Machine Learning with Talend: Decision Trees

fr.talend.com

46 points·markgainor1··7 comments

It's pretty easy to visualize decision trees and there are already open source usable tools for that.

It would be way cooler if they could visualize Random Forest that can have thousands of these trees. This is what is actually used more frequently in practice.

Yeah, decision trees really shouldn't be used in practice. The variance of any single decision tree is too high to give you reasonable results.

They have their place - if you're more interested in inference than in prediction they can be useful tools.

Do you have a source for this? Pretty curious to see how that would play out.

No formal sources unfortunately, it's similar to using a linear regression model trained on historical data for inference (with some cross-validation/penalization to make sure we're not just finding signals in the noise).

We know for sure that there's a ton of bias (in the bias-variance sense), and that the predictions won't be nearly as good as a blackbox model, but we may glean some useful insights out of it.

Same idea with decision trees, in the past I've used them to find what factors influence customer support scores on online tickets, and we found some surprising and useful insights that were used for process improvements.

You don't need a source -- just read or print a decision tree after training one.

Talend lost me when they started making you pay for their software.

Not because I'm opposed to paying for software, but because Talend isn't that good. If I want to use paid software, I want it to be super refined and usable. Talend feels very open source.

Reply on news.ycombinator.com