https://www.youtube.com/watch?v=jVKGYu0_OCA

I finished a bit of work to make the dataset/ml system a bit less opaque: * tech.ml.dataset walkthrough - https://github.com/techascent/tech.ml.dataset/blob/master/docs/walkthrough.md * http://tech.ml xgboost early stopping and metrics - https://github.com/techascent/tech.ml/blob/master/docs/xgboost_metrics.md

I think Weld is interesting. It is language to write math in that can do more optimizations and optimize across function boundaries similar to TVM. It has a clean C api and a python api and has a few examples where they show amazingly good performance: https://github.com/weld-project/weld

https://news.ycombinator.com/item?id=21036503