DerwenAI
Integrate with Snorkel for weak supervision use cases
Integrate with Snorkel for weak supervision use cases
- add a notebook to show use of weak supervision as a complement to PSL, KGE, etc.
- enhance the semantic tagging of the 14K ingredients in the full dataset for the recipe KG
While we're at this spike on weak supervision we should probably also go the extra step to integrate human-learn as an alternative approach for first-pass on a large set of annotations.
In other words, guid learners to build an HITL feedback loop that leverages PSL, KGE, Shape Prediction, etc., to get coverage on a much of the unlabeled cases as possible.
Eager to get this into spaCy pipeline through pytextrank integration.
Best case scenario: run scikit-learn pipelines in lale to leverage human-learn, snorkel, etc.
For example, see:
Might be useful to have a conversation with @koaning
AskMeAnything[tm]
That said, I'm also working on some bulk labeling tools at the moment. The idea is to build ipywidgets that can be combined in a lot of ways. I'm not 100% certain where they'll land. It might be human-learn but I might put them on another project. Human-learn is more meant for tabular data and scikit-learn pipelines.
Many thanks @koaning, I'll keep a watch on human-learn as this evolves.
And definitely, what we're trying to do with kglab is make other graph libraries easier to use as parts of pipelines that involve pandas, scikit-learn, spaCy, etc.
Along with snorkel above, it could be helpful to evaluation an integration with hover https://lnkd.in/g3ycUrH
Much more weak supervision is available on https://github.com/recognai/rubrix plus, this intended use case was more about data labeling (linking data into a graph) and fits better in specific labeling libraries.