🌐 Mailman 3 August 2017 - scikit-learn - python.org

any interest in incorporating a new Transformer?
by Michael Capizzi Aug. 22, 2017

Aug. 22, 2017

Hi all - Forgive me if this is the wrong place for posting this question, but I'd like to inquire about the community's interest in incorporating a new Transformer into the code base. This paper ( https://nlp.stanford.edu/pubs/sidaw12_simple_sentiment.pdf ) is a "classic" in Natural Language Processing and is often times used as a very competitive baseline. TL;DR it transforms a traditional count-based feature space into the conditional probabilities of a `Naive Bayes` classifier. These transformed features can then be used to train any linear classifier. The paper focuses on `SVM`. The attached notebook has an example of the custom `Transformer` I built along with a custom `Classifier` to utilize this `Transformer` in a `multiclass` case (as the feature space transformation differs depending on the label). If there is interest in the community for the inclusion of this `Transformer` and `Classifier`, I'd happily go through the official process of a `pull-request`, etc. -Michael

2 4

How can i write the birch prediction results to the file
by Sema Atasever Aug. 22, 2017

Aug. 22, 2017

Dear scikit-learn developers, I have a text file where the columns represent the 22 features and the rows represent the amino asid . (you can see in the attachment) I want to apply hierarchical clustering to this database usign *sklearn.cluster.Birch algorithm.* There are too many prediction results and it is not possible to see them on the screen. How can i write the birch prediction results to the file? I would appreciate if you could advise on some methods. Thanks. *Birch Codes:* from sklearn.cluster import Birch import numpy as np X=np.loadtxt(open("C:\class1.txt", "rb"), delimiter=";") brc = Birch(branching_factor=50, n_clusters=None, threshold=0.5,compute_labels=True,copy=True) brc.fit(X) centroids = brc.subcluster_centers_ labels = brc.subcluster_labels_ n_clusters = np.unique(labels).size brc.predict(X) print("\n brc.predict(X)") print(brc.predict(X))

4 6

Malformed input for SVC(kernel='precomputed').predict()
by Sam Barnett Aug. 17, 2017

Aug. 17, 2017

I am rolling classifier based on SVC which computes a custom Gram matrix and runs this through the SVC classifier with kernel = 'precomputed'. While this works fine with the fit method, I face a dilemma with the predict method, shown here: def predict(self, X): """Run the predict method of the previously-instantiated SVM classifier, returning the predicted classes for test set X.""" # Check is fit had been called check_is_fitted(self, ['X_', 'y_']) # Input validation X = check_array(X) cut_off = self.cut_ord_pair[0] order = self.cut_ord_pair[1] X_gram = seq_kernel_free(X, self.X_, \ pri_kernel=kernselect(self.kernel, self.coef0, self.gamma, self.degree, self.scale), \ cut_off=cut_off, order=order) X_gram = np.nan_to_num(X_gram) return self.ord_svc_.predict(X_gram) This will run on any dataset just fine. However, it fails the check_estimator test. Specifically, when trying to raise an error for malformed input on predict (in check_classifiers_train), it says that a ValueError is not raised. Yet if I change the order of X and self.X_ in seq_kernel_free (which computes the [n_samples_train, n_samples_test] Gram matrix), it passes the check_estimator test yet fails to run the predict method. How do I resolve both issues simultaneously?

2 2

Categorical handling
by Georg Heiler Aug. 17, 2017

Aug. 17, 2017

Hi, how can I properly handle categorical values in scikit-learn? https://stackoverflow.com/questions/45727934/pandas-categories-new-levels?n… goals - scikit-learn syle fit/transform methods to encode labels of categorical features of X - should handle unseen labels - should be faster than running a label encoder manually for each fold and manually checking if the label already was seen in the training data i.e. what I currently do ( https://stackoverflow.com/questions/45727934/pandas-categories-new-levels?n… which links to https://gist.github.com/geoHeil/5caff5236b4850d673b2c9b0799dc2ce ) - only some columns are categorical, and only these should be converted Regards, Georg

3 3

caching transformers during hyper parameter optimization
by Georg Heiler Aug. 17, 2017

