kenhktsui

Hi HN community,

In this release, I introduced synthetic data generation for text classification. As a result, one can build classifiers from scratch, even without a dataset. It achieves competitive results with synthetic data, comparable to using real data, in 5 benchmarks.

There are many open researches and implementation in the future: - research on synthetic data algorithm resulting in higher performance - agentic workflow of model evaluation, error analysis and model improvement - multilingual support If you are interested, please give a star, try it out, and contribute.

Detailed Blog: https://huggingface.co/blog/kenhktsui/anyclassifier

Best, Ken

Hi HN,

I would like to share this with all of you. AnyClassifier is a framework that helps you build a classifier without any label, with as limited coding as possible. As a machine learning engineer, one of the unavoidable but the most heavy lifting issues is to build and curate a high quality labelled data. By leveraging LLM annotation with permissive license, one can now label data at a better quality and at lightning speed ever. This is inspired by some of the challenges I faced daily in work and doing open source - it is built for machine learning engineer and software engineer, by machine learning engineer .

The project is still experimental, but I found it worked in some of my use cases. Feedbacks welcome, and feel free to contribute. See Future Roadmap.

I would love to have your feedback and also collaboration to make it better.

Best,

Ken

Thanks for links. Let me set up a battle between GNUGo and GoFormer. I was also inspired by how well it played chess, so I am trying to train one to challenge Go. It is still an early version as I am still enhancing the model with more data. But it already exhibited reasonable move as I played with it.