Mailman 3 September 2016 - scikit-learn

Github project management tools
by Joel Nothman Dec. 3, 2016

Dec. 3, 2016

One of the biggest issues with scikit-learn as a project is managing its backlog of issues; another is release scheduling. Some of this cannot be fixed as long as our model of voluntary contribution (with a couple of important exceptions) does not change. However, it may be worth considering the new project management features in Github. At the moment we have the following management: * labels corresponding to type (bug, enhancement, new feat, question), scope (API, Build/CI, ?Large Scale, Documentation), difficulty (easy, moderate), status/scheduling (needs contributor, needs review, sprint). * PR status management with title prefixes [WIP], [MRG], [MRG+1], [MRG+2] Firstly, we might benefit from prefixing labels by category, i.e. difficulty:easy so that complementary labels appear together. In truth, PRs have roughly these statuses: * WIP (not ready for review) * waiting for review * waiting for changes (with or without one of the following) * in dispute (i.e. fundamental doubts about the PR) * the above together with 1 or 2 "official" approvals * ready for merge (pending minor changes such as what's new documentation) New github features: * reviews with "approved" or "request changes". A list of approvers can be found in the merge/CI panel. We could replace the MRG+1 annotation with this and use it to track disputation too. I'm not sure how it works with changes that are added after approval. I think it would have avoided one improper merge by me... One downside is that there does not yet seem to be a way to search for PRs with a specified level of approval (while searching for "MRG+1" sort-of works). * Milestone prioritising: issues in a milestone, such as https://github.com/scikit-learn/scikit-learn/milestone/21, can be ranked with drag-and-drop. I think this could help with release scheduling as it would allow us to identify the top priorities for a release and see when enough of them are completed. * The Kanban-style workflow management of the new Projects tool https://github.com/scikit-learn/scikit-learn/projects is another way of managing status and, I think, priority, for a small set of related issues. This might be an alternative way of managing milestone scope, or of working towards big changes like the one just completed for model selection; like proposed expansions to get_feature_names expansion; like estimator tags; making utilities public/private... So with the goal of making it easier to track where attention is most needed, and when to move to release: What's worth trying?

13 37

ANN Scikit-learn 0.18 released
by Andreas Mueller Oct. 11, 2016

Oct. 11, 2016

Hi everybody. I'm happy to announce scikit-learn 0.18 has been released today. You can install from pipy or anaconda.org: pip install --upgrade scikit-learn --no-deps or if you prefer conda: conda update scikit-learn A big thank you to everybody who contributed. This one took us a while, but I think it's worth the wait. Highlights include: - A new GaussianProcessClassifier and GaussianProcessRegressor to learn complex kernels! - A much improved GaussianMixture and BayesianGaussianMixture mixture models. - We moved the content of the grid_search, cross_validation and validation_curve modules to the new model_selection module. - A Multi-layer perceptron. and soo much more that it's impossible to summarize. Check out the full changelog here: http://scikit-learn.org/stable/whats_new.html#version-0-18 Please don't update a (ana)conda installation using pip, as that might lead to problems. Let us know any issues you have on the issue tracker: https://github.com/scikit-learn/scikit-learn/issues Enjoy! Andy

10 13

understanding lasso performance ?
by Yoel Sanchez Araujo Sept. 30, 2016

Sept. 30, 2016

Hi all, Today I updated to the latest release of scikit-learn, and I went to test out the LassoCV module in linear_model. I've tried both approaches below, and my accuracy seems very poor, while using the same exact data with glmnet in R for example will give me ~ 75% accuracy: from sklearn import linear_model from sklearn.model_selection import StratifiedKFold, train_test_split lassocv1 = linear_model.LassoCV(cv=10, max_iter=10000, n_alphas=10000) xtrain, xtest, ytrain, ytest = train_test_split( endo_Xv, endo_y, test_size = .25, random_state = 1 ) lassocv1.fit(xtrain, ytrain) lassocv1.score(xtest, ytest) from this, lassocv1.coef_ returns all zero coefficients I've also tried this: k_fold_S = StratifiedKFold(n_splits=10, shuffle=False) lasso_cv = linear_model.LassoCV() alphas=[] scores=[] coefs=[] ks=[] for k, (train, test) in enumerate(k_fold_S.split(endo_Xv, endo_y)): lasso_cv.fit(endo_Xv[train], endo_y[train]) scores.append(lasso_cv.score(endo_Xv[test], endo_y[test])) alphas.append(lasso_cv.alpha_) coefs.append(lasso_cv.coef_) ks.append(k) for all k, the coef_ arrays are all zero and the scores array for example: [-1.3295256159340241e-05, -1.3295256159562285e-05, -1.3295256159784328e-05, -1.3295256159562285e-05, -1.3295256159562285e-05, -1.3295256159340241e-05, -6.4162287406910323e-05, -6.4162287406910323e-05, -6.4162287406910323e-05, -3.8436343168246623e-06]) Any insights would be greatly appreciated, not sure if this has anything to do with the update, but yesterday(unupdated) I was getting better performance.

