================================================================================================ FACULTY: khoa_cong_nghe_thong_tin | strategy=expanding | event y<5.0 DEVICE: cuda (Tesla V100-SXM2-32GB) | model=MultiTaskNet(reg+cls+blend) DEV folds=[1, 2, 3, 4, 5] | HOLDOUT folds=[6, 7] | tune=F1 (chon signal theo AP dev) ================================================================================================ >>> DEVELOPMENT (tune tren fold dev) [Class weights] n_low=613, n_high=19112 | raw_inverse_ratio=31.178 | CB_low=10.1327, CB_high=0.7071 | CB_ratio=14.330 [Class weights] n_low=802, n_high=24145 | raw_inverse_ratio=30.106 | CB_low=8.7669, CB_high=0.7420 | CB_ratio=11.815 [Class weights] n_low=954, n_high=28591 | raw_inverse_ratio=29.970 | CB_low=7.9554, CB_high=0.7679 | CB_ratio=10.360 [Class weights] n_low=1172, n_high=34650 | raw_inverse_ratio=29.565 | CB_low=6.9857, CB_high=0.7975 | CB_ratio=8.759 [Class weights] n_low=1416, n_high=39480 | raw_inverse_ratio=27.881 | CB_low=6.0754, CB_high=0.8180 | CB_ratio=7.428 signal blended : AP=0.169 | F1-opt tau=-0.033 flag_rate=0.069 (ratio=1.65) -> P=0.223 R=0.367 F1=0.278 signal P(y P=0.217 R=0.395 F1=0.280 <= CHON (AP cao nhat) ### DEVELOPMENT (validation) — BLENDED [signal=blended] fold test base% | AP AUC | ACC@t P@t R@t F1@t | rate% ACC@r P@r R@r F1@r 1 5222 3.6% | 0.120 0.661 | 0.963 0.429 0.063 0.111 | 6.9% 0.912 0.128 0.243 0.168 2 4598 3.3% | 0.374 0.853 | 0.964 0.445 0.401 0.422 | 6.0% 0.942 0.288 0.520 0.371 3 6277 3.5% | 0.157 0.811 | 0.879 0.147 0.514 0.228 | 5.4% 0.932 0.196 0.307 0.239 4 5072 4.8% | 0.217 0.808 | 0.897 0.235 0.500 0.319 | 5.7% 0.931 0.314 0.373 0.341 5 6398 5.5% | 0.187 0.761 | 0.911 0.258 0.333 0.291 | 7.9% 0.905 0.249 0.359 0.294 ------------------------------------------------------------------------------------------------------------ POOLED 27567 4.2% | 0.169 0.754 | 0.920 0.223 0.367 0.278 | 6.4% 0.923 0.231 0.354 0.279 ### DEVELOPMENT (validation) — CLASSIFICATION HEAD P(y>> DEVELOPMENT (tune tren fold dev) [Class weights] n_low=613, n_high=19112 | raw_inverse_ratio=31.178 | CB_low=10.1327, CB_high=0.7071 | CB_ratio=14.330 [Class weights] n_low=802, n_high=24145 | raw_inverse_ratio=30.106 | CB_low=8.7669, CB_high=0.7420 | CB_ratio=11.815 [Class weights] n_low=954, n_high=28591 | raw_inverse_ratio=29.970 | CB_low=7.9554, CB_high=0.7679 | CB_ratio=10.360 [Class weights] n_low=1172, n_high=34650 | raw_inverse_ratio=29.565 | CB_low=6.9857, CB_high=0.7975 | CB_ratio=8.759 [Class weights] n_low=1416, n_high=39480 | raw_inverse_ratio=27.881 | CB_low=6.0754, CB_high=0.8180 | CB_ratio=7.428 signal blended : AP=0.169 | F1-opt tau=-0.033 flag_rate=0.069 (ratio=1.65) -> P=0.223 R=0.367 F1=0.278 signal P(y P=0.217 R=0.395 F1=0.280 <= CHON (AP cao nhat) ### DEVELOPMENT (validation) — BLENDED [signal=blended] fold test base% | AP AUC | ACC@t P@t R@t F1@t | rate% ACC@r P@r R@r F1@r 1 5222 3.6% | 0.120 0.661 | 0.963 0.429 0.063 0.111 | 6.9% 0.912 0.128 0.243 0.168 2 4598 3.3% | 0.374 0.853 | 0.964 0.445 0.401 0.422 | 6.0% 0.942 0.288 0.520 0.371 