Re: [Scikit-learn-general] question on img_to_graph
Thanks! Le 25/06/2014 22:36, bthirion a écrit : > On 25/06/2014 22:00, Kevin Keraudren wrote: >> Hi, >> >> Would you know if img_to_graph is using 4-neighbourhoods in 2D, and >> 8-neigbourhoods in 3D, or 8 and 26? >> http://scikit-learn.org/stable/modules/generated/sklearn.feature_extraction.image.img_to_graph.html#sklearn.feature_extraction.image.img_to_graph > This is 6-neighbourhoods in 3D (4 in 2D). Best, > > Bertrand >> Thank you for your help, >> >> Kind regards, >> >> Kevin >> >> -- >> Open source business process management suite built on Java and Eclipse >> Turn processes into business applications with Bonita BPM Community Edition >> Quickly connect people, data, and systems into organized workflows >> Winner of BOSSIE, CODIE, OW2 and Gartner awards >> http://p.sf.net/sfu/Bonitasoft >> ___ >> Scikit-learn-general mailing list >> Scikit-learn-general@lists.sourceforge.net >> https://lists.sourceforge.net/lists/listinfo/scikit-learn-general > > -- > Open source business process management suite built on Java and Eclipse > Turn processes into business applications with Bonita BPM Community Edition > Quickly connect people, data, and systems into organized workflows > Winner of BOSSIE, CODIE, OW2 and Gartner awards > http://p.sf.net/sfu/Bonitasoft > ___ > Scikit-learn-general mailing list > Scikit-learn-general@lists.sourceforge.net > https://lists.sourceforge.net/lists/listinfo/scikit-learn-general -- Open source business process management suite built on Java and Eclipse Turn processes into business applications with Bonita BPM Community Edition Quickly connect people, data, and systems into organized workflows Winner of BOSSIE, CODIE, OW2 and Gartner awards http://p.sf.net/sfu/Bonitasoft ___ Scikit-learn-general mailing list Scikit-learn-general@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
[Scikit-learn-general] SVC.predict_proba result inconsistent with SVC.predict result
Hello all, I have the following code: . . . . # 'train' is a (M,N) numpy array (input) and 'traint' is a (M,) numpy array (target/label) clf = SVC(kernel="rbf", C=1.74, gamma=0.0023, probability=True) clf.fit(train, traint) print clf.classes_# Ensure our classes are [0,1] t1 = clf.predict_proba(matrix) t2 = clf.predict(matrix) print 'probabilities: ', t1 print 'class: ' , t2 #repeating for various matrices.. . . . . I suppose that t2 must be 0 when probability for the 1st class is bigger, else 1 However the two results (t1,t2) are inconsistent results: [0 1] probabilities: [[ 0.2629873 0.7370127]] class: [0] probabilities: [[ 0.22160846 0.77839154]] class: [0] *probabilities: [[ 0.17936026 0.82063974]] class: [1] * probabilities: [[ 0.28899413 0.71100587]] class: [0] *probabilities: [[ 0.17035052 0.82964948]] class: [1] * 3rd and 5th examples agree, but the rest? Any idea what's going on? Thanks for your time -- Open source business process management suite built on Java and Eclipse Turn processes into business applications with Bonita BPM Community Edition Quickly connect people, data, and systems into organized workflows Winner of BOSSIE, CODIE, OW2 and Gartner awards http://p.sf.net/sfu/Bonitasoft___ Scikit-learn-general mailing list Scikit-learn-general@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
[Scikit-learn-general] randomized l1
Hey, sorry I did not understand. Are you asking me to change the code in order to add these features? I can probably rewrite the randomized l1 class in order to include these. Only few changes are needed. On the other side I do not have a lot of experience with github so I am not sure how to do this. Let me know. Luca > > > Hi, > > I know that LARS is usually faster. > > On the other side CD is often considered more robust. In particular in > > situation p>>n the Lars is not able to include in the model more than n > > variables. > > > > Which doesn't mean that this is not a lasso solution - there always exists > a lasso solution with a support J such that X_J is injective, i.e. |J| \leq > n. On the other extreme, one can choose the minimum l_2 norm solution > (minimizing exactly the same functional), which maximizes the support. This > can also be done in homotopy algorithms such as LarsLasso, but happens to > not be implemented in scikit-learn. Any convex combination of the two is > also a solution, and there may be many others. CD may find a different one, > but it would be neither better nor worse than the mentioned options in > which concerns training error. In prediction, including as many correlated > variables as possible may yield more stability. > > > > > I think that the best think to do would be to include the possibility to > > choice which algorithm to use and leave Lars as the default choice. > > I think that should also be included the option to use as penalty path > the > > lasso penalty path. This will be closer to the original paper. > > > > Would you be able to add this functionality? Alex's usecase was rather > specific, and there may be other cases where it is indeed useful to have CD > as a possibility. The most helpful thing in assessing this would be a > benchmark showing the differences. > > > > I have seen that the current choice of using 'aic' or 'bic' alpha does > > not work well in some situations. > > Hope this could help, > > Luca > > > > > >> > >> > I was wondering if there is any reason of why the randomized l1 > >> algorithm > >> > from the stability selection paper is implemented only using Lars > Lasso > >> and > >> > not the coordinate descent algorithm. > >> > I think than including a version of the algorithm with the coordinate > >> > descent method would be very useful. > >> > >> because on our use case the lars was always faster. So there was no > >> point supporting both. > >> > >> Best, > >> Alex > >> > > > -- Open source business process management suite built on Java and Eclipse Turn processes into business applications with Bonita BPM Community Edition Quickly connect people, data, and systems into organized workflows Winner of BOSSIE, CODIE, OW2 and Gartner awards http://p.sf.net/sfu/Bonitasoft___ Scikit-learn-general mailing list Scikit-learn-general@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
