## Held-out labels (mean): V-measure / many-to-one / labels / share in a class / sense / sense-COLL / merges
3000 LV-I: 0.622 / 0.712 / 1180.700 / 0.532 / 0.748 / 0.758 / 88.8
3000 LV-A: 0.720 / 0.933 / 1211.300 / 0.164 / 0.855 / 0.855 / 56.1
10000 LV-I: 0.637 / 0.720 / 1178.100 / 0.520 / 0.704 / 0.735 / 100.0
10000 LV-A: 0.726 / 0.929 / 1172.300 / 0.227 / 0.842 / 0.842 / 93.8
30000 LV-I: 0.658 / 0.759 / 1190.000 / 0.472 / 0.753 / 0.750 / 100.0
30000 LV-A: 0.722 / 0.918 / 1170.200 / 0.255 / 0.828 / 0.828 / 100.0
## Re-parse link F1 and level 2 (mean): link F1 / ucomp types / ucomp type recall / ucomp precision
3000 words(round 0) F1 0.455 ucomp recall 0.113; L-MST F1 0.626
LV-I: 0.449 / 7379 / 0.157 / 0.308
LV-A: 0.462 / 7569 / 0.121 / 0.319
10000 words(round 0) F1 0.514 ucomp recall 0.185; L-MST F1 0.689
LV-I: 0.504 / 23820 / 0.213 / 0.296
LV-A: 0.528 / 25433 / 0.198 / 0.315
30000 words(round 0) F1 0.570 ucomp recall 0.254; L-MST F1 0.700
LV-I: 0.544 / 69388 / 0.267 / 0.296
LV-A: 0.572 / 75680 / 0.256 / 0.316
## Predictions (at 30000)
L1: FAIL — sense LV-I − LV-I-COLL +0.003 [-0.004, +0.011]
L2: FAIL — link F1 LV-I − word MST -0.026 [-0.035, -0.016]
L3: FAIL — link F1 LV-I − L-MST -0.156 [-0.169, -0.144] (bar: lower > -0.02)
L4: PASS — ucomp type recall LV-I − word MST +0.013 [+0.005, +0.021]