simpsom-dask 1.0.3

A mix of jcfaracco/xpysom-dask and fcomitani/simpsom, with heavy modifications

XPySom-dask

Self Organizing Maps with Dask Support and pretty hexagonal plot.

Simpsom-Dask is a partial rewrite of XPySom-Dask with vizualisation elements introduced in Simpsom. Some optimizations have been added, and the vizualisation elements have been improved, mostly to fit my needs.

Installation

You can download XPySom-dask from PyPi:

pip install simpsom-dask

By default, dependencies for GPU execution are not downloaded.

You can also choose to clone this repo and then run

cd simpsom-dask
pip install -e .

if you would like to modify things on the fly.

How to use it

The module interface is similar to MiniSom. In the following only the basics of the usage are reported, for an overview of all the features, please refer to the original MiniSom examples you can refer to: https://github.com/JustGlowing/minisom/tree/master/examples (you can find the same examples also in this repository but they have not been updated yet).

In order to use Simpsom you need your data organized as a Numpy matrix where each row corresponds to an observation or as list of lists like the following:

chunks = (4, 2)
data = [[ 0.80,  0.55,  0.22,  0.03],
        [ 0.82,  0.50,  0.23,  0.03],
        [ 0.80,  0.54,  0.22,  0.03],
        [ 0.80,  0.53,  0.26,  0.03],
        [ 0.79,  0.56,  0.22,  0.03],
        [ 0.75,  0.60,  0.25,  0.03],
        [ 0.77,  0.59,  0.22,  0.03]]      

Then you can train XPySom just as follows:

from simsom_dask import Simpsom

import dask.array as da

from dask.distributed import Client, LocalCluster

client = Client(LocalCluster())

dask_data = da.from_array(data, chunks=chunks)

som = Simpsom(6, 6, 4, sigma=0.3, learning_rate=0.5, use_dask=True, chunks=chunks) # initialization of 6x6 SOM
som.train(dask_data, 100) # trains the SOM with 100 iterations

You can obtain the position of the winning neuron on the map for a given sample as follows:

som.winner(data[0])

Differences with MiniSom

• The batch SOM algorithm is used (instead of the online used in MiniSom). Therefore, use only train to train the SOM, train_random and train_batch are not present.
• decay_function input parameter is no longer a function but one of 'linear', 'exponential', 'asymptotic'. As a consequence of this change, sigmaN and learning_rateN have been added as input parameters to represent the values at the last iteration.
• New input parameter std_coeff, used to calculate gaussian exponent denominator d = 2*std_coeff**2*sigma**2. Default value is 0.5 (as in Somoclu, which is different from MiniSom original value sqrt(pi)).
• New input parameter xp (default = cupy module). Back-end to use for computations.
• New input parameter n_parallel to set size of the mini-batch (how many input samples to elaborate at a time).
• Hexagonal grid support is recommended.

Authors

Copyright (C) 2025 Hugo Banderier