Re: [Scikit-learn-general] question on img_to_graph
On 25/06/2014 22:00, Kevin Keraudren wrote: > Hi, > > Would you know if img_to_graph is using 4-neighbourhoods in 2D, and > 8-neigbourhoods in 3D, or 8 and 26? > http://scikit-learn.org/stable/modules/generated/sklearn.feature_extraction.image.img_to_graph.html#sklearn.feature_extraction.image.img_to_graph This is 6-neighbourhoods in 3D (4 in 2D). Best, Bertrand > Thank you for your help, > > Kind regards, > > Kevin > > -- > Open source business process management suite built on Java and Eclipse > Turn processes into business applications with Bonita BPM Community Edition > Quickly connect people, data, and systems into organized workflows > Winner of BOSSIE, CODIE, OW2 and Gartner awards > http://p.sf.net/sfu/Bonitasoft > ___ > Scikit-learn-general mailing list > Scikit-learn-general@lists.sourceforge.net > https://lists.sourceforge.net/lists/listinfo/scikit-learn-general -- Open source business process management suite built on Java and Eclipse Turn processes into business applications with Bonita BPM Community Edition Quickly connect people, data, and systems into organized workflows Winner of BOSSIE, CODIE, OW2 and Gartner awards http://p.sf.net/sfu/Bonitasoft ___ Scikit-learn-general mailing list Scikit-learn-general@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
Re: [Scikit-learn-general] Help for learning/contributing
Hi Joel, > Is that along the lines of the sort of things you would like to start on? Yes, I think these look good. I'll start looking into them and pop around here or github if anything comes up. Thanks Ignacio 2014-06-25 16:54 GMT-03:00 Joel Nothman : > Hi Ignacio, > > A good starting place is often working on the documentation. For example, > https://github.com/scikit-learn/scikit-learn/pull/3084 is an attempt at > filling in a gap in the documentation, but it doesn't look like Raul is > going to complete the work any time soon. If you want to pull his branch, > finish responding to the comments (i.e. fixing up the changes), and create > a new PR out of it, that would be helpful. > > A similarly stalled documentation change is > https://github.com/scikit-learn/scikit-learn/pull/2529 > > Also in the space of cross-validation (which I figure is a fairly trivial > ML idea to understand), but working on code: > https://github.com/scikit-learn/scikit-learn/issues/2904 is mostly an > issue of rewriting the cross_validation module to have a different API. As > suggested by the discussion there, its design may not be trivial, but it > would be very valuable to see a first attempt at the rewrite. > https://github.com/scikit-learn/scikit-learn/issues/1848 is similarly > mostly about moving things around, and it will help making a clean change > together with #2904. You should make sure you have read the contributing > guidelines and are familiar with the deprecation procedures for either of > these restructure tasks. > > Is that along the lines of the sort of things you would like to start on? > > Joel > > On 25 June 2014 14:56, Ignacio Rossi wrote: > >> Hello, >> >> Today I dropped by the freenode channel, but no one was around at the >> time, so I'm relaying my message here, thanks for reading :) >> >>I’m a math undergraduate who works as a python developer. I’m >> currently between jobs, so I have one month worth of spare time and i was >> thinking about learning ML and contributing to the project during that time >>I’ve peeked the issue tracker (the easy tag), but I might need some >> guidance on where to start. I don't know much about ML yet - taking the >> course on coursera, but wishing to contribute some code in the meantime >> >>Any suggestions regarding learning/contributing will be appreciated >> >>Thanks >> >>Ignacio >> >> >> -- >> Open source business process management suite built on Java and Eclipse >> Turn processes into business applications with Bonita BPM Community >> Edition >> Quickly connect people, data, and systems into organized workflows >> Winner of BOSSIE, CODIE, OW2 and Gartner awards >> http://p.sf.net/sfu/Bonitasoft >> ___ >> Scikit-learn-general mailing list >> Scikit-learn-general@lists.sourceforge.net >> https://lists.sourceforge.net/lists/listinfo/scikit-learn-general >> >> > > > -- > Open source business process management suite built on Java and Eclipse > Turn processes into business applications with Bonita BPM Community Edition > Quickly connect people, data, and systems into organized workflows > Winner of BOSSIE, CODIE, OW2 and Gartner awards > http://p.sf.net/sfu/Bonitasoft > ___ > Scikit-learn-general mailing list > Scikit-learn-general@lists.sourceforge.net > https://lists.sourceforge.net/lists/listinfo/scikit-learn-general > > -- Open source business process management suite built on Java and Eclipse Turn processes into business applications with Bonita BPM Community Edition Quickly connect people, data, and systems into organized workflows Winner of BOSSIE, CODIE, OW2 and Gartner awards http://p.sf.net/sfu/Bonitasoft___ Scikit-learn-general mailing list Scikit-learn-general@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
[Scikit-learn-general] question on img_to_graph
Hi, Would you know if img_to_graph is using 4-neighbourhoods in 2D, and 8-neigbourhoods in 3D, or 8 and 26? http://scikit-learn.org/stable/modules/generated/sklearn.feature_extraction.image.img_to_graph.html#sklearn.feature_extraction.image.img_to_graph Thank you for your help, Kind regards, Kevin -- Open source business process management suite built on Java and Eclipse Turn processes into business applications with Bonita BPM Community Edition Quickly connect people, data, and systems into organized workflows Winner of BOSSIE, CODIE, OW2 and Gartner awards http://p.sf.net/sfu/Bonitasoft ___ Scikit-learn-general mailing list Scikit-learn-general@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
Re: [Scikit-learn-general] Help for learning/contributing
