2012/2/20 David Warde-Farley :
> On 2012-02-20, at 3:46 AM, Olivier Grisel wrote:
>
>> How high dimensional is this? GraphLasso works on the empirical
>> covariance matrix which is implemented as an 2D numpy array with shape
>> (n_features, n_features). It won't fit in memory for n_features >
>> 10
On 2012-02-20, at 3:46 AM, Olivier Grisel wrote:
> How high dimensional is this? GraphLasso works on the empirical
> covariance matrix which is implemented as an 2D numpy array with shape
> (n_features, n_features). It won't fit in memory for n_features >
> 1 and it might be intractably too lo
Thanks! And does it make sense to use L1 regularisation here (irrespective
of the graph structure)?
Best,
Mathias
On Mon, Feb 20, 2012 at 11:06 AM, Gael Varoquaux <
[email protected]> wrote:
> On Mon, Feb 20, 2012 at 10:35:51AM +0100, Mathias Verbeke wrote:
> > I would have around
On Mon, Feb 20, 2012 at 10:35:51AM +0100, Mathias Verbeke wrote:
> I would have around 1 features. I'm working on a sentence
> classification problem,
Graph lasso won't work on such a problem.
> I would like to do feature selection, to reduce the number of
> dimensions, and was thinking to ta
Hi Olivier,
Thanks for the fast reply!
> I have a high-dimensional feature set, where the features originate from
> > graphs. I was wondering if the use of GraphLasso applies and would be a
> good
> > idea in this case? And if it would be, can I then just apply it on the
> > feature vectors or do
2012/2/20 Mathias Verbeke :
> Hi all,
>
> I have a high-dimensional feature set, where the features originate from
> graphs. I was wondering if the use of GraphLasso applies and would be a good
> idea in this case? And if it would be, can I then just apply it on the
> feature vectors or do I need t
Hi all,
I have a high-dimensional feature set, where the features originate from
graphs. I was wondering if the use of GraphLasso applies and would be a
good idea in this case? And if it would be, can I then just apply it on the
feature vectors or do I need to input the originating graph structure
On Sat, Nov 12, 2011 at 12:28 AM, Gael Varoquaux
wrote:
> I don't like the results as much. Basically, we can either cluster on
> conditional relations, or on marginal relations: basically, as an input
> of affinity propagation, we can use the correlation matrix, which is a
> standard affinity me
On Sat, Nov 12, 2011 at 12:33 AM, Olivier Grisel
wrote:
> Indeed I thought the same, but as the current result is already good /
> interesting ...
We can see that Yahoo is clustered with Amazon and Apple, although it
would have been clustered with Cisco, Dell and IBM if the clustering
had been m
2011/11/11 Mathieu Blondel :
> GraphLasso seems really neat (and the associated CV object should
> prove very useful).
>
> I had a look at the stock market example but I am a bit confused by
> the fact that the clustering, graph structure and 2d-embedding seemed
> to be learned independently althou
On Sat, Nov 12, 2011 at 12:19:37AM +0900, Mathieu Blondel wrote:
> GraphLasso seems really neat (and the associated CV object should
> prove very useful).
Thanks!
> I see that the (dense) correlation matrix is used as input to affinity
> propagation. Wouldn't it be better if we used the partial c
GraphLasso seems really neat (and the associated CV object should
prove very useful).
I had a look at the stock market example but I am a bit confused by
the fact that the clustering, graph structure and 2d-embedding seemed
to be learned independently although they are clearly related
problems. I
On Thu, Nov 10, 2011 at 12:06:15AM +0100, Gael Varoquaux wrote:
> On Wed, Nov 09, 2011 at 11:43:40PM +0100, bthirion wrote:
> > > What do people think? Should I:
> > > 1. change graph_lasso to take the empirical covariance as an input
> > > 2. add an 'X_is_cov' parameter to the estimators
> >
On Wed, Nov 09, 2011 at 09:10:46PM -0500, Alexandre Gramfort wrote:
> to complexify a bit the pb note that in the SVM/Lasso/... case the
> precomputed gram
> is np.dot(X, X.T) which means that the cross-val can be done just with it
> [...]
OK, how about that: I try to solve only the situation for
indeed this is more correct...
