Hello @lala,
I agree with @P_Bartell on making a Wish List request.
One question for you is how deep is deep? JMP Pro currently offers a 2 layer approach with as many hidden nodes as you want to add in each layer and you have 3 activation functions to choose from for those layers and hidden nodes.
On top of that you can use boosting that fits an additive sequence of models for the residuals scaled by the learning rate. Basically you can make this number as big as you want and JMP will use only as many boosting steps as it needs to give the best fit. Just remember the bigger the number the longer it will take to fit the model.
There is also built in cross-validation or you can use three level validation to help avoid overfitting of the data.
Hope this is helpful.
Kind regards,
Bill