Bug #2740

CAH, the number of clusters is capped to 16 due of FactomineR

Added by Sebastien Jacquot 3 months ago.

Status:New Start date:01/23/2020
Priority:Normal Due date:
Assignee:- % Done:

0%

Category:Charts Spent time: -
Target version:TXM X.X

Description

The verification tests have been implemented in the UI.
The maximum number of clusters has been fixed as this : rows count - 1 and columns count -1, according to the computing mode (rows or columns) for CA and Lexical Table because FatoMineR can't compute a CAH where the cluster numbers is equals to the number of rows/cols. Is it normal ?

Other tasks to do:
- FactoMineR seems to manage a maximum of 16 clusters when plotting with plot.HCPC, it may be a FactoMineR bug linked to the plotting of inertia barplot
We can see when computing a CAH on rows that the barplot never exceeds 16 bars and trying to plot more than 16 clusters results in an error in R source code. After some tests plot.HCPC with argument tree.barplot=FALSE can exceed 16 clusters. A solution may be to plot the inertia barplot in another SWT component with plot.HCPC, as for the CA singular values.
The results generated by HCPC and used in CAH can contain more clusters than 16 but FactoMineR doesn't seems to manage to draw them with plot.HCPC. At this moment, I fixed the max clusters to 16 in the UI tool bar. We may investigate this and also limit the maximum clusters to 16 in the CAH preference page, if it's confirmed. NOTE: We could generate charts with more clusters directly form the HCPC result by using the JFC charts engine since it doesn't use plot.HCPC. In this case we need to find a way to define the max clusters count outside the CAH class, maybe at the charts engine levels, e.g. ChartsEngine.getMaxAvailableClusters(CAH ca) which could return 16 for the R implementation and the real maximum for the JFC implementation
- define if implement these verifications at CAH TBX level, before/during step computing, is useful

Also see old issue: #836

Also available in: Atom PDF