Alteryx ML Linear Regression Is Not Linear Regression - Misleading
- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Mute
- Printer Friendly Page
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Notify Moderator
Does anybody know why the ML Linear Regression within the Intelligence Suite is *actually* Elastic Net Regression?
Was using the Machine Learning tool within the Intelligence Suite lately and got positive results, so attempted to recreate in Python using sklearn and got fundamentally different results. Did some digging into the packages and found that what Alteryx calls "Linear Regression" is actually just a wrapper for the Elastic Net regressor, NOT the Linear Regressor. This feels super misleading and is a huge issue for converting things to or from Alteryx.
Also found out the "Random Forest" regressor tool is actually a wrapper for the "Extra Trees Regressor" from sklearn.
Would appreciate if anyone has some context on the decision to hide the names of the actual regressors. Thanks!
- Labels:
- Bugs & Issues
- General Feedback
- Question
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Notify Moderator
To answer the question, I'm not sure why it is... but it makes sense to be "Elastic Net Regression". Elastic Net Regression is a type of Linear regression and the model in Intelligence Suite has the default to normalize targets.
As for the Extra Trees regressor, this can have faster performance than Random Forest due to using the whole sample, however the IS model has bootstrapping on by default meaning it's very close to the standard Random Forest, while allowing for better performance and better results as there are more options to train the model in. I don't know how this works with parallel cores and my memory is rusty on this, but I think Random Forest wasn't very good with using parallel cores by itself.
Neither of these are technically wrong, they just use an enhanced version of the model, rather than the 19+ year old standard models.
