mloss.org MLPY Machine Learning Py

MLPY Machine Learning Py 1.2.7

Tue, 11 Nov 2008 12:13:33 -0000

We introduce mlpy, a high-performance Python package for predictive modeling. It makes extensive use of NumPy to provide fast N-dimensional array manipulation and easy integration of C code. Mlpy provides high level procedures that support, with few lines of code, the design of rich Data Analysis Protocols (DAPs) for predictive classification and feature selection. Methods are available for feature weighting and ranking, data resampling, error evaluation and experiment landscaping. The package includes tools to measure stability in sets of ranked feature lists, of special interest in bioinformatics for functional genomics, for which large scale experiments with up to 10^6 classifiers have been run on Linux clusters and on the Grid.

The modular structure of mlpy allows easily adding new algorithms to each of the 7 categories in which the package is organized. They are:

Classification. For each algorithm, distinct methods are deployed for the training and the testing phases (whenever possible, real valued prediction can be obtained). The implemented algorithms are in the families of SVMs-Support Vector Machines (four kernels available), DA-Discriminant Analysis (Fisher, Penalized and Spectral Regression) and Nearest Neighbours.

Feature weighting. A total of nine methods is made available to obtain weights from models such as SVMs or DAs; classifier-independent methods for weighting features are also implemented, including I-RELIEF and Discrete Wavelet Transform.

Feature ranking. Two main schemas are used for selecting and ranking purposes, belonging either to the Recursive Feature Elimination or the Recursive Forward Selection family (for a total of six variants).

Resampling methods. The classification and feature ranking operations can be organized within a sampling procedure such as Textbook/Monte-Carlo cross validation (stratification over labels is available), leave-one-out or user-defined train/test split schema.

Metric functions. Performance assessment can be evaluated by a set of different measures, including Error, Accuracy, Matthews Correlation Coefficient, Area Under the ROC Curve. Variability can assessed by Standard Deviation or Bootstrap Confidence Intervals.

Feature list analysis. The ordered lists from the feature ranking experiments can be analyzed in terms of stability (Canberra indicator, extraction/position indicator) and an optimal list can be retrieved (Borda count).

Landscaping tools. A system of executable scripts to be used off-the-shelf to tabulate performance (e.g. Error, MCC and stability measures) on a grid of different experimental conditions by a basic DAP implementation (resampling by k-fold or Monte Carlo CV, training, feature ranking, test).

mlpy is a project developed by the MPBA research unit at FBK, the Bruno Kessler Foundation in Trento, Italy (http://mpba.fbk.eu).

Comment by Cheng Soon Ong on 2007-10-12 16:11

Cheng Soon Ong — Fri, 12 Oct 2007 16:11:23 -0000

This was presented at NIPS 2006 MLOSS workshop.

Comment by mseeger on 2007-10-12 16:42

mseeger — Fri, 12 Oct 2007 16:42:34 -0000

The source tarballs feature a complicated configure-make system, which is too clumsy for (my) day-to-day work. Also, a significant amount of code is not in there yet for various reasons, mostly because I am not happy with its stability. I use a simpler make system, which can easily be configured to a new system, but this does not happen automatically.

If you are serious about using LHOTSE for your own development, please get in touch, and I can provide you with an extension containing the simpler make system, classes to make MEX file implementation easier, and other things. I can also hoffer limited help with setting it up (on Linux), as far as I find time. I will put such an extension up at some point anyway, but this may take some time if there is no real interest.

If you have questions about the system and its design, please do not hesitate. Interest by other developers may also convince me that writing a general developers documentation is worth the effort. In the meantime, all the code is well documented, and there is a lengthy text file telling you about the most important issues.

I would of course very happy about receiving constructive criticism, ports to other systems (say, WINDOWS), etc.

Comment by mseeger on 2007-10-22 17:00

mseeger — Mon, 22 Oct 2007 17:00:56 -0000

I have added the simpler LHOTSE standard make system to the distribution (version 0.14). It is in the subdirectory old. doc/simple-make-system.txt tells you how to set it up for your machine. Also check the comments in old/Makefile.

Comment by Gorden Jemwa on 2007-11-24 21:19

Gorden Jemwa — Sat, 24 Nov 2007 21:19:30 -0000

Excellent object-oriented design within MATLAB which provides for a intuitive interface for the beginner and a flexible extensible framework for advanced users. Some of the implementations have default values that could be problematic; for example a hard-margin SVM classifier is assumed (C=Inf), which may lead to the program stalling for real-world data sets. Also, I've had problems with the default mex optimizer interface to LIBSVM when using the one-class SVM. Changing the optimizer with the matlab interface from LIBSVM website solved the problem (one needs to extend (recommended) or modify the associated training.m accordingly). Overall, I think this is a superb effort from Jason, Goekhan, Andre and the rest of the developers in designing this environment. Most other algorithms one may need from other sources can easily be incorporated seamlessly into spider. WELL DONE!