Hi Ignacio, A good starting place is often working on the documentation. For example, https://github.com/scikit-learn/scikit-learn/pull/3084 is an attempt at filling in a gap in the documentation, but it doesn't look like Raul is going to complete the work any time soon. If you want to pull his branch, finish responding to the comments (i.e. fixing up the changes), and create a new PR out of it, that would be helpful. A similarly stalled documentation change is https://github.com/scikit-learn/scikit-learn/pull/2529 Also in the space of cross-validation (which I figure is a fairly trivial ML idea to understand), but working on code: https://github.com/scikit-learn/scikit-learn/issues/2904 is mostly an issue of rewriting the cross_validation module to have a different API. As suggested by the discussion there, its design may not be trivial, but it would be very valuable to see a first attempt at the rewrite. https://github.com/scikit-learn/scikit-learn/issues/1848 is similarly mostly about moving things around, and it will help making a clean change together with #2904. You should make sure you have read the contributing guidelines and are familiar with the deprecation procedures for either of these restructure tasks. Is that along the lines of the sort of things you would like to start on? Joel On 25 June 2014 14:56, Ignacio Rossi wrote: > Hello, > > Today I dropped by the freenode channel, but no one was around at the > time, so I'm relaying my message here, thanks for reading :) > >I’m a math undergraduate who works as a python developer. I’m currently > between jobs, so I have one month worth of spare time and i was thinking > about learning ML and contributing to the project during that time >I’ve peeked the issue tracker (the easy tag), but I might need some > guidance on where to start. I don't know much about ML yet - taking the > course on coursera, but wishing to contribute some code in the meantime > >Any suggestions regarding learning/contributing will be appreciated > >Thanks > >Ignacio > > > -- > Open source business process management suite built on Java and Eclipse > Turn processes into business applications with Bonita BPM Community Edition > Quickly connect people, data, and systems into organized workflows > Winner of BOSSIE, CODIE, OW2 and Gartner awards > http://p.sf.net/sfu/Bonitasoft > ___ > Scikit-learn-general mailing list > Scikit-learn-general@lists.sourceforge.net > https://lists.sourceforge.net/lists/listinfo/scikit-learn-general > > -- Open source business process management suite built on Java and Eclipse Turn processes into business applications with Bonita BPM Community Edition Quickly connect people, data, and systems into organized workflows Winner of BOSSIE, CODIE, OW2 and Gartner awards http://p.sf.net/sfu/Bonitasoft___ Scikit-learn-general mailing list Scikit-learn-general@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
Re: [Scikit-learn-general] Help for learning/contributing
Hello, Today I dropped by the freenode channel, but no one was around at the time, so I'm relaying my message here, thanks for reading :) I’m a math undergraduate who works as a python developer. I’m currently between jobs, so I have one month worth of spare time and i was thinking about learning ML and contributing to the project during that time I’ve peeked the issue tracker (the easy tag), but I might need some guidance on where to start. I don't know much about ML yet - taking the course on coursera, but wishing to contribute some code in the meantime Any suggestions regarding learning/contributing will be appreciated Thanks Ignacio -- Open source business process management suite built on Java and Eclipse Turn processes into business applications with Bonita BPM Community Edition Quickly connect people, data, and systems into organized workflows Winner of BOSSIE, CODIE, OW2 and Gartner awards http://p.sf.net/sfu/Bonitasoft___ Scikit-learn-general mailing list Scikit-learn-general@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
Re: [Scikit-learn-general] Sparse NMF
Vlad, if you are going to cite the Mazack paper, you should also look into the original paper by Kim and Park (2006) which proposes the sparse NMF model ( the one which is currently implemented by Scikit-learn). For the sparse NMF model by Hoyer (2004), some benchmarks can be found in my ICLR paper. They were written in Matlab. Will convert them to Python as well and send you the link. Thanks, ~Vamsi. On Wed, Jun 25, 2014 at 11:37 AM, Vlad Niculae wrote: > Vamsi, do you have any benchmarks for your implementation? The biggest > reason why we didn't change the current implementation yet is that it > was hard to find something else that is consistently faster & better. > > Here are some benchmarks I did last time I looked at this, when I > wanted to replace the hacky regularization currently implemented with > an actual elastic net penalty. [1] is for just the non-negative least > squares solver, [2] and [3] are NMF implementations on top of the PG > and multiplicative methods. > I was pretty disappointed with not managing to find one configuration > that is consistently better, so I stopped pursuing this, as I'm not > using NMF in my work at the moment. > > Vlad > > [1] > http://nbviewer.ipython.org/urls/gist.githubusercontent.com/vene/7224672/raw/f9ab27a74addcca9e5e911c47478641d137ebc35/nls_solvers.ipynb > [2] > http://nbviewer.ipython.org/urls/gist.githubusercontent.com/vene/7196622/raw/7c7764ca9d0a6effba4976c5818933ef058a01ac/compare_nmf.ipynb > [3] > http://nbviewer.ipython.org/urls/gist.github.com/vene/7196622/raw/63f0005cf37cccf6d3b58b3d63b8aa9f6701b36b/compare_nmf_faces.ipynb > > On Wed, Jun 25, 2014 at 11:28 AM, Michael Eickenberg > wrote: > > i will never post a docstring again :) > > sorry for the noise > > > > michael > > > > > > On Wednesday, June 25, 2014, Vlad Niculae wrote: > >> > >> Hi, > >> > >> Allow me to clarify. We don't implement Hoyer's sparse update rule > >> indeed (it shouldn't say "this implements", I initially cited Hoyer > >> for motivating sparseness constraints in NMF). Instead, we implement a > >> version of sparse NMF with a clear (but not particularly elegant) > >> objective function, found in [1]. > >> > >> I had somewhere an implementation that just alternatively optimizes > >> elastic net problems in both W and H, which is more elegant but I > >> didn't finish polishing and benchmarking it. > >> > >> I should probably update the docstring, sorry for the confusion. > >> > >> Yours, > >> Vlad > >> > >> [1] http://mazack.org/papers/mazack_nmf_paper.pdf > >> > >> On Wed, Jun 25, 2014 at 11:00 AM, Vamsi Krishna Potluru > >> wrote: > >> > Don't believe all you read :-). > >> > It says that but implements something else. > >> > > >> > ~Vamsi. > >> > > >> > > >> > On Wed, Jun 25, 2014 at 10:48 AM, Michael Eickenberg > >> > wrote: > >> >> > >> >> > >> >> > >> >> > https://github.com/scikit-learn/scikit-learn/blob/master/sklearn/decomposition/nmf.py#L346 > >> >> > >> >> > >> >> > >> >> References > >> >> > >> >> -- > >> >> This implements > >> >> > >> >> > >> >> C.