SOM algorithm binary for VO-CLOUD worker

Installation and Usage

1. Compile the application by running a compile script: ./compile.sh
2. Insert data files into input folder Wrapper/input/.
3. Arbitrary change the user configuration file config.json.
4. Run the application: $> python run.py
5. Look for a results in result file Wrapper/result.
6. Arbitrary run created web application by opening Wrapper/result/index.php file in your browser.

Parameters of configuration file

Name arbitrary name of the experiment.

• algorithm definition of algorithm
  • BMU "normal" or "probing" algorithm
  • threads select number of computational threads (integer)
• data definition of data input
  • path one or more paths to input files or whole directories.
  • file_type "csv" or "fits"
  • delimiter delimiter of csv file.
  • columns columns in file or dimension of input vector.
• parameters SOM parameters
  • topology "rect" for rectangular, "hex" for hexagonal
  • neighborhood_fcion "gaussian" or "bubble" function
  • neighborhood_radius size of initial neighborhood radius (integer)
  • map_size_x horizontal size of SOM grid (integer)
  • map_size_y vertical size of SOM grid (integer)
  • iteration number of iterations over whole data set (integer)
  • learning_rate initial learning rate (float in interval 0–1)
  • probing_iter number of iterations of probing algorithm (integer, used only if probing algorithm is selected)
• output selection of output files
  • visualization values true or false, create visualization or not.
  • error values true or false, count errors or not.
  • web values true or false, create web application or not.
• optional info selection of optional functionality
  • Names values true or false, add files with data names or not (number of files and rows must be equivalent to input data)
  • Names_path one or more paths to names files or whole directories
  • Classes values true or false, add files with classes or not (number of files and rows must be equivalent to input data)
  • Classes_path one or more paths to classes files or whole directories

Input Files

Input data has to be in CSV format. In case of having solely FITS files, an included converter can be used. The data sets can be separated in arbitrary number of files. They common size has no limitations as the files are read sequentially. However, the average time of running is slightly bigger when using more files, as a data have to be for each iteration repeatedly read from the disk. Optionally, it can be added the file containing data classes and names. It is option that can help recognize quality of chosen algorithm (by visualizating it). This data files should be equivalent to main data in a way how they are separated in files and rows. It is important to add, that class and name data are not used in training phase of algorithm. They are exclusively used for visualization.