<!-- TITLE: Scalinglogisticregression -->
<!-- SUBTITLE: A quick summary of Scalinglogisticregression -->
+ Logistic regression normally is amazing in Python through [statsmodels -> Logit](https://www.statsmodels.org/stable/generated/statsmodels.discrete.discrete_model.Logit.fit.html#statsmodels.discrete.discrete_model.Logit.fit)
+ [This paper](https://faculty.ucmerced.edu/frusu/Papers/Conference/2019-ipdps-sgd-cpu-gpu.pdf) shows a 100x speedup using the GPU vs. CPU
+ [Hyperlearn](https://github.com/danielhanchen/hyperlearn) uses CUDA to speed up Sklearn operations. Limited
+ [This paper](https://pdfs.semanticscholar.org/e8e6/103106cc1f76c8d5511b023af0cc723dffaf.pdf?_ga=2.139388795.954021285.1584974698-167895077.1584203771) uses OpenCL to do logistic regression on GPU ([Code here](https://github.com/hamadazahera/GPU-Training-Logistic-Regression))
+ [RAPIDS](https://rapids.ai/about.html) cares a lot about this stuff, [has lots of libraries](https://docs.rapids.ai/api) (including [networks](https://github.com/rapidsai/cugraph)), and [has all kinds](https://github.com/rapidsai/cuml) of regression, classification, etc. including logistic regression.
+ This was "incubated by NVIDIA," seems really useful
+ It's hard to set up, probably best to [use a prebuilt AMI](https://aws.amazon.com/marketplace/pp/B07Q88KH8C?ref=cns_srchrow)