CODEWITHSUNDEEP
c++pythonrmachine-learning
Guy Adini
Guy Adini

Reputation: 5494

Implementations of local regression and local likelihood methods

I am looking for an efficient implementation of local regression (LOESS) and local likelihood methods such as local logistic regression (local likelihood methods are discussed, for example, in section 6.5 of Elements of Statistical Learning by Hastie et. al.).

I would prefer a C++ or Python implementation, but pointers to R (where I know that LOESS is implemented, but I can't find a local likelihood method) or Java would also suffice.

Upvotes: 4

Views: 1584

Answers (3)

David R
David R

Reputation: 1044

Can't you essentially get a python version of LOESS by using sk.learn's RadiusNeighborsRegressor function and specifying a custom tricube weighting function for the weights parameter?

Upvotes: 1

IRTFM
IRTFM

Reputation: 263332

In R there are the 'locfit' and 'mgcv' packages that I would suggest do forms of local regression. I believe the locfit package is simply a syntactic bridge to an underlying C package. (But not C++.)

Upvotes: 3

alexgolec
alexgolec

Reputation: 28252

There's an implementation in github. Given that it's pure python, I'd suggest using pypy as an interpreter.

https://github.com/ali01/loess.py

Also, I would recommend pandas as a general purpose regression tool. I'm not sure it implements LOESS and I'm not familiar enough with the algorithm to suggest an alternative, but pandas is wonderful nonetheless:

http://pandas.pydata.org/

Upvotes: 2

Related Questions