Screenshots

The following screenshots have been taken on Windows 7 running the full, licensed version of the SOM Analyzer. With each screenshot, there is a short description of what the view represents.

Screenshot 1. The SOM Analyzer is visualizing a SOM consisting of 27 x 18 neurons having a dimension 50. The pane on the left shows the name of each parameter levels and as you can see they are all checked. Checked items are visualized on the visualization area. On top of the pane, you can also see other visualizations of which only the Cluster Detection is currently active.

Screenshot 2. This is a snapshot of the same SOM and data as screenshot 1. The difference here is that the left pane now shows the data of a SOM reference vector. The data pane here now shows the data of map unit (19, 8) and you can see it indicated by the white hexagon with each visualization. Although not visible here (I could not capture it with Greenshot for some reason), you see an informative tooltip whenever you take your mouse pointer over a map unit.

Screenshot 3. Here you see a rather large SOM having 80 x 60 map units but only having a reference vector dimension 5. By observing the parameter level visualizations, you can see strong correllations between them. The map has also been labelled for easier interpretation and as you can from the open label input dialog, the labelling is still under way.

Screenshot 4. In here a SOM is yet to be created. First, the data must be selected. This screenshot illustrates how data can be fetched from a MySQL database. After connecting to a server, you can specify the exact SQL command to get your data. The SOM Analyzer wizard will guide you all the way through the SOM creation process.

Screenshot 5. The SOM Analyzer features a wizard that helps you get started with SOMs even if you do not know anything about them. There are three modes to using the wizard and you can configure which one you want to use. The fully automatic wizard will only ask you to select your data and from then on will do the rest automatically. The tab pages you see in this screenshot illustrate this type of wizard. The semi automatic wizard mode asks you a bit more about your problem but still takes most of the work from your shoulders. The third and last option is called manual and it will let you have full control over the SOM. In that mode, you get to select all parameters and fine tune the results.

Screenshot 6. The final stage of the wizard in automatic and manual modes is running the iterator. The purpose of the iterator is to run the SOM algorithm against your data while varying the SOM algorithm parameters in the process. You can effectively control which parameters are mutated and how. After the iterator completes or even during the process, you can have a look at the results as shown by this screenshot. By inspecting the results, you can simply click a row and click Create SOM button and the SOM Analyzer creates that particular SOM for you. The iterator is very handy if you have large data sets and want to thoroughly explore various SOMs against them. Namely, you can leave the iterator running for several hours or days and check the results when ready.

Screenshot 7. Here you see how the SOM Analyzer can then be used to monitor process with the trained SOM. In here, a trajectory shows the path of last 77 spots (trajectory length) the yellow operating point has passed. The data set has consisted of 89421 vectors and it has been fully processed. This screenshot also shows you a new visualization called the Best-Matching Unit (BMU) Histogram on top right of the screen. It is very handing in showing the hot spots on the map. Via the context menu you can access by clicking the right mouse button, a data display has been enabled for the BMU Histogram. With that, each hexagon shows the number of hits it has received when processing the data. The cluster display on top left of the screen shows some labelled clusters. You can fully control the trajectory even after the data has been processed. It is just like using a MP3 player with play, stop, pause and advance buttons. For easier access, there are key shortcuts for trajectory control as well.

Screenshot 8. Another handy feature is Find BMU. With that, you can manually input multi-dimensional data to search for a BMU. SOM is capable of processing input data even if not all data components are present. In a similar fashion, you can look for a BMU even if you do not specify all data components.