Open johann-petrak opened 8 years ago
Copied over from redundant #47 : Have a look how difficult it would be to add ordinal regression. This would be useful for targets like ordered degrees of sentiment, reviews etc. There is a mention of a Python implementation here: http://fa.bianp.net/blog/2013/logistic-ordinal-regression/ (see also http://pythonhosted.org/mord/) and there are some R packages for that. Also, proper coding and definition of the cost function with NN may work?
not sure about this, needs more thinking! This is about predicting per-instance ordinal values where we only know the partial ordering of targets but where the target value is otherwise not on an interval or ratio scale.