Vincent D. WarmerdamLast week Jay Allamar interviewed me to discuss some of the tools I've been working on in the past two years.
If you haven't seen it, we discuss human-learn, doubtlab, embetter, and bulk!
Watch-able here:
https://www.youtube.com/watch?v=KRQJDLyc1uM
YouTube
The first tool, human-learn, gives you scikit-learn compatible tools to just turn your domain knowledge into classifiers/regressors/detectors/transformers.
One main feature: you can turn functions with keyword arguments into gridsearch-able components!
The second tool, doubtlab, gives you a suite of tools to try and discover doubtful labels in your training data.
There are a bunch of reasons to doubt a label, and this library makes it easy to just try some.
Embetter is a utility library to make it easier to use embeddings from scikit-learn. It currently supports text and image embeddings, and it makes it super easy to build few-short classifiers from sklearn.
Soon it will also have fine-tunable components!
Finally there's bulk, which gives you a user-interface to easily bulk label training data by re-using embetter with UMAP.
There are a bunch more features/tools in the pipeline too. But I wanted to give a shoutout to @explosion, who have been very supportive of these tools.
Also, I've seen some of the internal demos. There's a lotta cool new stuff on the way.