Comment by Gorden Jemwa on 2007-12-01 16:34

Gorden Jemwa — Sat, 01 Dec 2007 16:34:06 -0000

I've tried installing the software but running into problems; perhaps someone can help. Below is what I do and what I get. When I run the setup.py build command the installer complains of an unnamed module numpy but I already have the latest installed (see next output from setup-deps.py).

$ /usr/bin/python2.5 setup-deps.py

 Reading package lists...

Building dependency tree... python-wxgtk2.8 is already the newest version. 0 upgraded, 0 newly installed, 0 to remove and 0 not upgraded.

Reading package lists... Building dependency tree... python-matplotlib is already the newest version. 0 upgraded, 0 newly installed, 0 to remove and 0 not upgraded.

Reading package lists... Building dependency tree... python-numpy is already the newest version. 0 upgraded, 0 newly installed, 0 to remove and 0 not upgraded.

Reading package lists... Building dependency tree... python-scipy is already the newest version. 0 upgraded, 0 newly installed, 0 to remove and 0 not upgraded. pythonPackageDir :

...

Check installation : wx

Check installation : matplotlib

Check installation : numpy

Check installation : scipy

All modules successfully installed.

Comment by Gorden Jemwa on 2007-12-01 19:41

Gorden Jemwa — Sat, 01 Dec 2007 19:41:01 -0000

Please ignore my earlier post. I'm referring it to the more appropriate elefant-users forum.

Comment by Erna Maier on 2008-01-07 19:05

Erna Maier — Mon, 07 Jan 2008 19:05:55 -0000

RapidMiner is the most flexible and comprehensive data mining tool around.

Comment by Frank Hagemann on 2008-03-15 08:56

Frank Hagemann — Sat, 15 Mar 2008 08:56:13 -0000

RapidMiner offers an amazing functionality for rapidly desgning and automatically optimizing even large and nested and very complex data mining processes. It is a powerful tool for experts that want to get the most out of data mining in a reasonably short amount of time. RapidMiner covers the full data mining process from data loading over pre-processing, data mining process design, modelling, automated parameter optimization, automated feature selection and generation, evaluation, visualization, and deployment.

RapidMiner can be used interactively via its easy-to-use graphical user interface (GUI). The GUI supports breakpoints, online visualisations while optimization are running, graphics and results exports, etc.

RapidMiner can also be used on servers through its command line version.

RapidMiner can also be used as a data mining, text mining, machine learning, and predictive analytics library for your own programs. It is probably one of the most complete data mining libraries. It provides more than 400 data mining operations of its own plus the about 100 data mining operations of Weka, i.e. more than 500 in total.

From my point of view it is the best tool on the market.

Hope this review is helpful for somebody.

Best wishes, Frank

Comment by Yaroslav Halchenko on 2008-05-18 17:07

Yaroslav Halchenko — Sun, 18 May 2008 17:07:37 -0000

It is actively developed project at the moment, thus it is preferable to don't rely on releases but rather use master branch of git repository mentioned on the project homepage

Comment by Abhik Shah on 2008-06-26 22:18

Abhik Shah — Thu, 26 Jun 2008 22:18:49 -0000

If you have problems installing or using Pebl or have any questions or comments, please do not hesitate to contact me at: abhikshah@gmail.com

Comment by Cheng Soon Ong on 2008-08-04 15:23

Cheng Soon Ong — Mon, 04 Aug 2008 15:23:41 -0000

This software was recently used to demonstrate SVMs in the introductory tutorial: Asa Ben-Hur, Cheng Soon Ong, Sören Sonnenburg, Bernhard Schölkopf, and Gunnar Rätsch, (2008) Support Vector Machines and Kernels for Computational Biology, PLoS Computational Biology, 4.

http://svmcompbio.tuebingen.mpg.de

Comment by ALI EL AKADI on 2008-08-11 11:41

ALI EL AKADI — Mon, 11 Aug 2008 11:41:12 -0000

Thanks

Comment by Ramanathan S on 2008-08-11 11:46

Ramanathan S — Mon, 11 Aug 2008 11:46:19 -0000

Some one has rated this product as good. But how do I down load. It says the product is available in a CVS but the CVS location is unknown.

Ram S Ramanathan

Comment by Ronan Collobert on 2008-08-11 16:34

Ronan Collobert — Mon, 11 Aug 2008 16:34:50 -0000

On http://torch5.sourceforge.net/download.html you can see the CVS location at cvs -d :pserver:torch5.cvs.sourceforge.net:/cvsroot/torch5 co torch5.1

Comment by Dmitrey Kroshko on 2008-09-02 20:20

Dmitrey Kroshko — Tue, 02 Sep 2008 20:20:15 -0000

I try to run makefile (from linux KUBUNTU) but it seems like it requires MATLAB installed. Is it true?

