Hello people,
I'm desperate. For weeks I've been working on the process. After several remodeling and many attempts, it still does not work. :'( In my stand alone version, everything works fine. In my whole process with loops, an error appears: "The attribute "hungarian cluster" does not exist.... ", because my R script loses the data. So it's clear, that the following operator can not find an attribute.
My if statement in R intervenes. The error emerges in the fourth pass of inner loop and generate this error. The log says: "PM INFO: Hungarian Algorithm: Fehler in minWeightBipartiteMatching(clusterA, clusterB) : number of cluster or number of instances do not match"
Why it works the first three repetitions?! I hope you can help me!
DownloadError version of my process:
https://dl.dropbox.com/u/5861880/Error_version.xmlWorking minimal version without loops:
https://dl.dropbox.com/u/5861880/Working_version.xmlExample setI use the wine dataset from UCI:
http://archive.ics.uci.edu/ml/datasets/Wine.
What do I want?
- 10 splits of D to Dtraining and Dtest. I repeat it for k = 3 to 20
- For every split, I choose ten times k examples for every predefined cluster.
- For every inner loop, I run k-medois and relabel the training data with the hungarian method in R. For this purpose I use the k previously selected examples per cluster and optimize they. This bases on mapping computated cluster and predefined cluster.
- Afterwards I measure the performance of my SVM-Model with Dtest
At the end, I have 17(k)*10(splits)*10(iterations) perfomance vectors.
