Line 70 in evaluate_prediciton.py results in no value being produced for both precision and recall if there are zero generated notes or the true notes up to a given onset time. However, if there no generated notes we can still get the recall (it's zero). Likewise, if there are no true notes we can still get the precision (it's also zero!).
In relation to issue #6, how nan values are being plotted, this produces artificially high values for small onset values (where there are likely no onsets to observe, and thus missing values).
Line 70 in evaluate_prediciton.py results in no value being produced for both precision and recall if there are zero generated notes or the true notes up to a given onset time. However, if there no generated notes we can still get the recall (it's zero). Likewise, if there are no true notes we can still get the precision (it's also zero!).
In relation to issue #6, how nan values are being plotted, this produces artificially high values for small onset values (where there are likely no onsets to observe, and thus missing values).