Comment by Peter Gehler on 2008-09-03 09:12

Peter Gehler — Wed, 03 Sep 2008 09:12:40 -0000

You can als run "make shared" and use the python version or bind your own C code.

Comment by Haymo Kutschbach on 2008-09-07 01:41

Haymo Kutschbach — Sun, 07 Sep 2008 01:41:42 -0000

Plotting example for ILNumerics.Drawing:

Comment by Giuseppe Passino on 2008-09-10 13:14

Giuseppe Passino — Wed, 10 Sep 2008 13:14:13 -0000

I've been using it for a long time. Nice architecture, pretty good features. Some aspects may need optimisation in the implementation. Total lack of documentation makes the use of the library a bit hard at the beginning.

The library can be used also as a stand-alone program specifying factors as raw files. This can be useful in case basic inference tasks are required. In this case, the presence of some commented examples makes the use of the program easier.

Comment by Soeren Sonnenburg on 2008-09-12 16:14

Soeren Sonnenburg — Fri, 12 Sep 2008 16:14:36 -0000

In case you find bugs, feel free to report them at http://trac.tuebingen.mpg.de/shogun.

Comment by Soeren Sonnenburg on 2008-09-12 17:06

Soeren Sonnenburg — Fri, 12 Sep 2008 17:06:36 -0000

We are very much interested in feedback, so please drop us a note :-)

Comment by Soeren Sonnenburg on 2008-09-21 09:24

Soeren Sonnenburg — Sun, 21 Sep 2008 09:24:01 -0000

I would very much prefer official tarballs. So +1 for a real download link.

Comment by Ronan on 2008-09-21 15:19

Ronan — Sun, 21 Sep 2008 15:19:43 -0000

I am working on it!

Comment by Leon Bottou on 2008-09-23 21:42

Leon Bottou — Tue, 23 Sep 2008 21:42:34 -0000

Version 1.2 fixes bug in the preprocessing program.

Comment by Raj on 2008-09-24 09:43

Raj — Wed, 24 Sep 2008 09:43:49 -0000

I go through some journal papers and conference papers based on LWPR algorithm. From my knowledge in all the papers LWPR algorithm used in Controlling the robot arms. Now my question is can i use LWPR algorithm to control the movements of mobile robots?

waiting for your reply

Thank you

Comment by Joris Mooij on 2008-09-30 23:06

Joris Mooij — Tue, 30 Sep 2008 23:06:23 -0000

I am pleased to announce a new version of libDAI. It features several optimizations of implementation details and improved doxygen documentation.

Comment by Jerry on 2008-10-16 14:18

Jerry — Thu, 16 Oct 2008 14:18:22 -0000

"help mpi_kmeans_mex" does not work! I think there is no mpi_kmeans_mex.m file in the Archive:)

Comment by Stefan on 2008-10-28 13:17

Stefan — Tue, 28 Oct 2008 13:17:55 -0000

Dear Raj, In general, there is no reason why you should not be able to use LWPR for mobile robots. However, without a clearer picture of what you want to learn, it is hard to judge whether other algorithms wouldn't be more suitable. Cheers, Stefan

Comment by Danilo Mandic on 2008-10-29 11:15

Danilo Mandic — Wed, 29 Oct 2008 11:15:55 -0000

The Delay vector variance (DVV) method uses predictability of the signal in phase space to characterize the time series. Using the surrogate data methodology, so called DVV plots and DVV scatter diagrams can be generated using the DVV method, as a test statistic, to examine the determinism/stochastisity and linearity/nonlinearity within a signal simultaneously. In DVV scatter diagram, the target variance values of the original signal is plotted against the averaged variance values, calculated over a number of iAAFT surrogates. As a result, for linear signals, the scatter diagram coincides with the bisector line and conversely for nonlinear signals, the scatter diagram deviates from bisector lineas shown in the attached document. The DVV method has been successfully applied to analyse the nature of biometric signals (EEG and fMRI).

Comment by Frank Xavier on 2008-11-07 04:36

Frank Xavier — Fri, 07 Nov 2008 04:36:48 -0000

With more than 500 data mining, pre-processing, visualisation, and evaluation operators/modules for the complete data mining process and Weka fully integrated, RapidMiner probably is one of the most comprehensive data mining solutions available. It has significantly matured over the years, i.e. scalability, robustness, and usability for complex real-world data mining tasks are met better than by any other open source data mining tool I know.

Comment by Zoea on 2008-11-23 15:50

Zoea — Sun, 23 Nov 2008 15:50:57 -0000

Buy Cialis Online
Buy Phentermine Online
Buy Viagra Online
Buy Levitra Online
accupril 10 mg
drug effects lipitor side
geodon topamax effexor
effects melatonin pediatric
purchase generic zyban
side effects lopid