Orange: Self-Organizing Map
Sumber: https://docs.biolab.si//3/visual-programming/widgets/unsupervised/selforganizingmap.html
Komputasi dari self-organizing map.
Input
Data: input dataset
Output
Selected Data: instances selected from the plot Data: data with an additional column showing whether a point is selected
Self-organizing map (SOM) adalah tipe dari artificial neural network (ANN) yang di train menggunakan unsupervised learning untuk menghasilkan representasi dua-dimensi, discretized dari data. Ini adalah metoda untuk melakukan reduksi dimensi. Self-organizing maps menggunakan neighborhood function untuk menjaga property topological yang ada di bidang input.
The points in the grid represent data instances. By default, the size of the point corresponds to the number of instances represented by the point. The points are colored by majority class (if available), while the intensity of interior color shows the proportion of majority class. To see the class distribution, select Show pie charts option.
Just like other visualization widgets, Self-Organizing Maps also supports interactive selection of groups. Use Shift key to select a new group and Ctr+Shift to add to the existing group.
- SOM properties:
- Set the grid type. Options are hexagonal or square grid.
- If Set dimensions automatically is checked, the size of the plot will be set automatically. Alternatively, set the size manually.
- Set the initialization type for the SOM projection. Options are PCA initialization, random initialization and replicable random (random_seed = 0).
- Once the parameters are set, press Start to re-run the optimization.
- Set the color of the instances in the plot. The widget colors by class by default (if available).
- Show pie charts turns points into pie-charts that show the distributions of the values used for coloring.
- Size by number of instances scales the points according to the number of instances represented by the point.
Contoh
Self-organizing maps are low-dimensional projections of the input data. We will use the brown-selected data and display the data instance in a 2-D projection. Seems like the three gene types are well-separated. We can select a subset from the grid and display it in a Data Table.
