eval command sometimes reports precision as "nan"

[osma]

For example when using annif eval with the 20news dataset, the Precision@K values are reported as nan:

$ annif eval vw 20news100.tsv 
Precision (doc avg):	0.09399999999999999
Recall (doc avg):   	0.94
F1 score (doc avg): 	0.170909090909091
Precision (conc avg):	0.09934443130487655
Recall (conc avg):  	0.93
F1 score (conc avg):	0.17739629326278103
Precision (microavg):	0.09494949494949495
Recall (microavg):  	0.94
F1 score (microavg):	0.1724770642201835
NDCG:               	0.7578547650886536
NDCG@5:             	0.7387738112464632
NDCG@10:            	0.7578547650886536
Precision@1:        	nan
Precision@3:        	nan
Precision@5:        	nan
LRAP:               	0.7025119047619048
True positives:     	94
False positives:    	896
False negatives:    	6
Documents evaluated:	100

PR coming up soon...