instead of Nodes and raises typing errors. We patched this in https://github.com/mlcommons/croissant/pull/594/commits/93de08e9c6b111229fb0959e095171c5e1f4215d. A longer-term solution would...

- Look for the GitHub issue URL in the code. - The typing issue comes from the fact that we use `source.uuid` (type str) when we actually mean `source.target_uuid` (type...

Clean technical debt after `@id` migration

1

- Fix the editor (@marcenacp). - Hugging Face migration to 1.0 (@ccl-core). - Make name optional + make `self.name = self.id` by default (@ccl-core). - Use a proper `Uuid` class...

From: https://github.com/mlcommons/croissant/pull/559/files#r1500642879 Sources should have the form: ``` { "fileSet": ... # or "recordSet" or "fileObject" "extract": ... "transform": ... } ```