Re: [Scikit-learn-general] Finding conditional Gaussian Mixture Model using scikit-learn.mixture.GMM

2015-02-02 Thread Sturla Molden
Leon Palafox wrote: > If you want to stick to python, you could use pyMC to do that. emcee is also very good if you need MCMC, not to mention easier to use than PyMC. You just have to provide a callback that computes the loglikelihood. http://dan.iel.fm/emcee/current/ Sturla

Re: [Scikit-learn-general] Finding conditional Gaussian Mixture Model using scikit-learn.mixture.GMM

2015-02-02 Thread Leon Palafox
If you want to stick to python, you could use pyMC to do that. On Mon, Feb 2, 2015 at 10:02 AM, Tom Groves wrote: > On 02/02/2015, Alexander Fabisch > wrote: > > I implemented that > > in another library (https://github.com/AlexanderFabisch/gmr, example: > > > https://github.com/AlexanderFabi

Re: [Scikit-learn-general] Finding conditional Gaussian Mixture Model using scikit-learn.mixture.GMM

2015-02-02 Thread Tom Groves
On 02/02/2015, Alexander Fabisch wrote: > I implemented that > in another library (https://github.com/AlexanderFabisch/gmr, example: > https://github.com/AlexanderFabisch/gmr/blob/master/examples/plot_estimate_gmm.py). I tried out your code, but it crashed when attempting to generate samples from

Re: [Scikit-learn-general] Finding conditional Gaussian Mixture Model using scikit-learn.mixture.GMM

2015-02-02 Thread Tom Groves
Wow, fast responses, thanks! On 02/02/2015, Andy wrote: > I don't see how this would fit into the standard sklearn interface... Just what Alexander said. I don't know sklearn well enough to know what the standard pattern for interfacing might be, or why this presents a problem. I was expecting t

Re: [Scikit-learn-general] Finding conditional Gaussian Mixture Model using scikit-learn.mixture.GMM

2015-02-02 Thread Andy
I don't see how this would fit into the standard sklearn interface... From a probabilistic modelling perspective that is indeed quite basic, but sklearn is not a probabilistic modelling framework ;) -- Dive into the Worl

Re: [Scikit-learn-general] Classifier that is perfectly stable given shuffled training data

2015-02-02 Thread Andy
Hi Juan. For up to floating point precision, that is pretty hard as Gael mentioned. 1e-5 on sigma seems pretty low, though. Can you post data to reproduce? I would expect most classifiers to go to around 1e-8. Cheers, Andreas On 02/02/2015 10:46 AM, Juan Nunez-Iglesias wrote: Hi all, *TL;DR

Re: [Scikit-learn-general] Finding conditional Gaussian Mixture Model using scikit-learn.mixture.GMM

2015-02-02 Thread Alexander Fabisch
Hi Tom, as far as I know this is not implemented in sklearn. I implemented that in another library (https://github.com/AlexanderFabisch/gmr, example: https://github.com/AlexanderFabisch/gmr/blob/master/examples/plot_estimate_gmm.py). The implementation is very simple and not as efficient and usabl

[Scikit-learn-general] Finding conditional Gaussian Mixture Model using scikit-learn.mixture.GMM

2015-02-02 Thread Tom Groves
I'm using scikit-learn to fit a multivariate Gaussian Mixture Model to some data (which works brilliantly). But I need to be able to get a new GMM conditional on some of the variables, and the scikit toolkit doesn't seem to be able to do that, which surprised me because it seems like a pretty basic

Re: [Scikit-learn-general] Classifier that is perfectly stable given shuffled training data

2015-02-02 Thread Gael Varoquaux
Thinking about it, that's going to be hard: even floating points operations such as a sum of many floating point numbers is not permutation invariant, due to the rounding errors. Gaël -- Dive into the World of Parallel Pr

[Scikit-learn-general] Classifier that is perfectly stable given shuffled training data

2015-02-02 Thread Juan Nunez-Iglesias
Hi all, TL;DR version: I'm looking for a classifier that will get the *exact same model* for shuffled versions of the training data. I thought GaussianNB would do the trick but either I don't understand it, or some kind of numerical instability prevents it from achieving the same model on subs