Hi,
I think some of the algorithms already offer this (Laplacian Eigenmaps for
instance).
I'm -1 for LLE as LLE does not compute distances, but weights based on the
points directly.
Matthieu
2011/9/21 Jacob VanderPlas
> Hello,
> I recently was contacted by someone interested in using manifold
On Wed, Sep 21, 2011 at 8:56 AM, Robert Layton wrote:
> I do really like the metric='precomputed' concept, which allows both
> implementing actual metrics (euclidean, manhattan) as well as passing a
> precomputed array in. If the algorithm doesn't allow it for whatever reason*
> throw an error. T
Hi Satra,
On Tue, Sep 20, 2011 at 01:14:46AM -0400, Satrajit Ghosh wrote:
> exchange between me and a student on current documentation. i don't know
> what level you are expecting people to be at. this is somebody who is using
> sklearn and understands cross-validation in principle (a 2nd year
> u
On 21 September 2011 09:50, Gael Varoquaux wrote:
> On Tue, Sep 20, 2011 at 04:25:58PM -0700, Jacob VanderPlas wrote:
> > I recently was contacted by someone interested in using manifold
> > learning methods on abstract metric spaces: that is, the training data
> > is a matrix of pairwise distance
On Tue, Sep 20, 2011 at 04:25:58PM -0700, Jacob VanderPlas wrote:
> I recently was contacted by someone interested in using manifold
> learning methods on abstract metric spaces: that is, the training data
> is a matrix of pairwise distances rather than a set of points. It would
> be fairly str
On 21 September 2011 09:25, Jacob VanderPlas <
vanderp...@astro.washington.edu> wrote:
> Hello,
> I recently was contacted by someone interested in using manifold
> learning methods on abstract metric spaces: that is, the training data
> is a matrix of pairwise distances rather than a set of point
Hello,
I recently was contacted by someone interested in using manifold
learning methods on abstract metric spaces: that is, the training data
is a matrix of pairwise distances rather than a set of points. It would
be fairly straightforward to implement this for basic LLE and Isomap,
and could
2011/9/20 Gael Varoquaux :
> On Tue, Sep 20, 2011 at 11:20:03AM +0200, Olivier Grisel wrote:
>> > Look at it this way: this is an opportunity to teach people how to
>> > properly structure their scripts for properly for use with joblib on
>> > all platforms. :-)
>
> Yes, I agree with this. I'll add
On Tue, Sep 20, 2011 at 04:35:36PM +0200, Lars Buitinck wrote:
> There's also a minor efficiency issue where as_float_array is called
> on the result of atleast_2d, which might incur two copies. We may want
> to rethink this, maybe add a safe_ replacement for atleast_2d with a
> boolean copy argume
2011/9/20 Olivier Grisel :
> 2011/9/20 Fabian Pedregosa :
>
> I find the opposite:
>
> $ git grep "\bcopy=" | wc -l
> 72
> $ git grep "\boverwrite_X=" | wc -l
> 44
Well, there's also a few overwrite_Gram instances.
> I prefer the copy={True,False} naming convention.
It sounds friendlier, doesn't
On Tue, Sep 20, 2011 at 04:27:58PM +0200, Fabian Pedregosa wrote:
> So it's OK to change it. However, a quick grep shows that there are
> more overwrite_? uses than copy_? in sklearn.
That's however contrary to conventions in numpy. My personnal opinion is
that when there is no ambiguity on what i
2011/9/20 Fabian Pedregosa :
> On Tue, Sep 20, 2011 at 4:10 PM, Lars Buitinck wrote:
>> Hi all,
>>
>> I only just spotted sklearn.utils.as_float_array. I might have missed
>> a discussion about this, but I notice it and its clients use an
>> overwrite_Whatever convention, while older modules commo
On Tue, Sep 20, 2011 at 11:20:03AM +0200, Olivier Grisel wrote:
> > Look at it this way: this is an opportunity to teach people how to
> > properly structure their scripts for properly for use with joblib on
> > all platforms. :-)
Yes, I agree with this. I'll add a statement in the docs of joblib
On Tue, Sep 20, 2011 at 4:10 PM, Lars Buitinck wrote:
> Hi all,
>
> I only just spotted sklearn.utils.as_float_array. I might have missed
> a discussion about this, but I notice it and its clients use an
> overwrite_Whatever convention, while older modules commonly use copy
> and copy_Whatever. Wh
On Tue, Sep 20, 2011 at 04:09:54PM +0200, Lars Buitinck wrote:
> I only just spotted sklearn.utils.as_float_array. I might have missed
> a discussion about this, but I notice it and its clients use an
> overwrite_Whatever convention, while older modules commonly use copy
> and copy_Whatever. When h
Hi all,
I only just spotted sklearn.utils.as_float_array. I might have missed
a discussion about this, but I notice it and its clients use an
overwrite_Whatever convention, while older modules commonly use copy
and copy_Whatever. When has as_float_array been introduced (wrt.
releases) and can we c
Yes, the problem is gone! Thanks, and the discussion is great.
