On Thu, Nov 29, 2012 at 10:39 AM, Afik Cohen wrote:
>
> It's easy to see how with some slight modifications (wrapping that in a
> joblib
> Parallel() call) we could enable n_jobs for OneVsRestClassifier. This
> almost
> seems too simple, so there must be a good reason why this isn't done; could
>
Hi all,
We've been looking at ways to parallelize our classifier training, and we
looked at the n_jobs parameter as a possible way to do that. The classifier
we're currently using, SGDClassifier, supports that parameter, but since we're
using a OneVsRest (Ovr) strategy, our call is wrapped in a O
Dear, sklearn community,
1. The source code of the Block SBL algorithm is now available at bitbucket:
https://bitbucket.org/liubenyuan/pybsbl
any suggestion, optimization and test on the code are all welcome! as well
as your success stories on applying our methods.
2. Block-OMP is an extension to
Hi Liu,
This work is really nice and very fancy, but it is also very recent and
needs a bit more insight and benchmarking before it can enter
scikit-learn: we have a rule not to integrate any new approach that is
more than 2 years old. The reason is that if the approach is to be a
massive success,
> Remember that one-vs-rest trains each base classifier *independently* (a new
sample may play the role of a positive or negative example, depending on the
class). Therefore, you do need to update all base classifiers.HTH,Mathieu
>
>
>
Ah, so every sample is trained on every classifier? It
On Wed, Nov 28, 2012 at 5:02 PM, Peter Prettenhofer <
[email protected]> wrote:
>
> if the input data is float64 you need to take conversion to float32
> into account; furthermore sklearn will convert to fortran layout ->
> this will give a huge penalty in memory consumption.
>
In my e
> scikit-learn is really bad when n_jobs=10.
To avoid the memory copy you can try my branch of joblib:
https://github.com/joblib/joblib/pull/44
You need to hack the X_argsorted generation code to generate a memmap
array instead of a numpy array (I am planning to add a helper in
joblib to make th
2012/11/28 Mathieu Blondel :
> scikit-learn's RF is entirely written in Python (forest.py) so there may
> still be some slow code paths. Moreover, their parallel implementation is
> probably written with pthreads or OpenMP so they bypass the problems that we
> have with Python's multiprocessing mod
On Thu, Nov 29, 2012 at 12:50 AM, Andreas Mueller
wrote:
> Why should C++ be any faster than Cython?
> Templating number of bins in leafs?
>
scikit-learn's RF is entirely written in Python (forest.py) so there may
still be some slow code paths. Moreover, their parallel implementation is
probably
2012/11/28 Andreas Mueller :
> Am 28.11.2012 16:46, schrieb Mathieu Blondel:
>
>
>
> On Thu, Nov 29, 2012 at 12:33 AM, Andreas Mueller
> wrote:
>>
>> Do you see where the "sometimes 100x" comes from?
>> Not from what he demonstrates, right?
>>
> scikit-learn is really bad when n_jobs=10. I would b
Am 28.11.2012 16:46, schrieb Mathieu Blondel:
On Thu, Nov 29, 2012 at 12:33 AM, Andreas Mueller
mailto:[email protected]>> wrote:
Do you see where the "sometimes 100x" comes from?
Not from what he demonstrates, right?
scikit-learn is really bad when n_jobs=10. I would be inte
On Thu, Nov 29, 2012 at 12:33 AM, Andreas Mueller
wrote:
> Do you see where the "sometimes 100x" comes from?
> Not from what he demonstrates, right?
>
> scikit-learn is really bad when n_jobs=10. I would be interested in
knowing if the performance gains are mostly coming from the fact that
wiseRF
Nope they don't...
On 28 November 2012 16:39, Andreas Mueller wrote:
> Am 28.11.2012 16:33, schrieb Gilles Louppe:
>> Do they use the same value for the min_samples_split parameter? I see
>> they use a default value (hidden in their constructor I guess), but
>> theirs might not be the same as our
Am 28.11.2012 16:33, schrieb Gilles Louppe:
> Do they use the same value for the min_samples_split parameter? I see
> they use a default value (hidden in their constructor I guess), but
> theirs might not be the same as ours.
>
They don't even give the depth, do they?
-
Do they use the same value for the min_samples_split parameter? I see
they use a default value (hidden in their constructor I guess), but
theirs might not be the same as ours.
Gilles
On 28 November 2012 16:29, Andreas Mueller wrote:
> Am 28.11.2012 16:19, schrieb Peter Prettenhofer:
>> Some more
Am 28.11.2012 16:19, schrieb Peter Prettenhofer:
> Some more benchmarks from wise.io:
>
> http://continuum.io/blog/wiserf-use-cases-and-benchmarks
>
> quite impressive indeed - unfortunately I cannot post any comments on
> the blog - I wonder if they use some sort of binned split evaluation
> [1] i
Am 28.11.2012 16:19, schrieb Peter Prettenhofer:
> Some more benchmarks from wise.io:
>
> http://continuum.io/blog/wiserf-use-cases-and-benchmarks
>
> quite impressive indeed - unfortunately I cannot post any comments on
> the blog - I wonder if they use some sort of binned split evaluation
> [1] i
Some more benchmarks from wise.io:
http://continuum.io/blog/wiserf-use-cases-and-benchmarks
quite impressive indeed - unfortunately I cannot post any comments on
the blog - I wonder if they use some sort of binned split evaluation
[1] instead of exact split evaluation (wiseRF has slightly lower
a
Am 28.11.2012 12:00, schrieb Olivier Grisel:
> Also the pending patent of BSBL-BO applied to EEG decoding makes me a
> lot less interested in working on maintaining an open-source version
> of such method knowing that it could not be used without licensing the
> patent in the U.S.
>
> http://techtr
Also the pending patent of BSBL-BO applied to EEG decoding makes me a
lot less interested in working on maintaining an open-source version
of such method knowing that it could not be used without licensing the
patent in the U.S.
http://techtransfer.universityofcalifornia.edu/NCD/22688.html
--
2012/11/28 :
> Dear scikit-learn community:
>
> Block Sparse Bayesian Learning is a powerful CS algorithm for recovering
> block sparse signals with structures, and shows the additional benefits of
> reconstruct non-sparse signals, see Dr. zhilin zhang's websites:
> http://dsp.ucsd.edu/~zhilin/BSB
It looks like a bug. Can you please open a github issue (including
your code snippet + the links to the predictions)?
It's weird that this issue cannot be seen here:
http://scikit-learn.org/dev/auto_examples/plot_roc_crossval.html#example-plot-roc-crossval-py
--
Olivier
Hi,
There is the orthogonal matching pursuit algorithm (an another CS
algorithm) in scikit-learn which is classified as a regression model.
The notations are a bit different from the CS community.
Arnaud Joly
Le 28/11/2012 08:38, [email protected] a écrit :
Hi Meng:
It is a compressive se
Dear Liu Benuyan.
Thank you for offering to contribute to scikit-learn.
I am no expert in sparse signal recovery and/or matrix factorization,
so I can not really comment on the method.
I just wanted to mention that we include mostly widely-used or classical
algorithms.
I am not sure how far this
24 matches
Mail list logo
