3 ms·

Now you can contact the authors and tell them to replace the publication by your 4 lines. And tree based models is not what you described, at most that could b

Now you can contact the authors and tell them to replace the publication by your 4 lines. And tree based models is not what you described, at most that could be a tree classifier. EDIT: First of all sorry for my comment, I think you did not like it. My point is that what you state as obvious on how tree based methods work is most probably not what makes them powerful but the fact that you have a bunch of them. To me if there is some intuition, statistics is more prevalent than separating spaces in hyperplanes. Here you can see the boundaries of a random forest compared to a single tree, the more dimensions you add the more blurry and unintuitive in terms of hyperplanes it gets: https://scikit-learn.org/stable/_images/sphx_glr_plot_classifier_comparison_001.png https://scikit-learn.org/stable/_images/sphx_glr_plot_classi...

spywaregorilla 4y ago
- abduhl 4y ago
  So it sounds like you’d need your 4 lines of pseudo-code from your first post and then another few lines of comments to describe how your 4 lines replace the authors’ publication? Something like “All tree models are partitions of the space based on training data, and the extension to any particular tree model is an obvious exercise left to the reader”?
  - JoshCole 4y ago
    You are saying it sounds as if he his explicit denial is what he was saying? You're like a kid that hears another kid getting upset at a mispronunciation of their name. "My name is John," he whines. "Sounds like you are Joonnna," you taunt. It is an exceedingly childish thing to do. Maybe others here aren't mature enough to realize you are being a bully, but I'm not stupid enough to think that "not X" said with emotional force means something "sounds like X". I'd also rather give this response rather than making spyware go through the humiliation of repeating himself. He got downvoted because he cussed, not because intentionally distorting what people say is funny.
  - spywaregorilla 4y ago
    Why have 2 separate people asserted I have issue with the author's publication? My point is that tree based models are models that partition the space. This is not obvious. But the concept of partitioning the space is a very obvious thing to try because it just makes visual sense. Take a look at the example linked above: https://scikit-learn.org/stable/_images/sphx_glr_plot_classifier_comparison_001.png https://scikit-learn.org/stable/_images/sphx_glr_plot_classi... All models indirectly partition the space. Tree based models partition the space explicitly. Other model types do so implicitly.
    - JoshCole 4y ago
      Patience will be necessary; a bad reading of your meaning will be a bit sticky because it allows others to pick up computation where others left off. In the short term it looks like misunderstanding. Over time it means a great deal more thoughtfulness than any one person would have been able to give.
spywaregorilla 4y ago
I did not like the comment. It is needlessly aggressive and implies I have the arrogant view that my post is better than the author's paper. I do not. I also suspect your hypothesis is not correct. The top row of your link is a good example. The NN tries to infer a gradient but that's really tough to do with limited data. That is to say, the tree based models will locally fit their partitions to the exact training data and the NN will try to view it in the big picture filling in the gaps. Tree based model works better for most real world tabular data. The paper concludes something similar: > This superiority is explained by specific features of tabular data: irregular patterns in the target function, uninformative features, and non rotationally-invariant data where linear combinations of features misrepresent the information. I daresay my perspective is better aligned with the paper than yours. Are YOU trying to replace the publication with your 4 lines?
- melenaboija 4y ago
  Totally agree and sorry again for my wording as I probably miss understood what you meant. And no, I am trying to replace what I consider is a wrong intuition (tree methods are strong models mostly because single trees separate data in hyperplanes) with my 4 lines. This is just my opinion.
stonemetal12 4y ago
To be fair "Obvious" doesn't mean correct or proven. Even if their results boil down to obvious method is better than more complicated method it is still worth a paper.

spywaregorilla 4y ago

abduhl 4y ago
So it sounds like you’d need your 4 lines of pseudo-code from your first post and then another few lines of comments to describe how your 4 lines replace the authors’ publication? Something like “All tree models are partitions of the space based on training data, and the extension to any particular tree model is an obvious exercise left to the reader”?
- JoshCole 4y ago
  You are saying it sounds as if he his explicit denial is what he was saying? You're like a kid that hears another kid getting upset at a mispronunciation of their name. "My name is John," he whines. "Sounds like you are Joonnna," you taunt. It is an exceedingly childish thing to do. Maybe others here aren't mature enough to realize you are being a bully, but I'm not stupid enough to think that "not X" said with emotional force means something "sounds like X". I'd also rather give this response rather than making spyware go through the humiliation of repeating himself. He got downvoted because he cussed, not because intentionally distorting what people say is funny.
- spywaregorilla 4y ago
  Why have 2 separate people asserted I have issue with the author's publication? My point is that tree based models are models that partition the space. This is not obvious. But the concept of partitioning the space is a very obvious thing to try because it just makes visual sense. Take a look at the example linked above: https://scikit-learn.org/stable/_images/sphx_glr_plot_classifier_comparison_001.png https://scikit-learn.org/stable/_images/sphx_glr_plot_classi... All models indirectly partition the space. Tree based models partition the space explicitly. Other model types do so implicitly.
  - JoshCole 4y ago
    Patience will be necessary; a bad reading of your meaning will be a bit sticky because it allows others to pick up computation where others left off. In the short term it looks like misunderstanding. Over time it means a great deal more thoughtfulness than any one person would have been able to give.

spywaregorilla 4y ago

I did not like the comment. It is needlessly aggressive and implies I have the arrogant view that my post is better than the author's paper. I do not. I also suspect your hypothesis is not correct. The top row of your link is a good example. The NN tries to infer a gradient but that's really tough to do with limited data. That is to say, the tree based models will locally fit their partitions to the exact training data and the NN will try to view it in the big picture filling in the gaps. Tree based model works better for most real world tabular data. The paper concludes something similar: > This superiority is explained by specific features of tabular data: irregular patterns in the target function, uninformative features, and non rotationally-invariant data where linear combinations of features misrepresent the information. I daresay my perspective is better aligned with the paper than yours. Are YOU trying to replace the publication with your 4 lines?

melenaboija 4y ago
Totally agree and sorry again for my wording as I probably miss understood what you meant. And no, I am trying to replace what I consider is a wrong intuition (tree methods are strong models mostly because single trees separate data in hyperplanes) with my 4 lines. This is just my opinion.

stonemetal12 4y ago

To be fair "Obvious" doesn't mean correct or proven. Even if their results boil down to obvious method is better than more complicated method it is still worth a paper.