On 2012-03-07, at 8:45 AM, Olivier Grisel wrote:
> I think was DWF calls sparse coding is the LASSO implemented with
> coordinate descent (sparse coding with a fixed dictionary).
Indeed, the encoding step rather than the dictionary learning step. When
benchmarking unsupervised feature learning m
> I guess Alex and I can mentor.
approved
Alex
--
Virtualization & Cloud Management Using Capacity Planning
Cloud computing makes use of virtualization - but cloud computing
also focuses on allowing computing to be deli
On Wed, Mar 07, 2012 at 03:11:09PM +0100, Olivier Grisel wrote:
> > Obviously doing this right is quite a lot of work. I think that my group
> > could invest some efforts in this direction. We were starting to discuss
> > this a bit.
> Multinomial logistic regression with a tweaked coordinate des
2012/3/6 Gael Varoquaux :
> Yes, I was about to answer the same thing: SGD is great when n_samples >
> n_features, but the situation n_samples << n_features also exists.
>
> In such situation, I believe that a cyclic coordinate descent with a
> clever way of choosing the coordinates is the fastest
> We could indeed merge both proposals in a single GSoC proposal, but I
> would like to keep them as 2 separate steps with the two examples:
> group lens movie recommendation and out-of-core NMF for topic modeling
> on wikipedia text.
>
> We could also work on making the MiniBatchSparseDirectionary
2012/3/6 Mathieu Blondel :
> Even if they would be useful, I'd rather avoid projects like
> "maintenance" or "speed things up". I think projects with a
> well-identified goal are more likely to be accepted by the PSF.
>
> I like Olivier's proposals for SGD-based low-rank and non-negative
> matrix f
2012/3/6 Alexandre Gramfort :
>> a) sparse coding is about 2 orders of magnitude slower than competing
>> implementations right now, making it kind of useless except in toy
>> 1996-sized situations (I'm supposed to find a way to benchmark
>> this for Alex, but I can tell you that the situatio
> I love that :)
> Then I can finally put my MLP code somewhere ;)
give it a start then.
the convention should be that the gist contains 1 file with an "if
__name__ == '__main__':"
that contains an example that people can try. It can/should also
contain tests function that start with test_
Alex
On 03/07/2012 09:16 AM, Alexandre Gramfort wrote:
>> For n_features> n_samples, I believe that coordinate descent is
>> faster in the dual. A primal coordinate descent needs to optimize one
>> w_i at a time. Therefore, if your data is high dimensional it can take
>> time. Liblinear implements shri
Hi David,
There is a pull request that improves a lot the HMM implementation and
documentation:
https://github.com/scikit-learn/scikit-learn/pull/538
It should be merged anytime. Actually, I almost feel like merging it
_now_ as it does improve the situation quite a lot.
Additional help on docume
Hi,
As part of my masters project, I'm going to be working pretty heavily with
numpy/scipy and machine learning techniques like SVM and HMM. I noticed
that the current HMM implementation lacks docs and has some stability
issues and might be removed shortly. I'm interested in trying to revive it,
b
> For n_features > n_samples, I believe that coordinate descent is
> faster in the dual. A primal coordinate descent needs to optimize one
> w_i at a time. Therefore, if your data is high dimensional it can take
> time. Liblinear implements shrinking to avoid revisiting some
> coordinates. Maybe th
12 matches
Mail list logo
