GPU usage in Alteryx algorithms?

[Atabarezz]

Does Alteryx ML tools benefit from computers GPU in terms of parallel algorithm execution?

I have seen some R examples and in Revolution Analytics blog there is mention of such a capability...

But couldn't be sure of...

GPU Analytics thru R: http://www.r-tutor.com/gpu-computing
Revolution analytics: http://blog.revolutionanalytics.com/2015/06/computing-with-gpus-in-r.html

Why this capabilirt is expected is detailed in here: http://www.fuzzyl.com/products/gpu-analytics/

Best Regards

Altan

[JohnJPS]

Wow, time flies... looks like they got their clean build on Windows... however not much for recent activity...

• Last update 10 months ago: https://github.com/cdeterman/gpuR
• Last Update 2 years ago: https://github.com/nullsatz/gputools

I think the explosion of TensorFlow, Keras and so forth on Python, is probably what caused these to fall by the wayside.

[JohnJPS]

Granted, looks like R supports all that too... 

• CRAN ugly page: https://cran.r-project.org/web/views/HighPerformanceComputing.html
• NVidia nice write-up, (albeit 4 years old): https://devblogs.nvidia.com/accelerate-r-applications-cuda/

[JohnJPS]

Can you tell I'm just clicking around....?   Anyway, this also looks pretty cool: 

• https://www.rdocumentation.org/packages/h2o4gpu/versions/0.2.0

So here's a thought:

• Run Designer from a browser
• Input tool dump to Spark
• Use the above package against Spark for crazy fast ML stuff

Blue-sky Friday thoughts... LOL.

[Atabarezz]

+1 there is some flirting between H2O and alteryx these days...

[JohnJPS]

h2o driverless AI looks fascinating... if you could throw data prepped in Alteryx at a driverless AI system, that would be a very powerful solution.

[Alison_Cossette]

They just announced the partnership yesterday.  Driverless AI customers will now have an integration option so they can run DAI models on their alteryx data from within alteryx. 

https://youtu.be/LUPjAHV77PQ

[JohnJPS]

Hi @Alison_Cossette,

That's great stuff.  Elastic compute probably is the best direction for any enterprise scale machine learning, and that little demo ties it right in with Alteryx.  Very cool.

Thanks,

John

[JohnJPS]

Check out ludwig, a driverless deep learning solution recently open-sourced by Uber. It uses tensorflow and is based on python.

So... with the right python/tensorflow installation in place, (tricky but a one-time affair) you could probably build an analytic app to create a deep neural net by, literally, having the user pick a dataset, check the target, check/uncheck predictors, apply some range limits if necessary... and off it goes: user builds a deep learning model without knowing anything at all about deep learning.

You could probably build a shap explainer into the same app and provide that as output along with the trained model; still without the user knowing anything other than that they can get pretty good predictions along with answers to "why that prediction".

[JohnJPS]

(Also posted elsewhere but perhaps more relevant here)

KNIME (open source) has done some interesting thinking on Guided Analytics.

In particular they have an entire example (requiring enterprise version to run, so no longer open source) of Automated Machine Learning that looks not unlike like Ludwig, except it is ready-to-go as point-and-click from a browser, including ample room for user input; heck Ludwig could even be tied into the back-end as one of the models considered. Very robust. Would love to see something like that in Alteryx. Seems doable.