Mailman 3 February 2017 - scikit-learn

Control over the inner loop in GridSearchCV
by Ludovico Coletta Feb. 27, 2017

Feb. 27, 2017

Dear Scikit experts, we am stucked with GridSearchCV. Nobody else was able/wanted to help us, we hope you will. We are analysing neuroimaging data coming from 3 different MRI scanners, where for each scanner we have a healthy group and a disease group. We would like to merge the data from the 3 different scanners in order to classify the healthy subjects from the one who have the disease. The problem is that we can almost perfectly classify the subjects according to the scanner (e.g. the healthy subjects from scanner 1 and scanner 2). We are using a custom cross validation schema to account for the different scanners: when no hyper-parameter (SVM) optimization is performed, everything is straightforward. Problems arise when we would like to perform hyperparameter optimization: in this case we need to balance for the different scanner in the optimization phase as well. We also found a custom cv schema for this, but we are not able to pass it to GridSearchCV object. We would like to get something like the following: pipeline = Pipeline([('scl', StandardScaler()), ('sel', RFE(estimator,step=0.2)), ('clf', SVC(probability=True, random_state=42))]) param_grid = [{'sel__n_features_to_select':[22,15,10,2], 'clf__C': np.logspace(-3, 5, 100), 'clf__kernel':['linear']}] clf = GridSearchCV(pipeline, param_grid=param_grid, verbose=1, scoring='roc_auc', n_jobs= -1) # cv_final is the custom cv for the outer loop (9 folds) ii = 0 while ii < len(cv_final): # fit and predict clf.fit(data[?]], y[[?]]) predictions.append(clf.predict(data[cv_final[ii][1]])) # outer test data ii = ii + 1 We tried almost everything. When we define clf in the loop, we pass the -ith cv_nested as cv argument, and we fit it on the training data of the -ith custom_cv fold, we get an "Too many values to unpack" error. On the other end, when we try to pass the nested -ith cv fold as cv argument for clf, and we call fit on the same cv_nested fold, we get an "Index out of bound" error. Two questions: 1) Is there any workaround to avoid the split when clf is called without a cv argument? 2) We suppose that for hyperparameter optimization the test data is removed from the dataset and a new dataset is created. Is this true? In this case we only have to adjust the indices accordingly Thank your for your time and sorry for the long text Ludovico

2 1

GSOC call for mentors
by Andreas Mueller Feb. 21, 2017

Feb. 21, 2017

Hey all. It's that time of the year again. Are we planning on participating in GSOC? If so, we need mentors and projects. It's unlikely that I'll have time to help with either in any substantial way. If we want to participate, I think we should try to be a bit more organized than last year ;) Andy Sent from phone. Please excuse spelling and brevity.

7 9

Re: [scikit-learn] Request for single pass clustering algorithm
by Andreas Mueller Feb. 21, 2017

Feb. 21, 2017

Hi Gaurish. scikit-learn-owner is the email address for the mailing list administration. See the FAQ on contacting the project: http://scikit-learn.org/stable/faq.html#what-s-the-best-way-to-get-help-on-… For feature requests I would suggest the issue tracker. This sounds like "single pass" is the same as agglomerative clustering with the average linkage criterion. Or is it any different from that? Andy On 02/21/2017 04:39 AM, Gaurish Thakkar wrote: > > I would like to make a request to Scikit team to please implement and > incorporate the single pass clustering algorithm. It is one of the > most basic online algorithms and > the link below dicusses the process in detail. > [http://facweb.cs.depaul.edu/mobasher/classes/csc575/assignments/single-pass…] > > -- > /Regards:/ > Gaurish P Thakkar > >

1 0

can we have a slack team for scikit-learn
by SHUBHAM BHARDWAJ 15BCE0704 Feb. 21, 2017

Feb. 21, 2017

Hello Friends, I have tried Slack and its awesome. Things are more dynamic. I have faced some problems which I am sure slack can alleviate like- When working on some issue if I need some guidance I am not sure when I will get reply. That maybe usually within 2-3 days or more.Maybe some fellow programmer who is free can discuss and we may find a good solution. I think collaboration would be much better. Regards Shubham Bhardwaj

7 6

Confidence intervals on GaussianProcessRegressor hyperparameters estimates
by Vincent Dubourg Feb. 20, 2017