-J. Lin. Projected gradient methods > >> >> for non-negative matrix factorization. Neural > >> >> Computation, 19(2007), 2756-2779. > >> >> http://www.csie.ntu.edu.tw/~cjlin/nmf/ > >> >> > >> >> > >> >> P. Hoyer. Non-negative Matrix Factorization with > >> >> Sparseness Constraints. Journal of Machine Learning > >> >> Research 2004. > >> >> > >> >> > >> >> > >> >> On Wed, Jun 25, 2014 at 4:39 PM, Vamsi Krishna Potluru > >> >> wrote: > >> >>> > >> >>> Yes, thats a different sparse model than the one in Hoyer, 2004 > (most > >> >>> likely the one by Kim and Park). > >> >>> Maybe, Vlad or someone else can comment on that. > >> >>> > >> >>> ~Vamsi. > >> >>> > >> >>> > >> >>> On Wed, Jun 25, 2014 at 10:15 AM, Alexandre Gramfort > >> >>> wrote: > >> > >> hi, > >> > >> have you played with the sparseness parameter of the NMF estimator? > >> > >> Alex > >> > >> > >> On Wed, Jun 25, 2014 at 3:42 PM, Vamsi Krishna Potluru > >> wrote: > >> > Hello, > >> > > >> > I have worked a bit on the sparse NMF model proposed by Hoyer > [1]. > >> > The > >> > paper is mentioned in the Scikits NMF module but AFAIK the model > is > >> > currently not implemented. Recently, we proposed an efficient > >> > algorithm > >> > based on block coordinate descent [2]. A reference python > >> > implementation is > >> > available at: > >> > > >> > https://github.com/ismav/sparseNMF > >> > > >> > Kindly let me know if this would be of interest to the > >> > Scikits-learn > >> > community. > >> > > >> > Thanks, > >> > > >> > ~Vamsi. > >> > > >> > > >> > [1] Hoyer, P. O. (2004). Non-negative Matrix Factorization with > >> > Sparseness > >> > Constraints. Journal of Machine Learning Research, 5, 1457-1469. > >> > > >> >>
Re: [Scikit-learn-general] Sparse NMF
Vamsi, do you have any benchmarks for your implementation? The biggest reason why we didn't change the current implementation yet is that it was hard to find something else that is consistently faster & better. Here are some benchmarks I did last time I looked at this, when I wanted to replace the hacky regularization currently implemented with an actual elastic net penalty. [1] is for just the non-negative least squares solver, [2] and [3] are NMF implementations on top of the PG and multiplicative methods. I was pretty disappointed with not managing to find one configuration that is consistently better, so I stopped pursuing this, as I'm not using NMF in my work at the moment. Vlad [1] http://nbviewer.ipython.org/urls/gist.githubusercontent.com/vene/7224672/raw/f9ab27a74addcca9e5e911c47478641d137ebc35/nls_solvers.ipynb [2] http://nbviewer.ipython.org/urls/gist.githubusercontent.com/vene/7196622/raw/7c7764ca9d0a6effba4976c5818933ef058a01ac/compare_nmf.ipynb [3] http://nbviewer.ipython.org/urls/gist.github.com/vene/7196622/raw/63f0005cf37cccf6d3b58b3d63b8aa9f6701b36b/compare_nmf_faces.ipynb On Wed, Jun 25, 2014 at 11:28 AM, Michael Eickenberg wrote: > i will never post a docstring again :) > sorry for the noise > > michael > > > On Wednesday, June 25, 2014, Vlad Niculae wrote: >> >> Hi, >> >> Allow me to clarify. We don't implement Hoyer's sparse update rule >> indeed (it shouldn't say "this implements", I initially cited Hoyer >> for motivating sparseness constraints in NMF). Instead, we implement a >> version of sparse NMF with a clear (but not particularly elegant) >> objective function, found in [1]. >> >> I had somewhere an implementation that just alternatively optimizes >> elastic net problems in both W and H, which is more elegant but I >> didn't finish polishing and benchmarking it. >> >> I should probably update the docstring, sorry for the confusion. >> >> Yours, >> Vlad >> >> [1] http://mazack.org/papers/mazack_nmf_paper.pdf >> >> On Wed, Jun 25, 2014 at 11:00 AM, Vamsi Krishna Potluru >> wrote: >> > Don't believe all you read :-). >> > It says that but implements something else. >> > >> > ~Vamsi. >> > >> > >> > On Wed, Jun 25, 2014 at 10:48 AM, Michael Eickenberg >> > wrote: >> >> >> >> >> >> >> >> https://github.com/scikit-learn/scikit-learn/blob/master/sklearn/decomposition/nmf.py#L346 >> >> >> >> >> >> >> >> References >> >> >> >> -- >> >> This implements >> >> >> >> >> >> C.-J. Lin. Projected gradient methods >> >> for non-negative matrix factorization. Neural >> >> Computation, 19(2007), 2756-2779. >> >> http://www.csie.ntu.edu.tw/~cjlin/nmf/ >> >> >> >> >> >> P. Hoyer. Non-negative Matrix Factorization with >> >> Sparseness Constraints. Journal of Machine Learning >> >> Research 2004. >> >> >> >> >> >> >> >> On Wed, Jun 25, 2014 at 4:39 PM, Vamsi Krishna Potluru >> >> wrote: >> >>> >> >>> Yes, thats a different sparse model than the one in Hoyer, 2004 (most >> >>> likely the one by Kim and Park). >> >>> Maybe, Vlad or someone else can comment on that. >> >>> >> >>> ~Vamsi. >> >>> >> >>> >> >>> On Wed, Jun 25, 2014 at 10:15 AM, Alexandre Gramfort >> >>> wrote: >> >> hi, >> >> have you played with the sparseness parameter of the NMF estimator? >> >> Alex >> >> >> On Wed, Jun 25, 2014 at 3:42 PM, Vamsi Krishna Potluru >> wrote: >> > Hello, >> > >> > I have worked a bit on the sparse NMF model proposed by Hoyer [1]. >> > The >> > paper is mentioned in the Scikits NMF module but AFAIK the model