sim

SimpSOM

Python library for Self-Organizing Maps

Showing:

Popularity

Downloads/wk

0

GitHub Stars

103

Maintenance

Last Commit

4mos ago

Contributors

6

Package

Dependencies

0

License

MIT

Categories

Readme

SimpSOM (Simple Self-Organizing Maps)

DOI PyPI version Build Status Documentation Status

Version 1.3.4

SimpSOM is a lightweight implementation of Kohonen Self-Organizing Maps (SOM) for Python 2.7 and 3, useful for unsupervised learning, clustering and dimensionality reduction.

The package is now available on PyPI, to retrieve it just type pip install SimpSOM or download it from here and install with python setup.py install.

It allows you to build and train SOM on your dataset, save/load the trained network weights, and display or print graphs of the network with selected features. The function run_colorsExample() will run a toy model, where a number of colors will be mapped from the 3D RGB space to the 2D network map and clustered according to their similarity in the origin space.

Dependencies

  • Numpy 1.11.0 (older versions may work);
  • Matplotlib 1.5.1 (older versions may work);
  • Sklearn 0.15 (older versions may work);

Example of Usage

Here is a quick example on how to use the library with a raw_data dataset:

#Import the library
import SimpSOM as sps

#Build a network 20x20 with a weights format taken from the raw_data and activate Periodic Boundary Conditions. 
net = sps.somNet(20, 20, raw_data, PBC=True)

#Train the network for 10000 epochs and with initial learning rate of 0.01. 
net.train(0.01, 10000)

#Save the weights to file
net.save('filename_weights')

#Information on each node is stored in the .nodeList attribute of the network. These include each node position
#in the hexagonal grid (.pos) or its weights (.weights), i.e. the position of the node in the features space.
position_node0 = net.nodeList[0].pos
weights_node0 = net.nodeList[0].weights 

#Print a map of the network nodes and colour them according to the first feature (column number 0) of the dataset
#and then according to the distance between each node and its neighbours.
net.nodes_graph(colnum=0)
net.diff_graph()

#Project the datapoints on the new 2D network map.
net.project(raw_data, labels=labels)

#Cluster the datapoints according to the Quality Threshold algorithm.
net.cluster(raw_data, type='qthresh')   

A More Interesting Example: MNIST

Here is another example of SimpSOM capabilites: the library was used to try and reduce a MNIST handwritten digits dataset. A 50x50 nodes map was trained with 500 MINST landmark datapoints and 100000 epochs in total, starting from a 0.1 learning rate and without PCA Initialisation.

Projecting a few of those points on the map gives the following result, showing a clear distinction between cluster of digits with a few exceptions. Similar shapes (such as 7 and 9) are mapped closed together, while relatively far from other more distinct digits. The accuracy of this mapping could be further improved by tweaking the map parameters, by training the network for more epochs or with a more adequate choice of initial dataset.

Documentation

See here the full API documentation

Citation

If using this library, please cite it as

Federico Comitani, 2019. fcomitani/SimpSOM: v1.3.4. doi:10.5281/zenodo.2621560

What\'s New

  • It's now possible to explicitly choose the output folder.

Rate & Review

Great Documentation0
Easy to Use0
Performant0
Highly Customizable0
Bleeding Edge0
Responsive Maintainers0
Poor Documentation0
Hard to Use0
Slow0
Buggy0
Abandoned0
Unwelcoming Community0
100
No reviews found
Be the first to rate

Alternatives

No alternatives found

Tutorials

No tutorials found
Add a tutorial