Discussion of article "Advanced resampling and selection of CatBoost models by brute-force method" - page 6
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
I didn't see that in the textbooks. I know you can improve quality by combining good models. But not the other way round :)
Well, that's exactly what the forest does. All the good and the bad. And the forest was made by very cool mathematicians and statisticians.
And they probably tried to combine only the best models (if that's what you and I thought).Well, that's exactly what the forest does. All the good and the bad. And the forest was made by very cool mathematicians and statisticians.
And they probably tried to combine only the best models (if that's what you and I are thinking).even the forest has a reasonable limit of 50-100 trees, empirically derived by someone, it doesn't make sense anymore
pooling the best ones is common practice. On kaggle everyone likes to stack boosts. At least the meme used to go
even a forest has a reasonable limit of 50-100 trees, empirically deduced by someone, it doesn't make sense anymore.
You have just 50 clustering variants. It will be fine to average them.
Yes. There's no point. Time is wasted, and the quality gain is very little.
You have just 50 clustering variants. It would be fine to average them out.
I see the point of only clustering the best ones.
pooling the best is common practice. On kaggle everyone likes to stack boosts. At least the meme used to go
I guess it's not the Breiman's who sit on kaggle)))))) So they are experimenting...
I see the point of only putting the best ones together.
The Breiman's are probably not sitting on kaggle))))) So they are experimenting...
these are the ones who won the contests )
Try both and compare the result on the exam sample.
No, there's no point in adding bad models. By definition.
In training, averaging is one thing, but averaging the trained is another. Here you are deliberately shooting yourself in the foot by adding bad ones. They introduce error and that's it. And there's no such practice, I haven't seen it anywhere.
plus imagine the cost of getting a signal from 50 models, the brakes on testing.No, there's no point in adding bad models. By definition.
In training, averaging is one thing, but averaging the trained is another. Here you are deliberately shooting yourself in the foot by adding bad ones. They introduce error and that's it.
No, there's no point in adding bad models. By definition.
In training, averaging is one thing, but averaging the trained is another. Here you are deliberately shooting yourself in the foot by adding bad ones. They introduce error and that's it. And there's no such practice, I haven't seen it anywhere.
plus imagine the cost of getting a signal from 50 models, the brakes on testing.That's what happens in any random forest.