Evaluating precision@k/recall@k for users with <k "known positives"

Hi,

I have a question about the precision@k and recall@k evaluation functions. I can see from the source code that precision is calculated as:

precision = np.squeeze(np.array(ranks.sum(axis=1))) / k

Wouldn't this calculation underestimate performance if I have a users who have voted/rated

Start by reviewing the precision@k and recall@k evaluation functions and the shown ranks.sum(axis=1) / k calculation. Determine the expected behavior for users with fewer than k known positives and whether the current evaluation matches that definition; done means the behavior has a documented resolution or a targeted regression test if it is incorrect.

Written by the indexing model from the issue text.