Actual class | P_NINT | P_COMP | P_COOP | Recall |
---|
NINT | 21 | 17 | 13 | 41% |
COMP | 3 | 3 | 2 | 37% |
COOP | 15 | 2 | 46 | 73% |
NINTa | 5292 | 928 | 145 | 83% |
COMPa | 68 | 49 | 6 | 39% |
COOPa | 154 | 17 | 520 | 75% |
- Note: Upper - no calibration. Lower - with calibration for recall (also marked with a). The calibration is described in Recall calibration and evaluation and is performed by substituting to each prediction (whether correct or not) its weight. The weight is defined as the ratio between the number of interactions that have the same N_12 as that prediction and the same count done for the N12 value that has the least interactions. Note that specificity cannot be calibrated without some additional assumptions