is >> > currently not implemented. Recently, we proposed an efficient >> > algorithm >> > based on block coordinate descent [2]. A reference python >> > implementation is >> > available at: >> > >> > https://github.com/ismav/sparseNMF >> > >> > Kindly let me know if this would be of interest to the >> > Scikits-learn >> > community. >> > >> > Thanks, >> > >> > ~Vamsi. >> > >> > >> > [1] Hoyer, P. O. (2004). Non-negative Matrix Factorization with >> > Sparseness >> > Constraints. Journal of Machine Learning Research, 5, 1457-1469. >> > >> > [2] Block Coordinate Descent for Sparse NMF >> > >> > Vamsi K. Potluru, Sergey M. Plis, Jonathan Le Roux, Barak A. >> > Pearlmutter, >> > Vince D. Calhoun, Thomas P. Hayes. ICLR 2013. >> > >> > http://arxiv.org/abs/1301.3527 >> > >> > >> > >> > >> > -- >> > Open source business process management suite built on Java and >> > Eclipse >> > Turn processes into business applications with Bonita BPM Community >> > Edition >> > Quickly connect people, data, and systems into organized workflows >> > Winner of BOSSIE, CODIE, OW2 and Gartner awards >> > http://p.sf.net
Re: [Scikit-learn-general] Sparse NMF
i will never post a docstring again :) sorry for the noise michael On Wednesday, June 25, 2014, Vlad Niculae wrote: > Hi, > > Allow me to clarify. We don't implement Hoyer's sparse update rule > indeed (it shouldn't say "this implements", I initially cited Hoyer > for motivating sparseness constraints in NMF). Instead, we implement a > version of sparse NMF with a clear (but not particularly elegant) > objective function, found in [1]. > > I had somewhere an implementation that just alternatively optimizes > elastic net problems in both W and H, which is more elegant but I > didn't finish polishing and benchmarking it. > > I should probably update the docstring, sorry for the confusion. > > Yours, > Vlad > > [1] http://mazack.org/papers/mazack_nmf_paper.pdf > > On Wed, Jun 25, 2014 at 11:00 AM, Vamsi Krishna Potluru > > wrote: > > Don't believe all you read :-). > > It says that but implements something else. > > > > ~Vamsi. > > > > > > On Wed, Jun 25, 2014 at 10:48 AM, Michael Eickenberg > > > wrote: > >> > >> > >> > https://github.com/scikit-learn/scikit-learn/blob/master/sklearn/decomposition/nmf.py#L346 > >> > >> > >> > >> References > >> > >> -- > >> This implements > >> > >> > >> C.-J. Lin. Projected gradient methods > >> for non-negative matrix factorization. Neural > >> Computation, 19(2007), 2756-2779. > >> http://www.csie.ntu.edu.tw/~cjlin/nmf/ > >> > >> > >> P. Hoyer. Non-negative Matrix Factorization with > >> Sparseness Constraints. Journal of Machine Learning > >> Research 2004. > >> > >> > >> > >> On Wed, Jun 25, 2014 at 4:39 PM, Vamsi Krishna Potluru > >> > wrote: > >>> > >>> Yes, thats a different sparse model than the one in Hoyer, 2004 (most > >>> likely the one by Kim and Park). > >>> Maybe, Vlad or someone else can comment on that. > >>> > >>> ~Vamsi. > >>> > >>> > >>> On Wed, Jun 25, 2014 at 10:15 AM, Alexandre Gramfort > >>> > wrote: > > hi, > > have you played with the sparseness parameter of the NMF estimator? > > Alex > > > On Wed, Jun 25, 2014 at 3:42 PM, Vamsi Krishna Potluru > > wrote: > > Hello, > > > > I have worked a bit on the sparse NMF model proposed by Hoyer [1]. > > The > > paper is mentioned in the Scikits NMF module but AFAIK the model is > > currently not implemented. Recently, we proposed an efficient > > algorithm > > based on block coordinate descent [2]. A reference python > > implementation is > > available at: > > > > https://github.com/ismav/sparseNMF > > > > Kindly let me know if this would be of interest to the Scikits-learn > > community. > > > > Thanks, > > > > ~Vamsi. > > > > > > [1] Hoyer, P. O. (2004). Non-negative Matrix Factorization with > > Sparseness > > Constraints. Journal of Machine Learning Research, 5, 1457-1469. > > > > [2] Block Coordinate Descent for Sparse NMF > > > > Vamsi K. Potluru, Sergey M. Plis, Jonathan Le Roux, Barak A. > > Pearlmutter, > > Vince D. Calhoun, Thomas P. Hayes. ICLR 2013. > > > > http://arxiv.org/abs/1301.3527 > > > > > > > > > -- > > Open source business process management suite built on Java and > > Eclipse > > Turn processes into business applications with Bonita BPM Community > > Edition > > Quickly connect people, data, and systems into organized workflows > > Winner of BOSSIE, CODIE, OW2 and Gartner awards > > http://p.sf.net/sfu/Bonitasoft > > ___ > > Scikit-learn-general mailing list > > Scikit-learn-general@lists.sourceforge.net > > https://lists.sourceforge.net/lists/listinfo/scikit-learn-general > > > > > > -- > Open source business process management suite built on Java and > Eclipse > Turn processes into business applications with Bonita BPM Community > Edition > Quickly connect people, data, and systems into organized workflows > Winner of BOSSIE, CODIE, OW2 and Gartner awards > http://p.sf.net/sfu/Bonitasoft > ___ > Scikit-learn-general mailing list > Scikit-learn-general@lists.sourceforge.net > https://lists.sourceforge.net/lists/listinfo/scikit-learn-general > >>> > >>> > >>> > >>> > >>> > -- > >>> Open source business process management suite built on Java and Eclipse > >>> Turn processes into business applications with Bonita BPM Community > >>> Edition > >>> Quickly connect people, data, and systems into organized workflows > >>> Winner of BOSSIE, CODIE, OW2 and Gartner awards > >>> ht
Re: [Scikit-learn-general] Sparse NMF