> - data / design matrix, shape (n_samples, n_features)
> - kernel / Gram / similarity / affinity / connectivity, shape
> (n_samples, n_samples)
> - distance, shape (n_samples, n_samples) (same shape as kernel but
> opposite semantics)
> - covariance, shape (n_featur
2011/11/9 Alexandre Gramfort :
> to complexify a bit the pb note that in the SVM/Lasso/... case the
> precomputed gram
> is np.dot(X, X.T) which means that the cross-val can be done just with it
> while for the covariance estimation, like GraphLassoCV, the empirical
> covariance is np.dot(X.T, X) h
to complexify a bit the pb note that in the SVM/Lasso/... case the
precomputed gram
is np.dot(X, X.T) which means that the cross-val can be done just with it
while for the covariance estimation, like GraphLassoCV, the empirical
covariance is np.dot(X.T, X) hence the fit needs X as input.
so it see
On Wed, Nov 09, 2011 at 11:43:40PM +0100, bthirion wrote:
> > What do people think? Should I:
> > 1. change graph_lasso to take the empirical covariance as an input
> > 2. add an 'X_is_cov' parameter to the estimators
> +1 for the second one.
I actually was suggesting both, and 1 as a mean f
> What do people think? Should I:
>
> 1. change graph_lasso to take the empirical covariance as an input
>
> 2. add an 'X_is_cov' parameter to the estimators
+1 for the second one.
If we want to introduce some kind of automated guess of the
regularization parameter, we'll have to know the dim
On Wed, Nov 9, 2011 at 12:20 PM, Virgile Fritsch
wrote:
> Reminds me of the PR by Robert about performing clustering from similarity
> matrix or directly from the data.
> So I would be in favour of having a X_is_cov keyword.
>
> Sorry for biasing the discussion with cov_init, I answered to quikly
Reminds me of the PR by Robert about performing clustering from similarity
matrix or directly from the data.
So I would be in favour of having a X_is_cov keyword.
Sorry for biasing the discussion with cov_init, I answered to quikly ;)
On Wed, Nov 9, 2011 at 5:16 PM, Gael Varoquaux <
gael.varoqu..
On Wed, Nov 09, 2011 at 10:05:53AM -0500, [email protected] wrote:
> graph_lasso(X,) takes the data array as an argument, but except
> calculating the empirical_covariance at the beginning X is not used
> anymore, as far as I could see.
> The algorithm looks very interesting, but I would ha
On Wed, Nov 9, 2011 at 10:59 AM, Lars Buitinck wrote:
> 2011/11/9 Virgile Fritsch :
>> Did you notice the `cov_init` parameter?, or maybe it was added after your
>> comment?
>
> OOPS, sorry.
In my reading of the code cov_init is just the starting matrix, the
updating is still based on emp_cov.
J
2011/11/9 Virgile Fritsch :
> Did you notice the `cov_init` parameter?, or maybe it was added after your
> comment?
OOPS, sorry.
--
Lars Buitinck
Scientific programmer, ILPS
University of Amsterdam
--
RSA(R) Conference
Did you notice the `cov_init` parameter?, or maybe it was added after your
comment?
On Wed, Nov 9, 2011 at 4:13 PM, Lars Buitinck wrote:
> 2011/11/9 :
> > The graph looks very good, good advertising.
>
> Yep, pretty picture ;)
>
> > graph_lasso(X,) takes the data array as an argument, but
2011/11/9 :
> The graph looks very good, good advertising.
Yep, pretty picture ;)
> graph_lasso(X,) takes the data array as an argument, but except
> calculating the empirical_covariance at the beginning X is not used
> anymore, as far as I could see.
>
> The algorithm looks very interesting
On Wed, Nov 9, 2011 at 6:21 AM, Gael Varoquaux
wrote:
> Hi list,
>
> I'd like to ask for comments on the GraphLasso pull request that I have
> put in. I think that it is ready for merge, even though it has been in
> development for a short amount of time, because I have been working on
> similar a
Hi list,
I'd like to ask for comments on the GraphLasso pull request that I have
put in. I think that it is ready for merge, even though it has been in
development for a short amount of time, because I have been working on
similar algorithms for more than two years.
To give you a run-through, and
28 matches
Mail list logo
