Skip to content

Latest commit

 

History

History
87 lines (56 loc) · 2.87 KB

README.md

File metadata and controls

87 lines (56 loc) · 2.87 KB
Raw

SS-SOM

A Semi-Supervised Self-Organizing Map for Clustering and Classification

Sample code to the model proposed in https://ieeexplore.ieee.org/abstract/document/8489675

It was presented in the 2018 World Conference on Computational Intelligence (WCCI), precisely at the International Joint Conference on Neural Networks (IJCNN), in Rio de Janeiro, Brazil.

Cite:

Please cite our paper if you use this code in your own work:

@inproceedings{braga2018semi,
  title         = {A Semi-Supervised Self-Organizing Map for Clustering and Classification},
  author        = {Braga, Pedro HM and Bassani, Hansenclever F},
  booktitle     = {2018 International Joint Conference on Neural Networks (IJCNN)},
  pages         = {1--8},
  year          = {2018},
  organization  = {IEEE}
}

Requirements:

  1. You must have a file containing all the paths to the datasets you want to use. You can follow this example.

  2. You must have a parameters file:

To run SS-SOM, there are 11 parameters to set:

  • a_t
  • lp
  • dsbeta
  • age_wins
  • e_b
  • e_n
  • epsilon_ds
  • minwd
  • epochs
  • seed
  • push_rate

You can follow this example, where the first eleven rows represent the first set of parameters, the next 11 rows the second set and so on.

Also, it is important to update the constant noClass, if necessary. The default value is 999.

Parameters Generation

The sample code to generate the parameters with LHS, as in the paper, is available here.

Running:

  1. Make sure you fill the requirements.

  2. Open the NetbeansProject with Netbeans

  3. Set the main arguments for the program:

-i: this flag is used to get the path to the file containing all the paths to the datasets to be used.

-t: this flag is used to get the path of the folder with the test files

-r: this flag is used to get the path to the results folder

-p: this flag is used to get the path to the parameters file

-f[optional]: this flag disables the noisy filtering and all samples will be assigned to a cluster.

-n[optional]: this flag is used to define if the data needs to be normalized.

-c[optional]: this flag is used to execute a complete train and test experiment.

...and more.

For example, to run experiments for these real datasets the arguments will be as follows:

-i ../../Parameters/inputPathsTrain01 -t ../../Parameters/inputPathsTest -r sssom-l01/ -p ../../Parameters/sssom_0 -c

After that, you can evaluate the results folder files.

Metrics Calculation

See py-scripts.