PySOMVis framework

PySOMVis framework involves popular Self-Organizing Maps visualization techniques, which is inspired by Java based SOMToolbox (http://www.ifs.tuwien.ac.at/dm/somtoolbox/index.html)

Installation

Installation via commandline using pip: Installation can be performed by first cloning this repository and then navigating to the PySOMVis folder. Inside the folder, one then has to type:

pip install .

Import

After installation one can import it iby the following command:

import pysomvis

To get the ''main-class'':

from pysomvis import PySOMVis

from pysomvis import *

Current visualizations

Pictures bellow are based on the projected Chain Link Data set (http://ifs.tuwien.ac.at/dm/somtoolbox/datasets.html). It is synthetic data representing two intertwined rings, which presents the topology violations after projection.
The SOM map represents 18x12 neurons trained 10000 times with learning rate 0.7 and sigma 7 in SOMToolbox.

^{^{Activity Histogram}}	^{^{Cluster Connection}}	^{^{Component Plane}}	^{^D-Matrix}	^{^{Topology Error}}	^{^U-Matrix}
^{^{Graph based}}	^{^{Hit Histogram}}	^{^{Metro Map}}	^{^{Minimum Spanning Tree}}	^{^U*-Matrix}	^{^{Pie Chart}}
^{^{Quantization Error}}	^{^{Smoothed Data Histogram}}	^{^{Sky Metaphor}}	^{^Chessboard}	^{^{K-means (2 clusters)}}	^{^{average, complete, single, WARD}}

SOMStreamVis approach

SOMStreamVis approach helps to explore dynamic pattern with trained map. The example represents weather forecast, it includes 5 features of different temperature, pressure and wind speed values over 35 years (taken from the https://power.larc.nasa.gov/data-access-viewer/).
The features are following:

RH2M - Relative Humidity at 2 Meters (%)
PS - Surface Pressure (kPa)
T2M - Temperature at 2 Meters (C)
WS50M - Wind Speed at 50 Meters (m/s)
ALLSKY_SFC8_LW_DWN - Downward Thermal Infrared (Longwave) Radiative Flux (kW-hr/m^2/day)

Dynamic exploration with SOMStreamVis:

^{SOMStreamVis interface with trajectory-based approach}

^{Projection of 3 years (coloring is based on WARD clusterisation)}

^{Projection of 35 years (coloring is based on WARD clusterisation)}

Citation