Thank you all for your comments. Here is an illustration of the difference in performance between tacit and explicit definitions, where there is minimal difference in the computations demanded by each. Both calculate the mean along the last axis of an array and both have infinite rank.
tacit_mean=: (+/ % #)"1"_ explicit_mean=: 3 :'(+/ % #)"1 y' a is a large matrix a=: i.1000 1000 Watch each gobble it up, first as verbs with infinite rank: 10 (6!:2) 'tacit_mean a' 0.0024503 10 (6!:2) 'explicit_mean a' 0.0033749 The tacit one is a little quicker. Now make each eat it up as rank 1 verbs: 10 (6!:2) 'tacit_mean"1 a' 0.0047234 10 (6!:2) 'explicit_mean"1 a' 0.0096502 The tacit verb is now twice as quick. As Roger has pointed out more than once, factors of 2 in execution time are not very consequential, but it does appear that tacit definitions do have performance benefits, and these are more pronounced when the rank of the verb is less than the rank of its argument. Graham ---------------------------------------------------------------------- For information about J forums see http://www.jsoftware.com/forums.htm