Hi, Allow me to clarify. We don't implement Hoyer's sparse update rule indeed (it shouldn't say "this implements", I initially cited Hoyer for motivating sparseness constraints in NMF). Instead, we implement a version of sparse NMF with a clear (but not particularly elegant) objective function, found in [1]. I had somewhere an implementation that just alternatively optimizes elastic net problems in both W and H, which is more elegant but I didn't finish polishing and benchmarking it. I should probably update the docstring, sorry for the confusion. Yours, Vlad [1] http://mazack.org/papers/mazack_nmf_paper.pdf On Wed, Jun 25, 2014 at 11:00 AM, Vamsi Krishna Potluru wrote: > Don't believe all you read :-). > It says that but implements something else. > > ~Vamsi. > > > On Wed, Jun 25, 2014 at 10:48 AM, Michael Eickenberg > wrote: >> >> >> https://github.com/scikit-learn/scikit-learn/blob/master/sklearn/decomposition/nmf.py#L346 >> >> >> >> References >> >> -- >> This implements >> >> >> C.-J. Lin. Projected gradient methods >> for non-negative matrix factorization. Neural >> Computation, 19(2007), 2756-2779. >> http://www.csie.ntu.edu.tw/~cjlin/nmf/ >> >> >> P. Hoyer. Non-negative Matrix Factorization with >> Sparseness Constraints. Journal of Machine Learning >> Research 2004. >> >> >> >> On Wed, Jun 25, 2014 at 4:39 PM, Vamsi Krishna Potluru >> wrote: >>> >>> Yes, thats a different sparse model than the one in Hoyer, 2004 (most >>> likely the one by Kim and Park). >>> Maybe, Vlad or someone else can comment on that. >>> >>> ~Vamsi. >>> >>> >>> On Wed, Jun 25, 2014 at 10:15 AM, Alexandre Gramfort >>> wrote: hi, have you played with the sparseness parameter of the NMF estimator? Alex On Wed, Jun 25, 2014 at 3:42 PM, Vamsi Krishna Potluru wrote: > Hello, > > I have worked a bit on the sparse NMF model proposed by Hoyer [1]. > The > paper is mentioned in the Scikits NMF module but AFAIK the model is > currently not implemented. Recently, we proposed an efficient > algorithm > based on block coordinate descent [2]. A reference python > implementation is > available at: > > https://github.com/ismav/sparseNMF > > Kindly let me know if this would be of interest to the Scikits-learn > community. > > Thanks, > > ~Vamsi. > > > [1] Hoyer, P. O. (2004). Non-negative Matrix Factorization with > Sparseness > Constraints. Journal of Machine Learning Research, 5, 1457-1469. > > [2] Block Coordinate Descent for Sparse NMF > > Vamsi K. Potluru, Sergey M. Plis, Jonathan Le Roux, Barak A. > Pearlmutter, > Vince D. Calhoun, Thomas P. Hayes. ICLR 2013. > > http://arxiv.org/abs/1301.3527 > > > > -- > Open source business process management suite built on Java and > Eclipse > Turn processes into business applications with Bonita BPM Community > Edition > Quickly connect people, data, and systems into organized workflows > Winner of BOSSIE, CODIE, OW2 and Gartner awards > http://p.sf.net/sfu/Bonitasoft > ___ > Scikit-learn-general mailing list > Scikit-learn-general@lists.sourceforge.net > https://lists.sourceforge.net/lists/listinfo/scikit-learn-general > -- Open source business process management suite built on Java and Eclipse Turn processes into business applications with Bonita BPM Community Edition Quickly connect people, data, and systems into organized workflows Winner of BOSSIE, CODIE, OW2 and Gartner awards http://p.sf.net/sfu/Bonitasoft ___ Scikit-learn-general mailing list Scikit-learn-general@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/scikit-learn-general >>> >>> >>> >>> >>> -- >>> Open source business process management suite built on Java and Eclipse >>> Turn processes into business applications with Bonita BPM Community >>> Edition >>> Quickly connect people, data, and systems into organized workflows >>> Winner of BOSSIE, CODIE, OW2 and Gartner awards >>> http://p.sf.net/sfu/Bonitasoft >>> ___ >>> Scikit-learn-general mailing list >>> Scikit-learn-general@lists.sourceforge.net >>> https://lists.sourceforge.net/lists/listinfo/scikit-learn-general >>> >> >> >> >> -- >> Open source business process management suite built on Java and Eclipse >> Turn processes i
Re: [Scikit-learn-general] Sparse NMF
Don't believe all you read :-). It says that but implements something else. ~Vamsi. On Wed, Jun 25, 2014 at 10:48 AM, Michael Eickenberg < michael.eickenb...@gmail.com> wrote: > https://github.com/scikit-learn/scikit-learn/blob/master/sklearn/decomposition/nmf.py#L346 > > References > -- > This implements > > C.-J. Lin. Projected gradient methods > for non-negative matrix factorization. Neural > Computation, 19(2007), 2756-2779. > http://www.csie.ntu.edu.tw/~cjlin/nmf/ > > P. Hoyer. Non-negative Matrix Factorization with > Sparseness Constraints. Journal of Machine Learning > Research 2004. > > > > On Wed, Jun 25, 2014 at 4:39 PM, Vamsi Krishna Potluru < > vamsi.potl...@rutgers.edu> wrote: > >> Yes, thats a different sparse model than the one in Hoyer, 2004 (most >> likely the one by Kim and Park). >> Maybe, Vlad or someone else can comment on that. >> >> ~Vamsi. >> >> >> On Wed, Jun 25, 2014 at 10:15 AM, Alexandre Gramfort < >> alexandre.gramf...@telecom-paristech.fr> wrote: >> >>> hi, >>> >>> have you played with the sparseness parameter of the NMF estimator? >>> >>> Alex >>> >>> >>> On Wed, Jun 25, 2014 at 3:42 PM, Vamsi Krishna Potluru >>> wrote: >>> > Hello, >>> > >>> > I have worked a bit on the sparse NMF model proposed by Hoyer [1]. The >>> > paper is mentioned in the Scikits NMF module but AFAIK the model is >>> > currently not implemented. Recently, we proposed an efficient algorithm >>> > based on block coordinate descent [2]. A reference python >>> implementation is >>> > available at: >>> > >>> > https://github.com/ismav/sparseNMF >>> > >>> > Kindly let me know if this would be of interest to the Scikits-learn >>> > community. >>> > >>> > Thanks, >>> > >>> > ~Vamsi. >>> > >>> > >>> > [1] Hoyer, P. O. (2004). Non-negative Matrix Factorization with >>> Sparseness >>> > Constraints. Journal of Machine Learning Research, 5, 1457-1469. >>> > >>> > [2] Block Coordinate Descent for Sparse NMF >>> > >>> > Vamsi K. Potluru, Sergey M. Plis, Jonathan Le Roux, Barak A. >>> Pearlmutter, >>> > Vince D. Calhoun, Thomas P. Hayes. ICLR 2013. >>> > >>> > http://arxiv.org/abs/1301.3527 >>> > >>> > >>> > >>> -- >>> > Open source business process management suite built on Java and Eclipse >>> > Turn processes into business applications with Bonita BPM Community >>> Edition >>> > Quickly connect people, data, and systems into organized workflows >>> > Winner of BOSSIE, CODIE, OW2 and Gartner awards >>> > http://p.sf.net/sfu/Bonitasoft >>> > ___ >>> > Scikit-learn-general mailing list >>> > Scikit-learn-general@lists.sourceforge.net >>> > https://lists.sourceforge.net/lists/listinfo/scikit-learn-general >>> > >>> >>> >>> -- >>> Open source business process management suite built on Java and Eclipse >>> Turn processes into business applications with Bonita BPM Community >>> Edition >>> Quickly connect people, data, and systems into organized workflows >>> Winner of BOSSIE, CODIE, OW2 and Gartner awards >>> http://p.sf.net/sfu/Bonitasoft >>> ___ >>> Scikit-learn-general mailing list >>> Scikit-learn-general@lists.sourceforge.net >>> https://lists.sourceforge.net/lists/listinfo/scikit-learn-general >>> >> >> >> >> -- >> Open source business process management suite built on Java and Eclipse >> Turn processes into business applications with Bonita BPM Community >> Edition >> Quickly connect people, data, and systems into organized workflows >> Winner of BOSSIE, CODIE, OW2 and Gartner awards >> http://p.sf.net/sfu/Bonitasoft >> ___ >> Scikit-learn-general mailing list >> Scikit-learn-general@lists.sourceforge.net >> https://lists.sourceforge.net/lists/listinfo/scikit-learn-general >> >> > > > -- > Open source business process management suite built on Java and Eclipse > Turn processes into business applications with Bonita BPM Community Edition > Quickly connect people, data, and systems into organized workflows > Winner of BOSSIE, CODIE, OW2 and Gartner awards > http://p.sf.net/sfu/Bonitasoft > ___ > Scikit-learn-general mailing list > Scikit-learn-general@lists.sourceforge.net > https://lists.sourceforge.net/lists/listinfo/scikit-learn-general > > -- Open source business process management suite built on Java and Eclipse Turn processes into business applications with Bonita BPM Community Edition Quickly connect people, data, and systems into organized workflows Winner of BOSSIE, CODIE, OW2 and Gartner awards http://p.s
Re: [Scikit-learn-general] Sparse NMF
https://github.com/scikit-learn/scikit-learn/blob/master/sklearn/decomposition/nmf.py#L346 References -- This implements C.-J. Lin. Projected gradient methods for non-negative matrix factorization. Neural Computation, 19(2007), 2756-2779. http://www.csie.ntu.edu.tw/~cjlin/nmf/ P. Hoyer. Non-negative Matrix Factorization with Sparseness Constraints. Journal of Machine Learning Research 2004. On Wed, Jun 25, 2014 at 4:39 PM, Vamsi Krishna Potluru < vamsi.potl...@rutgers.edu> wrote: > Yes, thats a different sparse model than the one in Hoyer, 2004 (most > likely the one by Kim and Park). > Maybe, Vlad or someone else can comment on that. > > ~Vamsi. > > > On Wed, Jun 25, 2014 at 10:15 AM, Alexandre Gramfort < > alexandre.gramf...@telecom-paristech.fr> wrote: > >> hi, >> >> have you played with the sparseness parameter of the NMF estimator? >> >> Alex >> >> >> On Wed, Jun 25, 2014 at 3:42 PM, Vamsi Krishna Potluru >> wrote: >> > Hello, >> > >> > I have worked a bit on the sparse NMF model proposed by Hoyer [1]. The >> > paper is mentioned in the Scikits NMF module but AFAIK the model is >> > currently not implemented. Recently, we proposed an efficient algorithm >> > based on block coordinate descent [2]. A reference python >> implementation is >> > available at: >> > >> > https://github.com/ismav/sparseNMF >> > >> > Kindly let me know if this would be of interest to the Scikits-learn >> > community. >> > >> > Thanks, >> > >> > ~Vamsi. >> > >> > >> > [1] Hoyer, P. O. (2004). Non-negative Matrix Factorization with >> Sparseness >> > Constraints. Journal of Machine Learning Research, 5, 1457-1469. >> > >> > [2] Block Coordinate Descent for Sparse NMF >> > >> > Vamsi K. Potluru, Sergey M. Plis, Jonathan Le Roux, Barak A. >> Pearlmutter, >> > Vince D. Calhoun, Thomas P. Hayes. ICLR 2013. >> > >> > http://arxiv.org/abs/1301.3527 >> > >> > >> > >> -- >> > Open source business process management suite built on Java and Eclipse >> > Turn processes into business applications with Bonita BPM Community >> Edition >> > Quickly connect people, data, and systems into organized workflows >> > Winner of BOSSIE, CODIE, OW2 and Gartner awards >> > http://p.sf.net/sfu/Bonitasoft >> > ___ >> > Scikit-learn-general mailing list >> > Scikit-learn-general@lists.sourceforge.net >> > https://lists.sourceforge.net/lists/listinfo/scikit-learn-general >> > >> >> >> -- >> Open source business process management suite built on Java and Eclipse >> Turn processes into business applications with Bonita BPM Community >> Edition >> Quickly connect people, data, and systems into organized workflows >> Winner of BOSSIE, CODIE, OW2 and Gartner awards >> http://p.sf.net/sfu/Bonitasoft >> ___ >> Scikit-learn-general mailing list >> Scikit-learn-general@lists.sourceforge.net >> https://lists.sourceforge.net/lists/listinfo/scikit-learn-general >> > > > > -- > Open source business process management suite built on Java and Eclipse > Turn processes into business applications with Bonita BPM Community Edition > Quickly connect people, data, and systems into organized workflows > Winner of BOSSIE, CODIE, OW2 and Gartner awards > http://p.sf.net/sfu/Bonitasoft > ___ > Scikit-learn-general mailing list > Scikit-learn-general@lists.sourceforge.net > https://lists.sourceforge.net/lists/listinfo/scikit-learn-general > > -- Open source business process management suite built on Java and Eclipse Turn processes into business applications with Bonita BPM Community Edition Quickly connect people, data, and systems into organized workflows Winner of BOSSIE, CODIE, OW2 and Gartner awards http://p.sf.net/sfu/Bonitasoft___ Scikit-learn-general mailing list Scikit-learn-general@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
Re: [Scikit-learn-general] Sparse NMF
