🌐 Mailman 3 August 2018 - scikit-learn - python.org

Smoke and Metamorphic Testing of scikit-learn
by Steffen Herbold Aug. 27, 2018

Aug. 27, 2018

Dear developers, I am writing you because I applied an approach for the automated testing of classification algorithms to scikit-learn and would like to forward the results to you. The approach is a combination of smoke testing and metamorphic testing. The smoke tests try to find problems by executing the training and prediction functions of classifiers with different data. These smoke tests should ensure the basic functioning of classifiers. I defined 20 different data sets, some very simple (uniform features in [0,1]), some with extreme distributions, e.g., data close to machine precision. The metamorphic tests determine if classification results change as expected if the training data is modified, e.g., by reordering features, flipping class labels, or reordering instances. I generated 70 different Python unittest tests for eleven different scikit-learn classifiers. In summary, I found the following potential problems: - Two errors due to possibly infinite loops for the LogisticRegressionClassifier for data that approaches MAXDOUBLE. - The classification of LogisticRegression, MLPClassifier, QuadraticDiscriminantAnalysis, and SVM with a polynomial kernel changed if one is added to each feature value. - The classification of DecisionTreeClassifier, LogisticRegression, MLPClassifier, QuadraticDiscriminantAnalysis, RandomForestClassifier, and SVM with a linear and a polynomial kernel were not inverted when all binary class labels are flipped. - The classification of LogisticRegression, MLPClassifier, QuadraticDiscriminantAnalysis, and RandomForestClassifier sometimes changed when the features are reordered. - The classification of KNeighborsClassifier, MLPClassifier, QuadraticDiscriminantAnalysis, RandomForestClassifier, and SVM with a linear kernel sometimes changed when the instances are reordered. You can find details of our results online [1]. The provided resources include the current draft of the paper that describes the tests as well as detailed results in detail. Moreover, we provide an executable test suite with all tests we executed, as well as the export of our test results as XML file that contains all details of the test execution, including stack traces in case of exceptions. The preprint and online materials also contain the results for two other machine learning libraries, i.e., Weka and Spark MLlib. Additionally, you can find the atoml tool used to generate the tests on GitHub [2]. I hope that these tests may help with the future development of scikit-learn. You could help me a lot by answering the following questions: - Do you consider the tests helpful? - Do you consider any source code or documentation changes due to our findings? - Would you be interested in a pull request or any other type of integration of (a subset of) the tests into your project? - Would you be interested in more such tests, e.g., for the consideration of hyper parameters, other algorithm types like clustering, or more complex algorithm specific metamorphic tests? I am looking forward to your feedback. Best regards, Steffen Herbold [1] http://user.informatik.uni-goettingen.de/~sherbold/atoml-results/ [2] https://github.com/sherbold/atoml -- Dr. Steffen Herbold Institute of Computer Science University of Goettingen Goldschmidtstraße 7 37077 Göttingen, Germany mailto. herbold(a)cs.uni-goettingen.de tel. +49 551 39-172037

2 3

Regarding Contribution to scikit-learn
by Adwait Bauskar Aug. 24, 2018

Aug. 24, 2018

Hello everybody, I am a university student with a major in Mathematics and Computer Science interested in contributing to scikit-learn. I do not have any previous experience in open source development. I am willing to contribute to scikit-learn on a regular basis. I am looking for a mentor who would be ready to guide me in contributing. I do not need any help in setting up the project or any GitHub related things. I need help mainly in understanding the current code and a guide to help me along with my code. I have experience with Java and C++ before. If any developer is interested and has the time to guide me, please feel free to message me on this email. Also, I am sorry if this is not the right channel to contact developers on. Thank you, Adwait

1 0

Using GPU in scikit learn
by hoang trung Ta Aug. 21, 2018

Aug. 21, 2018

Dear all members, I am using Random forest for classification satellite images. I have a bunch of images, thus the processing is quite slow. I searched on the Internet and they said that GPU can accelerate the process. I have GPU NDVIA Geforce GTX 1080 Ti installed in the computer Do you know how to use GPU in Scikit learn, I mean the packages to use and sample code that used GPU in random forest classification? Thank you very much -- *Ta Hoang Trung (Mr)* *Master student* Graduate School of Life and Environmental Sciences University of Tsukuba, Japan Mobile: +81 70 3846 2993 Email : ta.hoang-trung.xm(a)alumni.tsukuba.ac.jp tahoangtrung(a)gmail.com s1626066(a)u.tsukuba.ac.jp *----* *Mapping Technician* Department of Surveying and Mapping Vietnam No 2, Dang Thuy Tram street, Hanoi, Viet Nam Mobile: +84 1255151344 Email : tahoangtrung(a)gmail.com

6 5

GMM.fit and GaussianMixture.fit()
by Dixeena Lopez Aug. 14, 2018

Aug. 14, 2018

Dear Sir/Madam, I have used GMM.fit() instead of GaussianMixture.fit() and got different answers. Please gives the advantage and disadvantage of these two. Please reply fast ‌Diixeena

