Timeline for Why a decision tree can't learn this simple rule?
Current License: CC BY-SA 4.0
10 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Oct 12, 2020 at 18:44 | history | edited | klumbard | CC BY-SA 4.0 |
cleaned things up a bit
|
| Oct 12, 2020 at 18:22 | comment | added | Ben Reiniger | fwiw, There were some discussions for model-based recursive partitioning in sklearn: github.com/scikit-learn/scikit-learn/issues/11647 | |
| Oct 12, 2020 at 16:42 | history | edited | klumbard | CC BY-SA 4.0 |
added 7 characters in body
|
| Oct 12, 2020 at 16:30 | comment | added | Sycorax♦ |
@HansHupe sklearn is convenient, but limiting yourself to the models that it implements will not give you very deep coverage of the kinds of problems you might need to solve. Indeed, you've discovered one such blindspot in the case of this problem.
|
|
| Oct 12, 2020 at 16:02 | comment | added | HansHupe | Sorry for complaining, but it looks like that no algorithm of the widely used ML toolbox sklearn in Python is able to learn that rule in a satisfying way (maybe MLPRegressor could but it would be overshooting I think and needs a lot of tuning). For me this structure of problem is a very practical challenge. | |
| Oct 12, 2020 at 15:54 | comment | added | Sycorax♦ | @HansHupe This is kind of like complaining that a screwdriver doesn't do a very good job of driving nails. There are many types of machine learning models because there are many types of problems, and a model which does well against one problem may not do well against another type of problem. For more information, you may be interested in reading about the so-called "No Free Lunch Theorem." en.wikipedia.org/wiki/No_free_lunch_theorem | |
| Oct 12, 2020 at 15:46 | history | edited | klumbard | CC BY-SA 4.0 |
edited body
|
| Oct 12, 2020 at 15:41 | comment | added | HansHupe | Thx. for sharing this I got the idea. I think model trees (M5) are a nice approach, unfortunately poorly supported in Python (sklearn). But honestly I am surprised that such simple interaction rules are obviously not an easy thing for standard ML methods. | |
| Oct 12, 2020 at 15:37 | vote | accept | HansHupe | ||
| Oct 12, 2020 at 15:35 | history | answered | klumbard | CC BY-SA 4.0 |