Feb. 20, 2017

Hi list, Did anyone ever considered using the Cramer-Rao lower bound estimate for the variance-covariance matrix of the GaussianProcess hyperparameters estimate? I have seen the gradient of the marginal log likelihood is already available. How about the hessian matrix? Looking at the theta values wrt to the one in the fitted kernel themselves, it looks like some normalization occurs, which is fine, though how do I get the true gradient back? Actually, I am more interested in infering the parameters rather than predicting. I have considered using pymc3 but MCMC is quite time expensive and I would like to be able to speed this with a reasonable approximation. George is also an alternative but out of the question since I am running Windows. Thank you, Vincent

1 0

Regarding scikit learn to take part in GSOC 2017
by Akshay Gupta Feb. 19, 2017

Feb. 19, 2017

Dear Programmers, I'm watching scikit learn on github from last few month and have also made some contribution. Just now I found that this year there is no final plan in community to take part in GSOC. Community like Scikit Learn which have a unique place in industry should promote open source contribution and must take part in events like GSOC. My appeal is that scikit learn should at least have a idea page.Though it is late but there still exist a chance to take part in GSOC. Regards Akshay GitHub Name - Akshay0724

2 1

OMP ended prematurely due to linear dependence in the dictionary
by Benjamin Merkt Feb. 18, 2017

Feb. 18, 2017

Hi everyone, I'm using OrthogonalMatchingPursuit to get a sparse coding of a signal using a dictionary learned by a KSVD algorithm (pyksvd). However, during the fit I get the following RuntimeWarning: /usr/local/lib/python2.7/dist-packages/sklearn/linear_model/omp.py:391: RuntimeWarning: Orthogonal matching pursuit ended prematurely due to linear dependence in the dictionary. The requested precision might not have been met. copy_X=copy_X, return_path=return_path) In those cases the results are indeed not satisfactory. I don't get the point of this warning as it is common in sparse coding to have an overcomplete dictionary an thus also linear dependency within it. That should not be an issue for OMP. In fact, the warning is also raised if the dictionary is a square matrix. Might this Warning also point to other issues in the application? Thanks, Ben

3 11

Modelling event rates
by Stuart Reynolds Feb. 17, 2017

Feb. 17, 2017

Does scikit provide any event-rate/time-to-event models, or other models that are specifically time-dependent? (e.g. models that output the # events per unit of time) Examples might include: Poisson model, or Cox proportional hazard. There was some discussion about pulling from statsmodels, https://github.com/scikit-learn/scikit-learn/issues/5975 but (AFAIK), this was not done.

2 1

Google Summer of code 2017
by Akshay Gupta Feb. 17, 2017

Feb. 17, 2017

Are we having any plans to take part in GSOC this year? If so then I would like to apply this year. Thanks

3 2

ANN: PyDataLondon Conference in May - Call for Proposals closing in 1 week
by Ian Ozsvald Feb. 17, 2017

Feb. 17, 2017

PyDataLondon 2017 runs in London this May 5-7th at Bloomberg's HQ near London Bridge. Our Call for Proposals is open until February 24th (next Friday), and I'd love to see sklearn talks and tutorial proposals: http://pydata.org/london2017/ This is our 4th annual conference, we'll have 330 active data scientists over the 3 days. Our conference builds on our 4,800+ member meetup which runs every month at hedge fund AHL: http://london.pydata.org/ I'd *love* to see a general sklearn tutorial at the conference, there's a real demand for this here in London. I'm also very interested in communicating complex data visually, applications of data science that "made a difference", data engineering and all the topics you'd expect at a strong data science conference. See last year's schedule if you'd like an idea of what to expect: http://pydata.org/london2016/schedule/ You may also be interested in PyDataAmsterdam (April 8-9th) and PyDataBerlin (June 30th- July 2nd), both have their CfP open at the moment: http://pydata.org/amsterdam2017/ http://pydata.org/berlin2017/ I'm hoping to see some interesting sklearn submissions, Ian (conference co-chair) ps. At our monthly meetups I'm also asking members to think on testimonials they could provide back to the sklearn testimonials page, I think that'll be a slow mission but I'll keep pushing the message. Hopefully a few companies will reciprocate to help with your grant applications -- Ian Ozsvald (Data Scientist, PyDataLondon co-chair) ian(a)IanOzsvald.com http://IanOzsvald.com http://ModelInsight.io http://twitter.com/IanOzsvald

1 0