What do you think is the explanation for that? I had assumed that using a lookup table would be faster considering that the loggam implementation has loops and makes calls to elementary functions in it. -- Sent from: http://numpy-discussion.10968.n7.nabble.com/