Ads by Google
Christian Borgelt's Web Pages

CoCoNAD and Frequent Item Set Mining

Download

coconad.pdf (4963 kb) CoCoNAD result diagrams
cocoexp.zip (63 kb) scripts and other source files
cocoexp.tar.gz (53 kb)
jitter.pdf (12397 kb) frequent item set mining result diagrams
jitter.zip (42 kb) scripts and other source files
jitter.tar.gz (35 kb)
prefer.pdf (103 kb) preference relations for pattern set reduction
prefer.zip (5 kb) sources for document prefer.pdf
prefer.tar.gz (4 kb)
psf+psr.zip (78 kb) Python scripts for Pattern Spectrum Filtering
psf+psr.tar.gz (70 kb) and Pattern Set Reduction

Description

The document coconad.pdf contains the result diagrams for the complete set of experiments with CoCoNAD (Continuous-time Closed Neuron Assembly Detection) to find inexactly synchronous spiking events in parallel neural spike trains, that were conducted for a paper that is currently under review. Only few of these diagrams are contained in the paper due to a lack of space. For the theory underlying the methods, please consult the paper and for the different patterns set reduction methods, the paper [Torre et al. 2013].

The archives cocoexp.{zip,tar.gz} contain scripts and other source files, with which the experiments were conducted and the document with the result diagrams was created.

The document jitter.pdf contains the diagrams for the complete set of experiments concerning the alternative method based on time binning and then applying standard frequent item set mining (FIM), to which CoCoNAD is compared in a paper that is currently under review. Only few of these diagrams are contained in the paper. For the theory underlying the methods, please consult the paper and for the different patterns set reduction methods, the paper Torre et al. 2013. The archive jitter.zip contains scripts and other source files, with which the experiments were conducted and the document with the result diagrams was created.

The document prefer.pdf contains diagrams that illustrate the various preference relations used for the pattern set reduction for signatures close to the detection border. For the theory underlying these methods, consult the paper Torre et al. 2013 referenced below. The archive prefer.zip contains scripts and other source files, with which these diagrams were created.

The scripts in the archives psf+psr.{zip,tar.gz} implement the full analysis process of parallel (spike) trains/point processes as it is described (although for standard frequent item set mining) in [Picado-Muiño et al. 2013] and [Torre et al. 2013]. A documentation of these scripts can be found here. Call the main script ccn+psf+psr.py without any arguments to obtain a help message that shows the invocation and the available options.

Note that the scripts etc. were developed on/for a GNU/Linux system (Ubuntu 12.10 or later) and thus are directly executable on such a system or a similar one (that is, some other GNU/Linux distribution). Although at least most of the Python scripts should also be working on a Windows system (with the possible exception of the parallelization scripts), most of the other scripts (like the run script, which is the main control script, and the makefile, which controls generating the diagrams from the result data) may need porting to batch files or something similar.

On a GNU/Linux system, the following software needs to be installed to run the experiments:

On such a system the experiments can be run by simply calling the main script run (in the directory cocoexp or jitter, respectively) on the command line, which does everything. The execution of the experiments exploits 4-fold parallelization, thus making full use of the quadcore processors basically all modern computers are equipped with. The progress of the experiments can be followed on the command line, to which regular progress messages are written. Once all experiments are completed (which, even on a modern computer system, can take more than 5 hours for the coconad experiments and more than 10 hours for the jitter experiments, mainly because of the huge number of individual experimental runs, namely in the hundreds of thousands), the result diagrams are created and compiled into the final documents, which are also directly available above.

References