Yes, thats a different sparse model than the one in Hoyer, 2004 (most likely the one by Kim and Park). Maybe, Vlad or someone else can comment on that. ~Vamsi. On Wed, Jun 25, 2014 at 10:15 AM, Alexandre Gramfort < alexandre.gramf...@telecom-paristech.fr> wrote: > hi, > > have you played with the sparseness parameter of the NMF estimator? > > Alex > > > On Wed, Jun 25, 2014 at 3:42 PM, Vamsi Krishna Potluru > wrote: > > Hello, > > > > I have worked a bit on the sparse NMF model proposed by Hoyer [1]. The > > paper is mentioned in the Scikits NMF module but AFAIK the model is > > currently not implemented. Recently, we proposed an efficient algorithm > > based on block coordinate descent [2]. A reference python implementation > is > > available at: > > > > https://github.com/ismav/sparseNMF > > > > Kindly let me know if this would be of interest to the Scikits-learn > > community. > > > > Thanks, > > > > ~Vamsi. > > > > > > [1] Hoyer, P. O. (2004). Non-negative Matrix Factorization with > Sparseness > > Constraints. Journal of Machine Learning Research, 5, 1457-1469. > > > > [2] Block Coordinate Descent for Sparse NMF > > > > Vamsi K. Potluru, Sergey M. Plis, Jonathan Le Roux, Barak A. Pearlmutter, > > Vince D. Calhoun, Thomas P. Hayes. ICLR 2013. > > > > http://arxiv.org/abs/1301.3527 > > > > > > > -- > > Open source business process management suite built on Java and Eclipse > > Turn processes into business applications with Bonita BPM Community > Edition > > Quickly connect people, data, and systems into organized workflows > > Winner of BOSSIE, CODIE, OW2 and Gartner awards > > http://p.sf.net/sfu/Bonitasoft > > ___ > > Scikit-learn-general mailing list > > Scikit-learn-general@lists.sourceforge.net > > https://lists.sourceforge.net/lists/listinfo/scikit-learn-general > > > > > -- > Open source business process management suite built on Java and Eclipse > Turn processes into business applications with Bonita BPM Community Edition > Quickly connect people, data, and systems into organized workflows > Winner of BOSSIE, CODIE, OW2 and Gartner awards > http://p.sf.net/sfu/Bonitasoft > ___ > Scikit-learn-general mailing list > Scikit-learn-general@lists.sourceforge.net > https://lists.sourceforge.net/lists/listinfo/scikit-learn-general > -- Open source business process management suite built on Java and Eclipse Turn processes into business applications with Bonita BPM Community Edition Quickly connect people, data, and systems into organized workflows Winner of BOSSIE, CODIE, OW2 and Gartner awards http://p.sf.net/sfu/Bonitasoft___ Scikit-learn-general mailing list Scikit-learn-general@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
Re: [Scikit-learn-general] Sparse NMF
hi, have you played with the sparseness parameter of the NMF estimator? Alex On Wed, Jun 25, 2014 at 3:42 PM, Vamsi Krishna Potluru wrote: > Hello, > > I have worked a bit on the sparse NMF model proposed by Hoyer [1]. The > paper is mentioned in the Scikits NMF module but AFAIK the model is > currently not implemented. Recently, we proposed an efficient algorithm > based on block coordinate descent [2]. A reference python implementation is > available at: > > https://github.com/ismav/sparseNMF > > Kindly let me know if this would be of interest to the Scikits-learn > community. > > Thanks, > > ~Vamsi. > > > [1] Hoyer, P. O. (2004). Non-negative Matrix Factorization with Sparseness > Constraints. Journal of Machine Learning Research, 5, 1457-1469. > > [2] Block Coordinate Descent for Sparse NMF > > Vamsi K. Potluru, Sergey M. Plis, Jonathan Le Roux, Barak A. Pearlmutter, > Vince D. Calhoun, Thomas P. Hayes. ICLR 2013. > > http://arxiv.org/abs/1301.3527 > > > -- > Open source business process management suite built on Java and Eclipse > Turn processes into business applications with Bonita BPM Community Edition > Quickly connect people, data, and systems into organized workflows > Winner of BOSSIE, CODIE, OW2 and Gartner awards > http://p.sf.net/sfu/Bonitasoft > ___ > Scikit-learn-general mailing list > Scikit-learn-general@lists.sourceforge.net > https://lists.sourceforge.net/lists/listinfo/scikit-learn-general > -- Open source business process management suite built on Java and Eclipse Turn processes into business applications with Bonita BPM Community Edition Quickly connect people, data, and systems into organized workflows Winner of BOSSIE, CODIE, OW2 and Gartner awards http://p.sf.net/sfu/Bonitasoft ___ Scikit-learn-general mailing list Scikit-learn-general@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
[Scikit-learn-general] Sparse NMF
Hello, I have worked a bit on the sparse NMF model proposed by Hoyer [1]. The paper is mentioned in the Scikits NMF module but AFAIK the model is currently not implemented. Recently, we proposed an efficient algorithm based on block coordinate descent [2]. A reference python implementation is available at: https://github.com/ismav/sparseNMF Kindly let me know if this would be of interest to the Scikits-learn community. Thanks, ~Vamsi. [1] Hoyer, P. O. (2004). Non-negative Matrix Factorization with Sparseness Constraints. Journal of Machine Learning Research, 5, 1457-1469. [2] Block Coordinate Descent for Sparse NMF Vamsi K. Potluru, Sergey M. Plis, Jonathan Le Roux, Barak A. Pearlmutter, Vince D. Calhoun, Thomas P. Hayes. ICLR 2013. http://arxiv.org/abs/1301.3527 -- Open source business process management suite built on Java and Eclipse Turn processes into business applications with Bonita BPM Community Edition Quickly connect people, data, and systems into organized workflows Winner of BOSSIE, CODIE, OW2 and Gartner awards http://p.sf.net/sfu/Bonitasoft___ Scikit-learn-general mailing list Scikit-learn-general@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
Re: [Scikit-learn-general] Scikit learn's multiprocessing
2014-06-25 4:50 GMT+02:00 Sturla Molden : > In general, only POSIX APIs are safe to use on both sides of a fork Actually, "only a short list of async-signal-safe library routines" [1, 2]. Practically all of POSIX is off-limits after fork in a multithreaded program. [1] http://pubs.opengroup.org/onlinepubs/9699919799/functions/fork.html [2] http://pubs.opengroup.org/onlinepubs/9699919799/functions/pthread_atfork.html -- Open source business process management suite built on Java and Eclipse Turn processes into business applications with Bonita BPM Community Edition Quickly connect people, data, and systems into organized workflows Winner of BOSSIE, CODIE, OW2 and Gartner awards http://p.sf.net/sfu/Bonitasoft ___ Scikit-learn-general mailing list Scikit-learn-general@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/scikit-learn-general