2 1

Unable to connect HDInsight hive to python
by Debabrata Ghosh Aug. 12, 2018

Aug. 12, 2018

Hi All, Greetings ! Wish you are doing good ! I am just reaching out to you in case if you have any answer or help me direct to the right forum please: We are facing with hive connectivity from the python on Azure HDinsights, We have installed required SASL,thrift_sasl(0.2.1) and Thirft (0.9.3) packages on Ubuntu , but some how when we are trying to connect Hive using following packages we are getting errors , It would be really great help if you could provide some pointers based on your experience Example 1: from impala.dbapi import connect conn=connect(host="localhost", port=10001 , auth_mechanism="PLAIN", user="admin", password="PWD") (tried both 127.0.0.1:10000/10001) Example 2: import pyhs2 conn = pyhs2.connect(host='localhost ', port=10000,authMechanism="PLAIN", user='admin', password=,database='default') Example 3: from pyhive import hive conn = hive.Connection(host="localhost", port=10001, username="admin", password=None, auth='NONE') Across all of the above examples we are getting the error message: thrift.transport.TTransport.TTransportException: Tsocket read 0 bytes Thanks, Debu

2 1

DBSCAN
by Prathusha Jonnagaddla Subramanyam Naidu Aug. 10, 2018

Aug. 10, 2018

Hi everyone, I'm trying to cluster 14000 samples using DBSCAN and want to know if there is a way to display the index of each data point along with it's label. I'm only able to access labels in the form of a list . When I look at the graph and see outliers (black points) , I'm not able to pinpoint as to which image/data that particular point belongs to. Thank you -- Regards, Prathusha JS Naidu

1 0

回覆： Using GPU in scikit learn
by blacklabel29 Aug. 9, 2018

Aug. 9, 2018

Hi, the scikit-learn random forest does not support GPUs. If you want to do image classification using GPU processing, the standard way in this day and age is to use a neural network library like TensorFlow/keras or pytorch. GPUs can be faster than CPUs when the task is SIMD (single instruction multiple data), meaning the same calculation is done many times just on different datapoints. Neural networks are well-suited for such an architecture, decision trees not so much (even though there have been attempts to speed up decision trees using GPUs). So my advice to you depends on how much time you have: If you are willing to invest time to learn about neural networks and the aforementioned libraries, then that is certainly a very valuable skill, especially when looking for a job later on. But if you just need to get your paper done as soon as possible, stick with random forest. Greetings,Patrick 從我的 Samsung Galaxy 智慧型手機傳送。-------- 原始訊息 --------自： hoang trung Ta <tahoangtrung(a)gmail.com> 日期： 2018/8/9 02:50 (GMT+01:00) 至： scikit-learn(a)python.org 主旨： [scikit-learn] Using GPU in scikit learn Dear all members, I am using Random forest for classification satellite images. I have a bunch of images, thus the processing is quite slow. I searched on the Internet and they said that GPU can accelerate the process. I have GPU NDVIA Geforce GTX 1080 Ti installed in the computer Do you know how to use GPU in Scikit learn, I mean the packages to use and sample code that used GPU in random forest classification? Thank you very much -- Ta Hoang Trung (Mr) Master studentGraduate School of Life and Environmental SciencesUniversity of Tsukuba, Japan Mobile: +81 70 3846 2993Email : ta.hoang-trung.xm(a)alumni.tsukuba.ac.jp tahoangtrung(a)gmail.com s1626066(a)u.tsukuba.ac.jp---- Mapping Technician Department of Surveying and Mapping VietnamNo 2, Dang Thuy Tram street, Hanoi, Viet Nam Mobile: +84 1255151344Email : tahoangtrung(a)gmail.com

2 1

Fwd: BIC using GMM.fit and GaussianMixture.fit()
by Dixeena Lopez Aug. 9, 2018

Aug. 9, 2018

---------- Forwarded message ---------- From: Dixeena Lopez <dixeenalopez(a)gmail.com> Date: 2 August 2018 at 23:53 Subject: BIC using GMM.fit and GaussianMixture.fit() To: scikit-learn(a)python.org Dear Sir/Madam, I have tried to fit the data using GaussianMixture.fit() and GMM.fit() and calculated the BIC score. The BIC score value and the number of clusters I got using each method is different. Do you have any idea? ‌ Sincerely, Dixeena <https://mailtrack.io/> Sent with Mailtrack <https://mailtrack.io?utm_source=gmail&utm_medium=signature&utm_campaign=sig…>

1 0

RidgeCV and cv=LeaveOneGroupOut
by Christophe Pallier Aug. 8, 2018

Aug. 8, 2018

Hello, I do not manage to find out how to use a LeaveOneGroupOut crossvalidation stratergy with RidgeCV. The problem is that I do not see how to pass the groups mapping. Any hint? The approach by GridSearch is way to slow to be practical. Thanks in advance! PS: Thanks Alex for the response to my previous question. By the way, the context is fMRI data, so I am running the Ridge regression in about 200.000 voxels. -- -- Christophe Pallier <christophe(a)pallier.org> INSERM-CEA Cognitive Neuroimaging Lab, Neurospin, bat 145, 91191 Gif-sur-Yvette Cedex, France Tel: 00 33 1 69 08 79 34 Personal web site: http://www.pallier.org Lab web site: http://www.unicog.org

2 1

Decision trees with only pure leaves
by Pranav Ashok Aug. 8, 2018

Aug. 8, 2018

I am trying to use scikit-learn for building decision trees representing fully defined many-to-one functions. i.e. f(x1, x2, ..., xn) = f(x1', x2', ..., xn') if and only if x1 = x1', x2 = x2' and so on. In such a scenario, it seems clear that it is possible to construct trees which have pure leaves as long as min_samples_split = 2 and I am not setting any other parameters which might stop the splitting. My question is whether the decision tree builder in scikit-learn can indeed give me a perfect representation, i.e where all leaf nodes are pure. This would imply that tree.predict(x) = f(x). Thanks and best regards.

1 0