Sorry, I did not mean to put it in such a way, I meant something else. But
yes I understand your point.
Thank you,
Ronnie Ghose
On Thu, Mar 21, 2013 at 1:30 AM, Olivier Grisel wrote:
> 2013/3/20 Ronnie Ghose :
> > Yes, however participation in a GSoC style program gives it credibility
> it
> >
2013/3/20 Ronnie Ghose :
> Yes, however participation in a GSoC style program gives it credibility it
> would not otherwise have I believe... no?
I would say that you should not engage in open source contribution to
gain credibility but rather because you are interested in moving the
project forwa
Yes, however participation in a GSoC style program gives it credibility it
would not otherwise have I believe... no?
Thank you,
Shomiron Ghose
On Thu, Mar 21, 2013 at 1:15 AM, Olivier Grisel wrote:
> 2013/3/20 Ronnie Ghose :
> > Mr. Grisel,
> >
> > Anyway I could sign up for this as well, but n
2013/3/20 Ronnie Ghose :
> Mr. Grisel,
>
> Anyway I could sign up for this as well, but not formally participate as a
> GSoC student?
If you cannot qualify formally for the GSoC program you are always
free to contribute using the regular contribution workflow:
http://scikit-learn.org/stable/devel
Mr. Grisel,
Anyway I could sign up for this as well, but not formally participate as a
GSoC student? .. :( I'm 2 months too .. young and they allow no exceptions,
though I'm allowed to work in the US.
Thank you,
Shomiron Ghose
On Thu, Mar 21, 2013 at 12:31 AM, Olivier Grisel
wrote:
> scikit-le
scikit-learn is officially registered for the 2013 edition of the
Google Summer of Code under the PSF umbrella.
http://wiki.python.org/moin/SummerOfCode/2013
Prospective students and mentors please feel free to register your
name on the wiki:
https://github.com/scikit-learn/scikit-learn/wiki/A-l
2013/3/16 Olivier Grisel :
> Hi Bradley,
>
> Thanks for the heads up. I'll try to show up on time for the BoF. In
> case I don't I just want to say that we are still interested in
> joining the conservancy.
Hi Bradley,
During PyCon / PyData I had the opportunity to speak to various
NumFOCUS board
Gilles, my understanding and experience is that RFs are fairly well
calibrated, ie the probability estimates are pretty good, for "reasonable"
number of trees, e.g. 255. Paul hasn't specified how accurate he expects
his probability measurements.
Another consideration, though, is that if a test sam
I'm not on the PIL mailing list but I'll sign up and post there.
The one thing I was thinking scikit-learn could do is to give a better
message when
> imread(file_path)
in sklearn/datasets/lfw.py
returns a 0-dm array. Something to the effect of "the file wasn't
successfully read" instead
Note that you can get perfect scores (either 0.0 or 1.0) simply be
setting n_estimators=1. This is why you should use this measure with
caution.
On 20 March 2013 15:27, Lars Buitinck wrote:
> 2013/3/20
>> > I was just about to say that discarding predictions in a range .5 -
>> > epsilon < p < .5
2013/3/20
> > I was just about to say that discarding predictions in a range .5 -
> > epsilon < p < .5 + epsilon can be a useful thing to do in some cases.
>
> Where does "epsilon" come from? I don't see it on the list of parameters
It's a parameter that I just made up. If you consider >0.9 or <0
> > You should use predict_proba with caution if what you want is a level
> > of confidence with respect to the true values. If your trees are fully
> > developed, then predict_proba is rather a level of agreement between
> > the trees, no matter they are right or wrong with the true values. It
>
@ogrisel , I'd be happy to lend a hand with this. Just say yay and I'll get
it going :)
2013/3/20 Andreas Mueller
> Olivier wrote it and is planning to merge it.
> I think he wouldn't resist if any one gave him a hand ;)
>
> I think it is quite urgent as this keeps getting up on the ml/irc/so
>
Olivier wrote it and is planning to merge it.
I think he wouldn't resist if any one gave him a hand ;)
I think it is quite urgent as this keeps getting up on the ml/irc/so
Cheers,
Andy
--
Everyone hates slow websites. So
2013/3/20 Jaques Grobler
> Definitely - who made that tut originally? Would be a good addition to the
> tutorial section.
>
We did, or rather, the other guys; I was lured into the project by this
tutorial, which was quite good when it was up to data.
> Else we could just update it and then lin
Actually, I think the last update just missed the line at the bottom of the
page about version 0.9..
There's a note by the index saying
Note
This document is meant to be used with *scikit-learn version 0.11+* (i.e.
the current state of the master branch at the time of writing: 2012-02-13).
So i
Definitely - who made that tut originally? Would be a good addition to the
tutorial section.
Else we could just update it and then link to it from the tutorial section
like with Jake's AstroML tutorial..
