Nick Becker

@DaPiePiece We've generally focused on enabling cuML with other HPO tools, but we're always looking to better understand user needs. Could you file a Github issue on cuML describing how built-in HPO tooling would affect your work? github.com/rapidsai/cuml/…

@becker_data Oooh alright, I'll give it a try, sounds good. Btw, any plans on native HPO in cuML? I'm a big fan of how cuML is practically a superset of sklearn, makes migrating sklearn code extremely easy.

God I despise PyTorch, who made this garbage? The literal only reason I'm using it over sklearn or cuML is cause sklearn doesn't run on a GPU and cuML doesn't have hyperparameter optimisation

@DaPiePiece SVMs are a perfect example of this scenario (results will vary based on data size, CPU, and GPU). In this example, my CPU is relatively weak so I can get benefits even with just 10K rows. gist.github.com/beckernick/ff0…

@DaPiePiece Using scikit-learn's GridSearchCV with cuML models should generally give you a large speedup if the model training itself is the bottleneck. If the cuML speedup is > the number of parallel models you can run via n_jobs without sacrificing performance, you'll come out ahead.

@becker_data Hi yeah my issue is that my data set is masisve (10000*64 10x10 images) so kinda need to pass this onto a GPU, spent 6+ hours trying to GridSearchCV using sklearn to no avail, and passing between GPU cuML data to CPU sklearn data is, from what I've heard, slower

🚀 Slideflow 2.0 is here! Take your #digitalpathology research to the next level with extended MIL support, expanded feature generators, and enhanced stain normalization. Deploy & visualize models with Slideflow Studio. 🔬💻 Check it out at slideflow.dev

@bilzrd @bilzard It looks like this may work smoothly in the cuDF 23.04 nightlies. Please let us know if things work for you!

@bilzrd @bilzard which version does wandb require? cuDF has relaxed its protobuf constraint to >=4.21.6,<4.22 in the 23.04 nightly packages

@StonewrightAI @RAPIDSai Hi @StonewrightAI , we've just updated the RAPIDS website and it includes new installation instructions and advanced install resources. You can learn more at #quick-start" target="_blank" rel="nofollow noopener">rapids.ai/#quick-start

@RAPIDSai hello! Why am I having a hard time installing cudf? I believe I have compatible versions of python, CUDA, cupy, cuDNN, zlib, and more. Can you please point me toward installation resources? Thank you.

@rblourenco You can now pip install cuDF, cuML, and cugraph on Colab, which should be much faster!

It took 29 minutes. Now able to work. But every time that I will restart, will need to reinstall 😭😭😭

22 minutes and counting while installing @RAPIDSai on @GoogleColab... 😭

@AIMLDS XGBoost now provides built-in support for categorical features. If you haven't revisited categorical handling in XGBoost in a while, it's worth a look. developer.nvidia.com/blog/categoric…

Deciding between #Catboost and #XGBoost for feature selection? Both are powerful gradient boosting algorithms, but Catboost has the added benefit of being able to handle categorical features natively. #MachineLearning #FeatureSelection

@PtrPomorski @BenHarlander XGBoost now supports categorical features via optimal partitioning. If you haven't revisited categorical handling in XGBoost in a while, it's worth a look #entry-content-comments" target="_blank" rel="nofollow noopener">developer.nvidia.com/blog/categoric…

Truth be told, currently there’s no significant performance difference between xgboost, random forest and lightgbm, so use whatever (just optimise hyperparams). Only when you have categorical features I highly recommend catboost, it beats them all.

@zaialamm You can use a GPU by clicking "Runtime -> Change runtime type" from the dropdown menu. You can then install and use XGBoost and other GPU-accelerated ML libraries like cuML rapids.ai/pip.html

@becker_data I know. But I use Google Colab for modelling and I think it doesn't support CUDA, does it? 🤔

Why is training with XGBoost very sloooww? I hate waiting ffs 😩

Here are the CLI 5 commands it takes to get a multi-node multi-GPU Data Science cluster running in the cloud with RAPIDS and Dask. 1/7

Moving between CPU and GPU environments just got a whole lot easier. Find out more in this new blog by @_rjzamora and @quasiben: medium.com/rapids-ai/easy…

Working with BERTopic and not meeting your New Year's wait loss goals? This blog by @MaartenGron and the RAPDIS team on speeding up BERTopic on CPU, and going the next mile with GPU, will meet you where you are at to keep those resolutions strong. medium.com/rapids-ai/fast…

@SNbarbat Are you experiencing a specific error or issue?

Does anyone know how to install rapids #cuML in a #AWS p3.2xlarge? Using the Deep Learning Pythorch image and Ubuntu 20.20

@tom_gxt @RAPIDSai @PyTorch You should now be able to smoothly combine the standard RAPIDS and PyTorch installation commands using the CUDA Toolkit version PyTorch wants. We now run tests every RAPIDS release to verify creating a joint environment works. This blog is separate, exciting new functionality!

We are kicking off an exciting year for RAPIDS with this great blog on using @PyTorch with RAPIDS Memory Manager! medium.com/rapids-ai/pyto…

Anyone interested in topic modeling or #NLP should check out BERTopic. The upcoming release will include deeper support for #RAPIDS and cuML to help you get results faster and process larger datasets (among other amazing updates)!

Preview #5: Heavyweight BERTopic! Last week was a lightweight update... this week a heavyweight💪 Better and more native integration of cuML's HDBSCAN in BERTopic! Use the full force of GPU acceleration to scale topic modeling to millions of documents. A preview thread👇🧵