-- Originally posted by (AT sourceforge.net): burfee -- -- This ticket was imported from http://sourceforge.net/p/rkward/feature-requests/109 on 2017-05-31 14:48:58 +0100 -- Hi, Cluster analysis would be great to have implemented in the menu. Something like this things here http://statmethods.net/advstats/cluster.html but automated from the menu. Also, scoring procedures would be very useful to have too. -- Labels: RKWard user interface --
hi burfee, this shouldn't be too hard, generally. as of version 0.5.7, RKWard supports additional plugins in the form of special R packages: http://rkward.sf.net/R/pckg/ and the package rkwarddev enables you to write such a plugin as a single R script, and even does the packaging for you. i'd be interested in writing a cluster analysis plugin, too. please drop a note if you'd like to help or even try yourself \(with some assistance, if you like\), especially if you have ideas regarding layout and workflow.
-- Originally posted by (AT sourceforge.net): burfee -- Hi m-eik, Great to see you're interested\! :D I can't write the plugin myself because i don't really know anything about programming in R, but i will think about the layout and workflow and come back with a few ideas in new comment shortly.
-- Originally posted by (AT sourceforge.net): burfee -- Hi again m-eik and sorry for the delay. I've been pretty caught up with work lately. Obviously it should be in the analysis menu. The cluster analysis window should allow you to choose the method to use, the number of clusters, the variables to include in the segmentation, the threshold to determine if a variable is worth enough depending on the statistic you're using. The output should contain something like this: -a pie chart showing the size of the segments -a table containing the statistics of the segments \(number of observations contained in the cluster, percentage of total, etc\) -barcharts for each segment, showing each of the selected variables' contribution to the cluster -overlaid hystograms for each of the variables in the clusters vs the overall sample - this feature would be great as it would allow you to see what really differentiates that cluster from the rest of the sample. That's what i could thing of based on the experience i had with other statistical software