Consulting Something About HLS4ML

[Lxh]

Dear hls4ml group,

I’m Lxh, a major student, who is very interested in the hls4ml project. I also want to implement a ML algorithm on a FPGA board, and I’m so happy to find your work.

I carefully read the gitbook you published, and I have noticed that I have to get a Keras or Pytorch Model before I want to use hls4ml project to transform my ML algorithm. Correspondingly, the file must be saved in .h5 or .pt format.

However, I just want to transform a ML model which use scikit-learn library, such as decision tree or SVM but not deep learning which use Keras and Pytorch. And I only can save these model file in .pkl format. And on this condition, how can I use the hls4ml project? Because I noticed you mentioned the scikit-learn library, so I try to send this email for help.

Thanks for your time, I’m so sorry to disturb your work. I’m looking forward to your reply.

Best wishes,
Lxh.

[sioni summers]

Hi,

Thanks for getting in touch. You're right hls4ml is only for neural networks. For ensembles of decision trees like GradientBoosting{Classifier, Regressor}, RandomForest{Classifier, Regressor}, I can send you to this package, where we now maintain the BDT code: https://github.com/thesps/conifer. The API is similar to hls4ml, and there are examples there to get started. You can use scikit-learn .pkl files for those types of models. Having said that, we don't actually have support for conversion of 'DecisionTree{Classifier, Regressor}', only the ensemble methods I mentioned above. Although that's just a limitation of the conversion code, so we could look at adding it if that's what you're using. We don't have an implementation of SVM, either. I hope that helps!

Thanks,

Sioni

--
You received this message because you are subscribed to the Google Groups "hls4ml help" group.
To unsubscribe from this group and stop receiving emails from it, send an email to hls4mlhelp+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/hls4mlhelp/F5617ED2-AE91-44FD-875B-E64AD91AADA7%40njnet.edu.cn.