Aug. 17, 2017

There is a new option in the pipeline: http://scikit-learn.org/stable/modules/pipeline.html#pipeline-cache How can I use this to also store the transformed data as I only want to compute the last step i.e. estimator during hyper parameter tuning and not the transform methods of the clean steps? Is there a possibility to apply this for crossvalidation? I would want to see all the folds precomputed and stored to disk in a folder. Regards, Georg

2 3

Question-Early Stopping MLPClassifer RandomizedSearchCV
by fabian.sippl＠gmx.net Aug. 14, 2017

Aug. 14, 2017

2 1

No module named crluster.hierarchical
by Ariani A Aug. 13, 2017

Aug. 13, 2017

Dear all, I am writing this import: from sklearn.crluster.hierarchical import (_hc_cut, _TREE_BUILDERS, linkage_tree) But it gives this error: ImportError: No module named crluster.hierarchical Any clue? Best regards, -Noushin

2 2

Truncated svd not working for complex matrices
by André Melo Aug. 12, 2017

Aug. 12, 2017

Hello all, I'm trying to use the randomized version of scikit-learn's TruncatedSVD (although I'm actually calling the internal function randomized_svd to get the actual u, s, v matrices). While it is working fine for real matrices, for complex matrices I can't get back the original matrix even though the singular values are exactly correct: >>> import numpy as np >>> from sklearn.utils.extmath import randomized_svd >>> N = 3 >>> a = np.random.rand(N, N)*(1 + 1j) >>> u1, s1, v1 = np.linalg.svd(a) >>> u2, s2, v2 = randomized_svd(a, n_components=N, n_iter=7) >>> np.allclose(s1, s2) True >>> np.allclose(a, u1.dot(np.diag(s1)).dot(v1)) True >>> np.allclose(a, u2.dot(np.diag(s2)).dot(v2)) False Any idea what could be wrong? Thank you! Best regards, Andre Melo

7 8

scikit-learn 0.19.0 is out!
by Olivier Grisel Aug. 12, 2017

Aug. 12, 2017

Grab it with pip or conda ! Quoting the release highlights from the website: We are excited to release a number of great new features including neighbors.LocalOutlierFactor for anomaly detection, preprocessing.QuantileTransformer for robust feature transformation, and the multioutput.ClassifierChain meta-estimator to simply account for dependencies between classes in multilabel problems. We have some new algorithms in existing estimators, such as multiplicative update in decomposition.NMF and multinomial linear_model.LogisticRegression with L1 loss (use solver='saga'). Cross validation is now able to return the results from multiple metric evaluations. The new model_selection.cross_validate can return many scores on the test data as well as training set performance and timings, and we have extended the scoring and refit parameters for grid/randomized search to handle multiple metrics. You can also learn faster. For instance, the new option to cache transformations in pipeline.Pipeline makes grid search over pipelines including slow transformations much more efficient. And you can predict faster: if you’re sure you know what you’re doing, you can turn off validating that the input is finite using config_context. We’ve made some important fixes too. We’ve fixed a longstanding implementation error in metrics.average_precision_score, so please be cautious with prior results reported from that function. A number of errors in the manifold.TSNE implementation have been fixed, particularly in the default Barnes-Hut approximation. semi_supervised.LabelSpreading and semi_supervised.LabelPropagation have had substantial fixes. LabelPropagation was previously broken. LabelSpreading should now correctly respect its alpha parameter. Please see the full changelog at: http://scikit-learn.org/0.19/whats_new.html#version-0-19 Notably some models have changed behaviors (bug fixes) and some methods or parameters part of the public API have been deprecated. A big thank you to anyone who made this release possible and Joel in particular. -- Olivier

9 8

Overflow Error with Cross-Validation (but not normally fitting the data)
by Sam Barnett Aug. 11, 2017

Aug. 11, 2017

To all, I am working on a scikit-learn estimator that performs a version of SVC with a custom kernel. Unfortunately, I have been presented with a problem: when running a grid search (or even using the cross_val_score function), my estimator encounters an overflow error when evaluating my kernel (specifically, in an array multiplication operation). What is particularly strange about this is that, when I train the estimator on the whole dataset, this error does not occur. In other words: the problem only appears to occur when the data is split into folds. Is this something that has been seen before? How ought I fix this? I have attached the source code below (in particular, see the notebook for how the problem arises). Best, Sam

1 0