2013/3/20 Lars Buitinck
> 2013/3/20 Jaques Grobler
>
>> same thought as Lars.. also, that
2013/3/20 Jaques Grobler
> same thought as Lars.. also, that tutorial mentions at the bottom that
> it's for scikit-learn 0.9
>
We should update that (or actually, we should merge the tutorial into the
mainline docs...)
--
Lars Buitinck
Scientific programmer, ILPS
University of Amsterdam
-
same thought as Lars.. also, that tutorial mentions at the bottom that it's
for scikit-learn 0.9
2013/3/20 Lars Buitinck
> 2013/3/19 Zvika Marx
>
>> /usr/lib/pymodules/python2.7/sklearn/linear_model/stochastic_gradient.pycin
>> _fit_multiclass(self, X, y, sample_weight) 162 strategy is called
Hi Andy,
yes I'm working on 32bit. I thought that maybe it was me who did something
wrong during the installation.
Thanks for your help.
Pamela
On Wed, Mar 20, 2013 at 12:02 PM, Andreas Mueller
wrote:
> Hi Pamela.
> Unfortunately this is a known problem. You are on 32bit, right? It is an
> in
Hi Pamela.
Unfortunately this is a known problem. You are on 32bit, right? It is an
instability in the blas in scipy.
Don't worry about it to much. We are trying to make the test more robust.
Hope that helps,
Andy
On 03/20/2013 11:59 AM, Pamela Carreño wrote:
Hi Andy,
I'm using 0.13.1 versi
Hi Andy,
I'm using 0.13.1 version. This is the output I get
.../usr/local/lib/python2.7/dist-packages/scikit_learn-0.13.1-py2.7-linux-i686.egg/sklearn/manifold/spectral_embedding.py:225:
UserWarning: Graph is not fully connected, spectra
Hi Pamela.
Could you please give us the full traceback / output of the test that is
failing?
Which version are you using? 0.13.1?
Cheers,
Andy
On 03/20/2013 11:40 AM, Pamela Carreño wrote:
Hi,
I have installed scikit-learn following the "installing an official
release" tutorial, however when
2013/3/19 Zvika Marx
> /usr/lib/pymodules/python2.7/sklearn/linear_model/stochastic_gradient.pycin
> _fit_multiclass(self, X, y, sample_weight) 162 strategy is called OVA:One
> Versus All
> . 163 """ --> 164 X = np.asarray(X, dtype=np.float64, order='C') 165 166 #
> Use joblib to run OVA in par
No, I am saying that. I am just saying that what you compute is not
really "a probability with respect to the true value". What this measure
represents highly depends on the number of trees, the size of your
data and the depth of the trees. Provided that you build an infinite
number of trees, with
Hi,
I have installed scikit-learn following the "installing an official
release" tutorial, however when I try to test it using
nosetests sklearn --exe
I get
Ran 1603 tests in 70.591s
FAILED (SKIP=11, failures=1)
Am I missing something?
--
Pamela
-
2013/3/20 Gilles Louppe :
> You should use predict_proba with caution if what you want is a level
> of confidence with respect to the true values. If your trees are fully
> developed, then predict_proba is rather a level of agreement between
> the trees, no matter they are right or wrong with the t
Hi Paul,
You should use predict_proba with caution if what you want is a level
of confidence with respect to the true values. If your trees are fully
developed, then predict_proba is rather a level of agreement between
the trees, no matter they are right or wrong with the true values. It
is likely
It would perhaps be good to add these, atleast point 1 and 3, as 2 is still
a bit confusing, to the development docs. I can do this if you not on it
already, Vlad
2013/3/19 Vlad Niculae
> >> II. Sometimes if attribute descriptions have multiple lines, a backtick
> >> is needed at the end of con
Thanks for reporting.. This is perhaps more a matter for the PIL mailing
list? http://www.pythonware.com/products/pil/ (See Free Support section)
Unless, people know of any way we can help solve it on this side?
2013/3/19 Patrick Flaherty
>
> I'm experimenting with the examples/tutorials t
I need the random matrix to evaluate the predictiveness of a particular
model - this time, not in terms of a "domain of applicability" :)
Just to clarify the question/put us on the same line: please see the
attached PDF for the question I'm having in mind:
Cheers & Thanks,
Paul
>
> What do y
Hi
I am trying to run through
scikit-learn.github.com/scikit-learn-tutorial/working_with_text_data.html
as is.
Calls to SGDClassifier(...).fit(...), either within the pipe or directly,
invoke
"ValueError: setting an array element with a sequence."
Thanks much (great tool!!!)
---
The term "domain of applicability" is frequently used in the field of
cheminformatics to judge the reliability of predictions for new (unseen)
compounds.
=> the model should perform not very well for very dissimilar compounds
In my simple understanding, the predict_proba function gives me at le
33 matches
Mail list logo
