Re: [Scikit-learn-general] Defining a custom correlation kernel for GaussianProcess in the form K(x, x')

On Wed, Dec 11, 2013 at 2:42 AM, Ralf Gunter wrote: > > Mathieu: I did not understand your comment about the gist being > "essentially kernel ridge." Do you mean that KRR is a simpler special > case of GP? In what way does that gist differ from the implementation > in your KRR branch? > I wanted

Re: [Scikit-learn-general] HMM contribution

Hi Matt, Thanks a lot for the offer. It's heart warming to hear people proposing improvements. Unfortunately, across the past years, we have not been able to make the HMM functionality respect our own guidelines. The core problem is that HMMs are intrinsically a structured data learning task, tha

Re: [Scikit-learn-general] Contributing ensemble selection

>> Yes, you should probably start by reading the source code of related >> ensemble models such as Adaboost and Bagging models (and maybe GBRT >> and forest models but they are more tree related hence less decoupled >> from the base estimator). Thanks a lot for the advice ! On Tue, Dec 10, 2013

[Scikit-learn-general] HMM contribution

I tried using the existing HMM module for a project, but it didn't work because it used too much memory. I implemented my own module (it currently just has a Multinomial HMM) that is memory-efficient and leverages numpy for fast vector-based operations. I'd like to contribute this module, but I

Re: [Scikit-learn-general] Contributing ensemble selection

2013/12/10 magellane a : > >>>We are still missing a stacking ensemble meta-estimator: >>>http://www.machine-learning.martinsewell.com/ensembles/stacking/Wolpert1992.pdf >>> (2748 citations) > > I would be glad to work on this, beside the guidelines here > (http://scikit-learn.org/stable/developers

Re: [Scikit-learn-general] Contributing ensemble selection

>>We are still missing a stacking ensemble meta-estimator: >>http://www.machine-learning.martinsewell.com/ensembles/stacking/Wolpert1992.pdf >> (2748 citations) I would be glad to work on this, beside the guidelines h

Re: [Scikit-learn-general] Defining a custom correlation kernel for GaussianProcess in the form K(x, x')

Thanks for all the responses! Indeed, taking a precomputed kernel matrix would be extremely useful to us since some of the terms in the kernel expression take some time to generate for the whole original training set. Unfortunately I don't know if GPs could actually work like that -- do they need

Re: [Scikit-learn-general] Defining a custom correlation kernel for GaussianProcess in the form K(x, x')

Hi Ralf, Peter, sklearn's implementation of GPML already enables custom correlation function as far as they are callables with signature matching corr(theta, dx). I can't see anything unusual in the arguments of the sklearn correlation function. GPML theory (or at least what the GaussianProcess e

Re: [Scikit-learn-general] Suggested technique for 1 D clustering

This post also relates to the postregarding using KMeans for large 1d data I asked today on the Mailing List. I just read this answer on Cross Validated, which suggests using Kernel Densi

Re: [Scikit-learn-general] Defining a custom correlation kernel for GaussianProcess in the form K(x, x')

Hi Ralf, You might find the following gist useful: https://gist.github.com/mblondel/6230778 It should support the usual kernel="precomputed". However, my implementation doesn't support the usual features that one would expect from a GP implementation such as kernel parameter tuning or ARD. In fac

Re: [Scikit-learn-general] Defining a custom correlation kernel for GaussianProcess in the form K(x, x')

Hi Ralf, unfortunately, I cannot answer your question but it would be indeed very valuabe to allow custom correlation functions. best, Peter 2013/12/9 Ralf Gunter > Hi all, > > We're trying to use a custom correlation kernel with GP in the usual > form K(x, x'). However, by looking at the bu