Use k-fold cross-validation with CRFSuite and save it in a model file

Question

I'm new with the CRFSuite library but I know how to train a model and save it in a file thanks to the "-m" option. However, I try to master de k-fold cross-validation but the "-m" option seem to not working to save de process in a file. The command simply doesn't create the model file.

Can I have a simple example of the good use of k-fold cross-validation with CRFSuite that write the process in a file (not the log file, but the model file). In the documentation, the author explain how to do the k-fold cross-validation, but not specify a file to save the model. I really wonder why performing a cross-validation if we don't save the model generated after...

I found documentation about CRFSuite here: http://www.chokkan.org/software/crfsuite/manual.html

But it's not enough for my needs. Thanks.

Answer 1

I think you are a little confused regarding cross validation and model generation.

Let's say you perform a 10-fold cross validation. The data will be divided in 10 train-test sets. As a result, 10 models are created (1 with each partial set of data).

So which of the 10 models would you like to store? You probably want the one that uses all the data for training, i.e., none of the 10 that the cross-validation process performs.

Cross validation is only an approximated measure of the performance you will get with the model that uses all data.