1 1

nmf with multiple cores
by Roberto Pagliari Sept. 30, 2016

Sept. 30, 2016

I’m running nmf on a machine with 16 cores. Is there an option to run nmf with multithreading? I know numpy does, but all I see is one single process with 100% CPU usage. Thanks,

2 1

Question about Python's L2-Regularized Logistic Regression
by Kristen M. Altenburger Sept. 29, 2016

Sept. 29, 2016

Hi All, I am trying to understand Python’s code [function ‘_fit_liblinear' in https://github.com/scikit-learn/scikit-learn/blob/master/sklearn/svm/base.py] for fitting an L2-logistic regression for a ‘liblinear’ solver. More specifically, my [approximately balanced class] dataset is such that the # of predictors [p=2000] >> # of observations [n=100]. Therefore, I am currently confused that when I increase C [and thus decrease the regularization strength] in fitting the logistic regression model to my training data why I then still obtain high AUC results when the model is then applied to my testing data. Is python internally doing a feature selection when fitting this model for high C values? Or why is it that the almost unregularized model [high C values] versus regularized [cross-validated approach to selecting C] model both result in similar AUC and accuracy results when the model is applied to the testing data? Should I be coding my predictors as +1/-1? Any pointers/explanations would be much appreciated! Thanks, Kristen

3 2

PyData cook book
by Andreas Mueller Sept. 29, 2016

Sept. 29, 2016

Hey all. Numfocus and some folks are working on a PyData cookbook: https://github.com/pydata/pydata-cookbook Unfortunately so far it has no scikit-learn yet. Is anyone interested in contributing? I think I did enough writing for a while ;) Andy Sent from phone. Please excuse spelling and brevity.

3 2

Re: [scikit-learn] ANN Scikit-learn 0.18 released
by Jiří Fejfar Sept. 29, 2016

Sept. 29, 2016

Thank you very much for the new version of this great SW! J. Dne 29.9.2016 15:55 napsal uživatel "Raghav R V" <ragvrv(a)gmail.com>: Congrats everyone :) On 29 Sep 2016 9:41 a.m., "Jaques Grobler" <jaquesgrobler(a)gmail.com> wrote: > Congrats everyone! > > 2016-09-29 8:39 GMT+02:00 bthirion <bertrand.thirion(a)inria.fr>: > >> Congrats ! >> >> Bertrand >> >> >> On 29/09/2016 07:28, Gael Varoquaux wrote: >> >>> Hurray! >>> >>> Congratulations to everybody, and in particular the release time! >>> >>> Gaël >>> >>> On Wed, Sep 28, 2016 at 05:01:45PM -0400, Andreas Mueller wrote: >>> >>>> Hi everybody. >>>> I'm happy to announce scikit-learn 0.18 has been released today. >>>> You can install from pipy or anaconda.org: >>>> pip install --upgrade scikit-learn --no-deps >>>> >>> >>> or if you prefer conda: >>>> conda update scikit-learn >>>> >>> >>> A big thank you to everybody who contributed. >>>> This one took us a while, but I think it's worth the wait. >>>> Highlights include: >>>> - A new GaussianProcessClassifier and GaussianProcessRegressor to learn >>>> complex kernels! >>>> - A much improved GaussianMixture and BayesianGaussianMixture mixture >>>> models. >>>> - We moved the content of the grid_search, cross_validation and >>>> validation_curve modules to the new model_selection module. >>>> - A Multi-layer perceptron. >>>> and soo much more that it's impossible to summarize. >>>> Check out the full changelog here: >>>> http://scikit-learn.org/stable/whats_new.html#version-0-18 >>>> Please don't update a (ana)conda installation using pip, as that might >>>> lead >>>> to problems. >>>> Let us know any issues you have on the issue tracker: >>>> https://github.com/scikit-learn/scikit-learn/issues >>>> Enjoy! >>>> Andy >>>> _______________________________________________ >>>> scikit-learn mailing list >>>> scikit-learn(a)python.org >>>> https://mail.python.org/mailman/listinfo/scikit-learn >>>> >>> >> >> _______________________________________________ >> scikit-learn mailing list >> scikit-learn(a)python.org >> https://mail.python.org/mailman/listinfo/scikit-learn >> > > > _______________________________________________ > scikit-learn mailing list > scikit-learn(a)python.org > https://mail.python.org/mailman/listinfo/scikit-learn > > _______________________________________________ scikit-learn mailing list scikit-learn(a)python.org https://mail.python.org/mailman/listinfo/scikit-learn

