- Python 3.5.1
- NumPy 1.10.4
- matplotlib
This script generates datasets to be used with the SOMToolbox. The focus is on clusters. The following settings can be configured:
- name of the output files
- number of dimensions
- number of clusters
- for each cluster:
- distribution type
- cardinality
- density
Both the cardinality and density range from 1 (low) to 10 (high). The following distribution types are supported:
- Binomial
- Exponential
- Gaussian
- Pareto
- Poisson
If the program is run with <= 2 dimensions it will also display a plot of the data.
The resulting vector and template vector files can be used by the SOMToolbox to train a SOM. The goal is to analyze the resulting maps and visualize how different cluster configuration can be detected.
First you'll want to install the dependencies. Afterwards you can run the program itself:
pip install -r requirements.txt
python generator.py <path-to-input-file>
In case the installation of the dependencies doesn't work out-of-the-box, they can also be installed manually:
The input file is a JSON file. The format should be self-explanatory given an example:
{
"dimensions": 2,
"export_name": "example1",
"clusters": [
{
"type": "binomial",
"cardinality": "4",
"density": "8"
},
{
"type": "exponential",
"cardinality": "5",
"density": "7"
},
{
"type": "pareto",
"cardinality": "9",
"density": "6"
},
{
"type": "poisson",
"cardinality": "9",
"density": "2"
},
{
"type": "gaussian",
"cardinality": "3",
"density": "10"
}
]
}
The results are written to the output folder with the name configured in the input file.