Please pardon my lack of experience with the Mahout Java library. I have 
recently started using RecommenderIRStatsEvaluator to check the accuracy of our 
model/data combinations. I am a bit surprised that the Precision and Recall 
figures are identical to each other, every time I have run the evaluator. For 
example (taken from an early stage in the evaluation): 

INFO: Precision/recall/fall-out/nDCG/reach: 0.18461538461538465 / 
0.18461538461538465 / 1.0560608928695964E-4 / 0.2039069945815918 / 1.0

If my understanding of precision/recall is correct, this would mean that FP 
would have to be equal to FN in our model/data—or could there be some other 
thing affecting the evaluator?

Rafal Lukawiecki

Reply via email to