On Tue, Sep 20, 2011 at 5:20 PM, Olivier Grisel wrote:
> 2011/9/20 Robert Kern :
> > On Tue, Sep 20, 2011 at 02:49, Olivier Grisel
> wrote:
> >> 2011/9/20 Robert Kern :
> >
> >>> The fix is straightforward on sklearn's part: move the
2011/9/20 Robert Kern :
> On Tue, Sep 20, 2011 at 02:49, Olivier Grisel
> wrote:
>> 2011/9/20 Robert Kern :
>
>>> The fix is straightforward on sklearn's part: move the code into a
>>> function and call that function under an "if __name__ == '__main__':"
>>> test. Or just move everything under th
2011/9/20 Robert Kern :
>
> I would additionally suggest that the test-running script also be
> fixed to run correctly under Windows.
Vlad, how you running the tests on windows? Do you experiment the
"infinite dot" issue using nosetests?
Or using the command:
python -c "import sklearn; sklearn.
On Tue, Sep 20, 2011 at 03:17, wrote:
>>But then people will just copy the file and modify it to set n_jobs to
> be greater, and then ask here why it doesn't work on Windows. It's a
> teaching moment. Embrace it!
>
> Yes and no. If I download sklearn on windows and the run make, it doesn't
> fin
>But then people will just copy the file and modify it to set n_jobs to
be greater, and then ask here why it doesn't work on Windows. It's a
teaching moment. Embrace it!
Yes and no. If I download sklearn on windows and the run make, it doesn't
finish the tests. To then tell me that I need to mod
On Tue, Sep 20, 2011 at 02:49, Olivier Grisel wrote:
> 2011/9/20 Robert Kern :
>> The fix is straightforward on sklearn's part: move the code into a
>> function and call that function under an "if __name__ == '__main__':"
>> test. Or just move everything under that __main__ test.
>
> Indeed I kee
Also in test suites, it restarts the suite in every process and you
get infinite dots.
On Tue, Sep 20, 2011 at 10:49 AM, Olivier Grisel
wrote:
> 2011/9/20 Robert Kern :
>> On Mon, Sep 19, 2011 at 20:00, xinfan meng wrote:
>>> Hi:
>>> I ran the sklearn sentiment classification codes
>>> (http
2011/9/20 Robert Kern :
> On Mon, Sep 19, 2011 at 20:00, xinfan meng wrote:
>> Hi:
>> I ran the sklearn sentiment classification codes
>> (https://github.com/scikit-learn/scikit-learn-tutorial/blob/master/solutions/exercise_02_sentiment.py)
>> and found it keep creating new python.exe instance
On Mon, Sep 19, 2011 at 20:00, xinfan meng wrote:
> Hi:
> I ran the sklearn sentiment classification codes
> (https://github.com/scikit-learn/scikit-learn-tutorial/blob/master/solutions/exercise_02_sentiment.py)
> and found it keep creating new python.exe instance, and then my computer
> crash
One of the drawcards of using python is that it is relatively platform
independent - so windows users have a reasonable expectation that their code
will run.
Windows itself is not a problem - I've written many C++ programs using
boost::thread that work just fine on windows. Perhaps this require
2011/9/20 xinfan meng :
> Hi:
> I ran the sklearn sentiment classification codes
> (https://github.com/scikit-learn/scikit-learn-tutorial/blob/master/solutions/exercise_02_sentiment.py)
> and found it keep creating new python.exe instance, and then my computer
> crashed because ran out of memor
Hello
That is the way joblib multiprocessing behaves on Windows, as far as I
know. Gaël is aware of this, but there are not that many scikit-learn
users running windows on the mailing list.
Best,
Vlad
On Tue, Sep 20, 2011 at 4:00 AM, xinfan meng wrote:
> Hi:
> I ran the sklearn sentiment cl
28 matches
Mail list logo
