Default Parameters for Elastic Net Logistic Regression

[#1]

Hi all,

I used to use JMP for data mining when I was in school, but have switched to using Python since then.  I recently used JMP to build an Elastic Net Logistic regression with my data since its output is more intuitive.  I attempted to replicate my results in python using the "SGDClassifier function" but my results were way off.  I believe it may be because this function requires specifying several parameters such as alpha, L1 ratio, initial learning rate for "adaptive" models, etc. Does anyone know the default settings for JMP's elastic net regression? I understand that the alpha is 0.9 but I'm unsure about the others.

Thank you in advance for any help!

Information on python function:

SGDClassifier Function in Python

[#2]

Since you are having a Python issue, I believe you should be asking a Python user discussion group for help.

[#3]

Well I was more so asking if anyone knows what these default parameters are in JMP so that I may be able to apply this to my python function because I’m trying to replicate the results that JMP provides

[#4]

I understand your question, but it basically boils down to the question:

"Can someone tell me the JMP default parameters, so I no longer have to use JMP?"  That doesn't seem to be an appropriate question for the JMP Discssion Group

[#5]

I don't agree.  I increasingly find that the default settings in software are quite important - there are so many parameters to set that good defaults mean a lot.  So, I would be very interested to compare the defaults between different programs.  This could be an important consideration in what software program to use.  While I am a devoted and happy JMP users, the reality is that the world is increasingly turning to open-source software, often under the belief that it is "free."  But it is not free - the cost involves (among other things) these very default settings that the question is being asked about.  So, I'd be interested in seeing an answer to the question as well as seeing the comparison the questioner is attempting.

[#6]

The parameters and hyper-parameters are dependent on the data. One way to proceed would be to ask the user to supply all the starting values. The other way is to use heuristics and default values to start with. JMP uses the second way because it reduces the burden placed upon users.

Different software developers and users might use different heuristics and default values for a given procedure. I think that such differences are moot because you can change the starting values to suit your preference or purpose.

In some procedures, JMP has used extensive research based on simulation to determine the best default values. There will still be cases where the default values are not appropriate or the best.