Fix handling of empty sets in set matching metrics

[wgantt]

Hi @wanmok and @ctongfei. In using this package for another project, I noticed that the current implementation of the set matching metrics gives undesired behavior in the case where:

• matching is one-to-one
• either the predicted or the reference set (but not both) is empty
• the inner metric is not discrete (so this line is not triggered)

In such a case, it’s possible to reach this line with a gram “matrix” that is a 1D array rather than a 2D array, yielding an error like the following when the call to spo.linear_sum_assignment is made:

File "/brtx/601-nvme2/wgantt/.conda/envs/ote/lib/python3.10/site-packages/metametric/core/matching.py", line 30, in score
    row_idx, col_idx = spo.linear_sum_assignment(
ValueError: expected a matrix (2-D array), got a 1 array

It feels like we should just add logic to return 0 if we reach here and either x or y is empty. What do you think?

[ctongfei]

@wgantt I think that you are right. If either x or y is empty, we should bypass the following logic and just return 0.

[ctongfei]

Fixed by #24 .