932021_en-US

liliconcom

Multivariate Fundamentals

Hello SE's from across the SEas

 

got a quick regarding multivariate analysis, & what actually is useful in your previous experience as engineers/scientists & what topics you think people SHOULD know. 

For Example: Is PCA actually really useful when you have computers that can handles a bunch of variables? - or are there fundamentals of this technique used in cool & relevantway (bonus points if it works in jmp pro) 

or cool techs you know

**like paring a clustered group with a parallel plot** 

thanks and have a good feb.

-lovy 

Advanced Statistical ModelingData Blending and CleanupPredictive Modeling and Machine LearningRe: Multivariate Fundamentals

3 thoughts come to mind:

1. Years ago when I was in the real world, I took the Multivariate course from Di Michelson. It really opened my eyes to what was going on in the whole "Multivariate" menu in JMP. I don't see that course in the On Demand section. I wonder if the notes are available somewhere. I'm tagging @ruthhummel since I'm not smart enough to find Di's tag.
2. To your question on Principal Components, I think Cluster Variables is mostly a better tool for variable reduction, but I have a fuzzy memory of (I think?) @Phil_Kay doing a video or blog showing how Principal Components can be very useful for seeing broad groupings that can be or should have been derived from process knowledge.
3. Multivariate Embedding is a fast tool for visualizing very wide data, and it's a JMP Pro tool. The sample file "Prostate Cancer.jmp" shows something really interesting and takes about 2 seconds. If my data had something like this, I'd be really interested in those 11 rows:
4. 

Re: Multivariate Fundamentals

awesome! thanks jed 

i personally haven't used these techniques much in my engineering roles, so its nice to hear others' perspectives 

 

i agree with the cluster variable over PCA since we're not dealing with abstract columns but find "the most representative" column for a cluster (a real column)

Re: Multivariate Fundamentals

You can get to the materials from here: Education Course Materials for Internal Use Only - JMP User Community. One day, I'll convert to some sort of online content.

Re: Multivariate Fundamentals

It's a good question. In Europe, the makers of SIMCA and Unscrambler did a really good job evangelizing for multivariate methods, especially PCA and PLS. Particularly in the biotech and chemicals industries. But I'm not sure they ever really proved the value. The community of users has stayed small but very passionate! If you ever meet someone using PCA/PLS and they are not using JMP, be very careful before committing any time selling to them.

Anyway, I've attached a journal of some case studies that I've used to talk about this topic. The final case study data table is big, so I've attached that separately. Happy to walk through these. They really should be in the TRL of course! 

Re: Multivariate Fundamentals

thanks phil! i'll look into through these examples 

Re: Multivariate Fundamentals

PCA is widely used in population genetics/genomics.

I used PCA during my master's to study HIV population evolution. I did it in Python, there are also many R packages available. I wish I knew about JMP back then.