cancel
Showing results for 
Show  only  | Search instead for 
Did you mean: 
%3CLINGO-SUB%20id%3D%22lingo-sub-780795%22%20slang%3D%22en-US%22%20mode%3D%22UPDATE%22%3EBuilt-in%20model%20hyperparameter%20tuning%3C%2FLINGO-SUB%3E%3CLINGO-BODY%20id%3D%22lingo-body-780795%22%20slang%3D%22en-US%22%20mode%3D%22UPDATE%22%3E%3CP%3E%3CSPAN%3E%3CSTRONG%3EThanks%20to%20Russ%20Wolfinger%20and%20JMP%20Super-Users%20like%20%3CA%20href%3D%22https%3A%2F%2Fcommunity.jmp.com%2Ft5%2Fuser%2Fviewprofilepage%2Fuser-id%2F12549%22%20target%3D%22_blank%22%3E%40SDF1%3C%2FA%3E%2C%26nbsp%3Bthere%20are%20now%20add-ins%20that%20allow%20easy%20tuning%20of%20hyperparameters%20for%20XGBoost%2C%20neural%20nets%2C%20bootstrap%20forests%2C%20and%20others%2C%20but%20I%20would%20really%20like%20to%20see%20these%20integrated%20into%20JMP.%20One%20thing%20I%20would%20like%20to%20do%20is%20test%20out%20thousands%20of%20neural%20networks%20with%20multiple%20Y's%2C%20which%20cannot%20yet%20be%20done%20with%20the%20add-in%20(and%20is%20not%20possible%20at%20all%20for%20XGBoost%2C%20as%20far%20as%20I%20can%20tell).%26nbsp%3B%3C%2FSTRONG%3E%3C%2FSPAN%3E%3C%2FP%3E%0A%3CP%3E%26nbsp%3B%3C%2FP%3E%0A%3CP%3E%26nbsp%3B%3C%2FP%3E%0A%3CP%3E%3CSTRONG%3ESometimes%2C%20we%20need%20to%20predict%20more%20than%20one%20thing.%20For%20instance%2C%20maybe%20you%20want%20to%20optimize%20a%20process%20while%20minimizing%20costs.%20With%20several%20JMP%20modeling%20platforms%2C%20you%20can%20create%20a%20two-Y%20model%2C%20then%20use%20the%20desirability%20analysis%20under%20the%20profiler%20to%20do%20this%2C%20but%20you%20cannot%20build%20thousands%20of%20dual-Y%20models%20using%20neural%20networks%2C%20bootstrap%20forests%2C%20or%20even%20standard%20least%20squares.%20I%20think%20this%20would%20really%20increase%20the%20capacity%20of%20JMP%20Pro%20to%20do%20some%20hardcore%20predictive%20modeling.%26nbsp%3B%3C%2FSTRONG%3E%3C%2FP%3E%0A%3CP%3E%26nbsp%3B%3C%2FP%3E%0A%3CP%3E%26nbsp%3B%3C%2FP%3E%3C%2FLINGO-BODY%3E%3CLINGO-LABS%20id%3D%22lingo-labs-780795%22%20slang%3D%22en-US%22%20mode%3D%22UPDATE%22%3E%3CLINGO-LABEL%3EAdvanced%20Statistical%20Modeling%3C%2FLINGO-LABEL%3E%3CLINGO-LABEL%3EPredictive%20Modeling%20and%20Machine%20Learning%3C%2FLINGO-LABEL%3E%3C%2FLINGO-LABS%3E%3CLINGO-SUB%20id%3D%22lingo-sub-799573%22%20slang%3D%22en-US%22%20mode%3D%22CREATE%22%3ERe%3A%20Built-in%20model%20hyperparameter%20tuning%20-%20Status%20changed%20to%3A%20Acknowledged%3C%2FLINGO-SUB%3E%3CLINGO-BODY%20id%3D%22lingo-body-799573%22%20slang%3D%22en-US%22%20mode%3D%22CREATE%22%3E%3CP%3EHi%26nbsp%3B%3CA%20href%3D%22https%3A%2F%2Fcommunity.jmp.com%2Ft5%2Fuser%2Fviewprofilepage%2Fuser-id%2F12111%22%20target%3D%22_blank%22%3E%40abmayfield%3C%2FA%3E%26nbsp%3B%2C%20thank%20you%20for%20submitting%20this%20request%2C%20it%20has%20been%20captured%20in%20our%20system.%20You're%20right%20that%20Russ%20and%20team%20have%20done%20a%20really%20nice%20job%20with%20this%20in%20XGBoost%20and%20Torch!%26nbsp%3B%3C%2FP%3E%0A%3CP%3EIf%20you%20haven't%20seen%20it%2C%20this%20add-in%20might%20help%3A%26nbsp%3B%3CA%20href%3D%22https%3A%2F%2Fcommunity.jmp.com%2Ft5%2FJMP-Add-Ins%2FNeural-Network-Tuning%2Fta-p%2F662666%22%20target%3D%22_blank%22%3Ehttps%3A%2F%2Fcommunity.jmp.com%2Ft5%2FJMP-Add-Ins%2FNeural-Network-Tuning%2Fta-p%2F662666%3C%2FA%3E%3C%2FP%3E%3C%2FLINGO-BODY%3E
Choose Language Hide Translation Bar

Built-in model hyperparameter tuning

Thanks to Russ Wolfinger and JMP Super-Users like @SDF1, there are now add-ins that allow easy tuning of hyperparameters for XGBoost, neural nets, bootstrap forests, and others, but I would really like to see these integrated into JMP. One thing I would like to do is test out thousands of neural networks with multiple Y's, which cannot yet be done with the add-in (and is not possible at all for XGBoost, as far as I can tell). 

 

 

Sometimes, we need to predict more than one thing. For instance, maybe you want to optimize a process while minimizing costs. With several JMP modeling platforms, you can create a two-Y model, then use the desirability analysis under the profiler to do this, but you cannot build thousands of dual-Y models using neural networks, bootstrap forests, or even standard least squares. I think this would really increase the capacity of JMP Pro to do some hardcore predictive modeling. 

 

 

1 Comment
mia_stephens
Staff
Status changed to: Acknowledged

Hi @abmayfield , thank you for submitting this request, it has been captured in our system. You're right that Russ and team have done a really nice job with this in XGBoost and Torch! 

If you haven't seen it, this add-in might help: https://community.jmp.com/t5/JMP-Add-Ins/Neural-Network-Tuning/ta-p/662666