8000 Request more criterion for random forest regression · Issue #5368 · scikit-learn/scikit-learn · GitHub
[go: up one dir, main page]
Skip to content

Request more criterion for random forest regression #5368

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Closed
boscotsang opened this issue Oct 8, 2015 · 16 comments
Closed

Request more criterion for random forest regression #5368

boscotsang opened this issue Oct 8, 2015 · 16 comments
Labels
Enhancement Moderate Anything that requires some knowledge of conventions and best practices

Comments

@boscotsang
Copy link

Current random forest regressor only support for 'mse'. Can more criterions such as mean square of percentage error can be support by scikit-learn?
@amueller
Copy link
Member
amueller commented Oct 9, 2015

Is that the same as mean squared relative error? Do you have a reference?

@jmschrei
Copy link
Member

New criteria can be supported fairly easily. They need to be added to sklearn/tree/_crtierion.pyx, and then have the appropriate dictionaries updated.

@boscotsang
Copy link
Author

@amueller https://www.kaggle.com/c/rossmann-store-sales/details/evaluation defined the RMSPE criterion.

@amueller
Copy link
Member

Not sure that is the most common name for that (also the formula is not really for percentages).
I think we could add that, it seems interesting. See what @jmschrei said ;)

@glouppe
Copy link
Contributor
glouppe commented Oct 19, 2015

New criteria can be supported fairly easily.

It is easy when you have understood the codebase, but it is true that people may not know where to look at when they arrive. Maybe we could make it possible to pass Criterion object directly, as we do for Splitter? Then one would not have to hack around.

@glouppe glouppe added Enhancement Moderate Anything that requires some knowledge of conventions and best practices labels Oct 19, 2015
@jmschrei
Copy link
Member

This is true. What do you mean pass Criterion object directly?

@Sandy4321
Copy link

so when it will be done?
lets do it pls!!

On Wed, Oct 21, 2015 at 6:58 PM, Jacob Schreiber notifications@github.com
wrote:

This is true. What do you mean pass Criterion object directly?

—
Reply to this email directly or view it on GitHub
#5368 (comment)
.

@jmschrei
Copy link
Member
jmschrei commented Nov 1, 2015

I think that is kind of the wrong attitude to have. If there's a particular feature you'd like, you should attempt to submit a PR incorporating it, or be very specific about which criterion you'd like and maybe someone will take it up for you.

@betatim
Copy link
Member
betatim commented Jan 12, 2016

If there is a concrete need/idea for a new criterion I'd be interested in doing the coding work to implement it (for my education about how the decision tree internals work).

@glouppe
Copy link
Contributor
glouppe commented Jan 12, 2016

There was some attempt to implement MAE at #6039. I think this criterion really is missing at the moment and could be used at many occasions.

@fjanoos
Copy link
fjanoos commented Jan 26, 2016

As a follow up, how much effort would it be to implement the set of criteria currently available in GradientBoostingRegressor, namely lad and huber losses ?

@fjanoos
Copy link
fjanoos commented Feb 3, 2016

Hi - I've written up a cython extension for a LAD (L1 norm) criterion that plugs into the tree based classifiers. Would this be of wider interest - to include back into sklearn ?

@raghavrv
Copy link
Member
raghavrv commented Feb 3, 2016

Could you make a PR? (An attempt at LMAD is also done at #6039. You could refer that)

@fjanoos
Copy link
fjanoos commented Feb 3, 2016

The implementation there is for MAD - namely the L1 norm about the mean of each class. This is a LAD implementation that uses medians.
I'll create a PR.

@raghavrv
Copy link
Member
raghavrv commented Feb 3, 2016

Ah okay! thanks for the clarification :) I'm pretty much new to all this ;)

@rth rth mentioned this issue Oct 3, 2019
Open
@lorentzenchr
Copy link
Member

Meanwhile we have MAE #6667 and the Poisson deviance #17386. More concrete proposals could be opened separately. Therefore closing this issue.

@jimthompson5802 jimthompson5802 mentioned this issue Dec 11, 2023
Closed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
Enhancement Moderate Anything that requires some knowledge of conventions and best practices
Projects
None yet
Milestone
No milestone
9 participants
@amueller @glouppe @fjanoos @betatim @jmschrei @boscotsang @raghavrv @Sandy4321 @lorentzenchr
0