Calibration thresholds are off

[ebursztein]

The threshold selected in the labels don't match the one displayed:

Label would say ['matching'] == 0.333
but the table would say -> 0.3111

[ebursztein]

num positive matches 1416/2000
label distance

---

possible 0.0288222
likely 0.00458199
very_likely 0.000248671
match 0.0359477

distance f1 precision recall

---

0.000248671 0.309069 1 0.18
0.000498652 0.438326 0.99 0.28
0.00051856 0.451965 1 0.29
0.000980318 0.619186 0.99 0.45
0.00103468 0.630163 0.98 0.46
0.0010711 0.635407 0.99 0.47
0.00142562 0.709213 0.98 0.56
0.00171316 0.737118 0.97 0.6
0.00231791 0.769101 0.96 0.64
0.00234365 0.769296 0.95 0.64
0.00237638 0.771368 0.96 0.65
0.00276017 0.783292 0.95 0.67
0.0030542 0.793328 0.94 0.69
0.00326419 0.795491 0.93 0.7
0.00376821 0.804434 0.92 0.72
0.0042907 0.80984 0.91 0.73
0.0042969 0.809988 0.9 0.73
0.00431204 0.810452 0.91 0.73
0.00458175 0.810518 0.9 0.74
0.00458199 0.810205 0.89 0.74
0.00464582 0.812042 0.9 0.74
0.00505996 0.812834 0.89 0.75
0.00543237 0.81077 0.88 0.75
0.0059306 0.808431 0.87 0.76
0.00784302 0.815663 0.86 0.78
0.0100467 0.824593 0.85 0.81
0.0119423 0.826134 0.84 0.82
0.0134963 0.826516 0.83 0.83
0.0135162 0.826638 0.82 0.83
0.013761 0.827878 0.83 0.83
0.0138208 0.828411 0.82 0.83
0.0138208 0.828822 0.83 0.83
0.016178 0.83062 0.82 0.85
0.0184268 0.830253 0.81 0.86
0.0184957 0.829969 0.8 0.86
0.0185173 0.830769 0.81 0.86
0.0215253 0.831536 0.8 0.87
0.0288222 0.834051 0.79 0.89
0.0359477 0.837179 0.78 0.91
0.0404364 0.833815 0.77 0.92
0.0495148 0.836066 0.76 0.94
0.0576776 0.833385 0.75 0.94
0.0579422 0.833126 0.74 0.94
0.0599245 0.834318 0.75 0.95
0.0713266 0.832005 0.74 0.96
0.114868 0.832029 0.73 0.98
0.210144 0.830222 0.72 0.99
0.395062 0.82904 0.71 1

[ebursztein]

Still some issues with the calibration even after the rewerite.