Hi Scikit-Learn
I just wanted to follow up with contributing an EM algorithm for Bernoulli
Mixture models. We can prepare a pull request so that you have something
to look at. If that makes this any easier.
Cheers,
Mark
---------- Forwarded message ----------
From: Mark Stoehr <[email protected]>
Date: Thu, Aug 28, 2014 at 1:18 PM
Subject: Re: [Scikit-learn-general] Multivariate Bernoulli EM
To: [email protected]
Cc: Gustav Larsson <[email protected]>
Hi Kyle,
We use binary features for computer vision and speech recognition. The
Bernoulli mixture model is the basic workhorse for performing
classification and learning mid-level features. We've done preliminary
work on using it to build multi-layered networks. It's been used by others
in a de-noising context (such as in Bishop's book), and its a good baseline
for comparison for recent work in deep learning using tractable probability
estimators (such as Pedro Domingos' work).
These are some papers using it:
Daniel Lowd, Pedro Domingos. "Naive Bayes Models for Probability
Estimation", ICML, 2005
http://machinelearning.wustl.edu/mlpapers/paper_files/icml2005_LowdD05.pdf
It was used by Pedro Domingos in this later paper as a baseline:
http://arxiv.org/pdf/1405.0501.pdf
Mathias Niepert, Pedro Domingos
"Exchangeable Variable Models", ICML, 2005
Joan Bruna, Stephane Mallat "Geometric Models with Co-occurrence Groups", ESANN
2010,
http://www.cmapx.polytechnique.fr/~bruna/Publications_files/esann_v6.pdf
A couple more recent papers
http://www.sciencedirect.com/science/article/pii/S0925231213004578#
http://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=6460335
On Thu, Aug 28, 2014 at 2:57 AM, Kyle Kastner <[email protected]> wrote:
> This sounds interesting - what do you normally use it for? Do you have
> any references for papers to look at?
>
> On Wed, Aug 27, 2014 at 12:44 PM, Mark Stoehr <[email protected]>
> wrote:
> > Hi scikit-learn,
> >
> > Myself and a colleague put together an implementation of the EM algorithm
> > for mixtures of multivariate Bernoulli vectors (the algorithm in Chris
> > Bishop's "Pattern Recognition") which is useful for classification and
> > clustering of binary data. We have based the implementation on the
> style of
> > the sklearn.mixture.GMM and subclassed it from the BaseEstimator class.
> It
> > probably needs a few more tweaks before it would be ready for general
> use,
> > but we just wanted to hear from people whether this would be a good
> addition
> > to scikit-learn.
> >
> > Cheers,
> > Mark Stoehr
> > Gustav Larsson
> >
> >
> >
> >
> ------------------------------------------------------------------------------
> > Slashdot TV.
> > Video for Nerds. Stuff that matters.
> > http://tv.slashdot.org/
> > _______________________________________________
> > Scikit-learn-general mailing list
> > [email protected]
> > https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
> >
>
>
> ------------------------------------------------------------------------------
> Slashdot TV.
> Video for Nerds. Stuff that matters.
> http://tv.slashdot.org/
> _______________________________________________
> Scikit-learn-general mailing list
> [email protected]
> https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
>
------------------------------------------------------------------------------
Slashdot TV.
Video for Nerds. Stuff that matters.
http://tv.slashdot.org/
_______________________________________________
Scikit-learn-general mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
