search

ahoshiyar / ordPens

Ordinal PCA, Selection and/or Smoothing of Ordinal Predictors
GNU General Public License v2.0
2 stars 2 forks source link

Possible related methods #8

Closed mingzehuang closed 3 years ago

mingzehuang commented 3 years ago

It's nice to use penalty term's to deal with ordinal data:) It's straightforward and solid in theory. I wonder if it's possible to compare penalty method with Gaussian copula model for latent correlation (of ordinal data) as the package we did before latentcor ? Thanks!

ahoshiyar commented 3 years ago

Thank you for this interesting reference! Actually, the methods differ in their point of view. The latent Gaussian copula model has been cited in the paper.md now with a few notes on latent variables and potential future research. However, I think a detailed discussion or comparison of the ordinal penalty & the latentcor would lie beyond the scope of the JOSS manuscript.

mingzehuang commented 3 years ago

Thank you for your reply, @ahoshiyar ! I definitely agree with you! And also many thinks for your citation! JOSS does focus on software rather than math foundation:) I was asking this question because we're doing related things and I want to hear your opinion about this:) Thank you for your comment:) Also I'm definitely fine with your math and code in "statement of need" part and "ordPens in action":) Since JOSS focus on software, I'm afraid that someone may have difficulty to understand what your math is doing at the beginning. A possible suggestion may be switching your order of two parts in "statement of need". First part you're going to explain there's no good package to address non-linearity and biasness caused by ordinal data in PCA even in IBM SPSS. (This reminds readers what you're going to do.) Then you go to your math:) But this is just my personal opinion. Overall it's definitely great explanation:)

mingzehuang commented 3 years ago

Looks great, thanks!