imStudd / Self-Organizing-Map

Implementation of artificial neural network Self-Organizing Map (SOM).

Self-Organizing Map

Prerequisites

• Python 3.8 (OPTIONAL)
  • nltk
  • numpy

Python is an optional prerequisite, it will be used for preprocessing if the data to be processed is text.
The method used for preprocessing is TF-IDF which is a fairly basic method.

Usage

make
./som.out [OPTIONS]

Options

Option	Argument(s)	Description
-h, --help	-	Display options.
-d, --data	<path>	Set data file (If not specified here, the data ).
-t, --text	<path> <max_features> <0|1>	Use text data file, it will be preprocessed. Mode 0 keep m recurrent terms for reduce dimensionality and mode 1 use random projection.
-c, --config	<path>	Set configuration file (config.ini by default).
-l, --load	<path>	Load neurons file (.dat file).
-v, --verbose	-	Print progression.
--print_data	-	Print all data.
--print_classes	-	Print all classes.
--no_print_map	-	Do not print the map.

Configuration

Name	Type	Description
DATA_PATH	string	Path of data file.
DATA_NAMES	string	Names of each data column (csv format). (OPTIONAL)
NEURONS_NUMBER	unsigned int	Number of neurons.
MAP_WIDTH	unsigned int	Neuron map width.
MAP_HEIGHT	unsigned int	Neuron map height.
NEIGHBORHOOD_RADIUS	unsigned int	Initial neighborhood radius.
NB_ITERATION_PHASE_2	unsigned int	Number of training iteration.
NB_ITERATION_PHASE_1	unsigned int	Number of training iteration for the first phase.
LEARNING_RATE_PHASE_1	unsigned int	Learning rate for the first phase.
LEARNING_RATE_PHASE_2	unsigned int	Learning rate for the second phase.
RANDOM_MIN	double	Minimum random value for neurons initialization. (OPTIONAL)
RANDOM_MAX	double	Maximum random value for neurons initialization. (OPTIONAL)
SHUFFLE	boolean	Get random data during training.
GAUSSIAN	boolean	Gaussian neighbourhood function.

Use Iniparser library as INI file parser.

Data format

Data must be in csv format.
Each value must be a numerical value except the last one which corresponds to the class, can be a string.

4.9, 3.0, 1.4, 0.2, Iris-setosa
4.7, 3.2, 1.3, 0.2, Iris-setosa
4.6, 3.1, 1.5, 0.2, Iris-setosa
...

Example

Example with Fisher's Iris data set (downloadable here).

License

This project is licensed under the MIT License - see the LICENSE file for details.