How to get 'Plot view' for SOM?
My Data mining process contains the following operators:
- DatabaseExampleSource
- SOMDimensionalityReduction
- ChangeAttributeRole
I have got the clustered data output in the Data view. But, I would like to get the SOM model for it (a grid structure showing the clusters). How do I get it in the 'Plot view' (what to select/mention for Plotter, x-axis, y-axis etc.)?
- DatabaseExampleSource
- SOMDimensionalityReduction
- ChangeAttributeRole
I have got the clustered data output in the Data view. But, I would like to get the SOM model for it (a grid structure showing the clusters). How do I get it in the 'Plot view' (what to select/mention for Plotter, x-axis, y-axis etc.)?
Find more posts tagged with
Sort by:
1 - 2 of
21
The problem here is that there are no documentation available about the 'Plot view'. For example, for the 'SOM' plotter, I do not know what best value to select for the 'Matrix', 'Adaptation radius' etc.
Hi,
that's a great example for the more urgent features
The best setting depends on your data set. There are no rules for gaining the best results, just start experimenting.
Originally the SOM was implemented for visualisation purposes only so your clustering results will not be show as you might expect. Instead of showing the values of each grid point this implementation based on the ESOM of Ultsch and hence view a distance matrix between the nodes or a matrix showing the empirical distribution densities at the nodes' values. These are then interpolated for a more smooth impression.
So it will be difficult to extract the values of one single dimension.
In my diploma thesis I have written a much improved version, but did not finde the time to include it into the program. But perhabs this will be possible in the future.
Greetings,
Sebastian
that's a great example for the more urgent features
The best setting depends on your data set. There are no rules for gaining the best results, just start experimenting.
Originally the SOM was implemented for visualisation purposes only so your clustering results will not be show as you might expect. Instead of showing the values of each grid point this implementation based on the ESOM of Ultsch and hence view a distance matrix between the nodes or a matrix showing the empirical distribution densities at the nodes' values. These are then interpolated for a more smooth impression.
So it will be difficult to extract the values of one single dimension.
In my diploma thesis I have written a much improved version, but did not finde the time to include it into the program. But perhabs this will be possible in the future.
Greetings,
Sebastian