3 2

always Squash and Merge?
by Andreas Mueller Sept. 28, 2016

Sept. 28, 2016

Hey. This is a continuation of the discussion we had on squashing in June: https://mail.python.org/pipermail/scikit-learn/2016-June/000121.html I thought we discussed this again after the "squash and merge" feature was introduced, but I couldn't find the thread. I think Joel, me and some others where recently using the github "squash and merge" feature, which I think is great. It removes burden from the contributors and makes for a "clean" (or fake) history. I like it because it makes cherry-picking easy and allows a pretty simple analysis of what's happening. When doing some backports, I realized that some people (including Gael) didn't use it. Is there a reason not to use squash and merge? Should we make it policy? The one case where I think we might not want it is in case there are multiple authors in a PR. Other than that, I don't really see a downside. Wdyt? Andy

4 5

Re: [scikit-learn] scikit-learn Digest, Vol 6, Issue 40
by Afarin Famili Sept. 28, 2016

Sept. 28, 2016

Hi David, When applying Train_test_split to the sample space, we have a single row per subject. I am looking for some other function like Train_test_split that can deal with pairs of rows (for each subject), which does not lead to a biased accuracy. We are studying memory and have a row of features for successful memory encoding, and a second row for unsuccessful memory encoding in each of the subjects. Our target space being 1 for successful and 0 for unsuccessful encoding respectively. How do you recommend me to split this set of data in order to get a reasonable/unbiased accuracy? Thanks, Afarin ________________________________________ From: scikit-learn <scikit-learn-bounces+afarin.famili=utsouthwestern.edu(a)python.org> on behalf of scikit-learn-request(a)python.org <scikit-learn-request(a)python.org> Sent: Monday, September 26, 2016 2:43 PM To: scikit-learn(a)python.org Subject: scikit-learn Digest, Vol 6, Issue 40 Send scikit-learn mailing list submissions to scikit-learn(a)python.org To subscribe or unsubscribe via the World Wide Web, visit https://mail.python.org/mailman/listinfo/scikit-learn or, via email, send a message with subject or body 'help' to scikit-learn-request(a)python.org You can reach the person managing the list at scikit-learn-owner(a)python.org When replying, please edit your Subject line so it is more specific than "Re: Contents of scikit-learn digest..." Today's Topics: 1. header intact (Afarin Famili) 2. Is there a built-in function for pairs of data? (Afarin Famili) 3. Re: Is there a built-in function for pairs of data? (Pedro Pazzini) 4. Re: Is there a built-in function for pairs of data? (David Nicholson) 5. Large computation time for homogeneous data with agglomerative clustering (Md. Khairullah) ---------------------------------------------------------------------- Message: 1 Date: Mon, 26 Sep 2016 18:03:27 +0000 From: Afarin Famili <Afarin.Famili(a)UTSouthwestern.edu> To: "scikit-learn(a)python.org" <scikit-learn(a)python.org> Subject: [scikit-learn] header intact Message-ID: <1474913007611.80841(a)UTSouthwestern.edu> Content-Type: text/plain; charset="iso-8859-1" ? ________________________________ UT Southwestern Medical Center The future of medicine, today.

5 4

Fwd: [Scikit-learn-general] MultinomialNB Scikit-learn question
by Bharat Didwania 4-Yr B.Tech. Electrical Engg. Sept. 27, 2016

Sept. 27, 2016

---------- Forwarded message ---------- From: Bharat Didwania 4-Yr B.Tech. Electrical Engg. < bharat.didwania.eee14(a)itbhu.ac.in> Date: Mon, Sep 26, 2016 at 11:04 PM Subject: Re: [Scikit-learn-general] MultinomialNB Scikit-learn question To: scikit-learn-general(a)lists.sourceforge.net It seems you are using the sample_weights with OneVsAll classifier. I think it should be an issue as the base classifier, i.e MultinomialNB supports sample_weights but OneVsAll does not. Sent with Mailtrack <https://mailtrack.io/install?source=signature&lang=en&referral=bharat.didwa…> On Mon, Sep 26, 2016 at 4:10 PM, Diego Vergara <reiby.viper(a)gmail.com> wrote: > Hi Scikit-learn developer team. I have a query in which I need help. > > How does it work sample weight in MultinomialNB ?, Is there any > documentation or equation?. > > In the original equation (http://scikit-learn.org/stabl > e/modules/naive_bayes.html#multinomial-naive-bayes), weights applied to > individual samples, throws me bad results. > Best regards, thank you for your answers. > > ------------------------------------------------------------ > ------------------ > > _______________________________________________ > Scikit-learn-general mailing list > Scikit-learn-general(a)lists.sourceforge.net > https://lists.sourceforge.net/lists/listinfo/scikit-learn-general > > Sent with Mailtrack <https://mailtrack.io/install?source=signature&lang=en&referral=bharat.didwa…>

1 0