3 6277 3.5% | 0.157 0.811 | 0.879 0.147 0.514 0.228 | 5.4% 0.932 0.196 0.307 0.239 4 5072 4.8% | 0.217 0.808 | 0.897 0.235 0.500 0.319 | 5.7% 0.931 0.314 0.373 0.341 5 6398 5.5% | 0.187 0.761 | 0.911 0.258 0.333 0.291 | 7.9% 0.905 0.249 0.359 0.294 ------------------------------------------------------------------------------------------------------------ POOLED 27567 4.2% | 0.169 0.754 | 0.920 0.223 0.367 0.278 | 6.4% 0.923 0.231 0.354 0.279 ### DEVELOPMENT (validation) — CLASSIFICATION HEAD P(y>> DEVELOPMENT (tune tren fold dev) [Class weights] n_low=613, n_high=19112 | raw_inverse_ratio=31.178 | CB_low=10.1327, CB_high=0.7071 | CB_ratio=14.330 [Class weights] n_low=802, n_high=24145 | raw_inverse_ratio=30.106 | CB_low=8.7669, CB_high=0.7420 | CB_ratio=11.815 [Class weights] n_low=954, n_high=28591 | raw_inverse_ratio=29.970 | CB_low=7.9554, CB_high=0.7679 | CB_ratio=10.360 [Class weights] n_low=1172, n_high=34650 | raw_inverse_ratio=29.565 | CB_low=6.9857, CB_high=0.7975 | CB_ratio=8.759 [Class weights] n_low=1416, n_high=39480 | raw_inverse_ratio=27.881 | CB_low=6.0754, CB_high=0.8180 | CB_ratio=7.428 signal blended : AP=0.169 | F1-opt tau=-0.033 flag_rate=0.069 (ratio=1.65) -> P=0.223 R=0.367 F1=0.278 signal P(y P=0.217 R=0.395 F1=0.280 <= CHON (AP cao nhat) ### DEVELOPMENT (validation) — BLENDED [signal=blended] fold test base% | AP AUC | ACC@t P@t R@t F1@t | rate% ACC@r P@r R@r F1@r 1 5222 3.6% | 0.120 0.661 | 0.963 0.429 0.063 0.111 | 6.9% 0.912 0.128 0.243 0.168 2 4598 3.3% | 0.374 0.853 | 0.964 0.445 0.401 0.422 | 6.0% 0.942 0.288 0.520 0.371 3 6277 3.5% | 0.157 0.811 | 0.879 0.147 0.514 0.228 | 5.4% 0.932 0.196 0.307 0.239 4 5072 4.8% | 0.217 0.808 | 0.897 0.235 0.500 0.319 | 5.7% 0.931 0.314 0.373 0.341 5 6398 5.5% | 0.187 0.761 | 0.911 0.258 0.333 0.291 | 7.9% 0.905 0.249 0.359 0.294 ------------------------------------------------------------------------------------------------------------ POOLED 27567 4.2% | 0.169 0.754 | 0.920 0.223 0.367 0.278 | 6.4% 0.923 0.231 0.354 0.279 ### DEVELOPMENT (validation) — CLASSIFICATION HEAD P(y=5 Failure <5 ------------------------------------------------------------------ Normal >=5 4,743 345 Failure <5 325 238 ------------------------------------------------------------------ Normal correctly rejected: 93.22% Failures successfully detected: 42.27% Confusion matrix — fold 6 — drift-aware @rate=10.002% ------------------------------------------------------------------ Actual / Predicted Normal >=5 Failure <5 ------------------------------------------------------------------ Normal >=5 4,757 331 Failure <5 329 234 ------------------------------------------------------------------ Normal correctly rejected: 93.49% Failures successfully detected: 41.56% 7 6926 15.1% | 0.408 0.788 | 0.835 0.452 0.440 0.446 | 18.2% 0.818 0.417 0.500 0.454 Confusion matrix — fold 7 — fixed threshold @t=0.16697 ------------------------------------------------------------------ Actual / Predicted Normal >=5 Failure <5 ------------------------------------------------------------------ Normal >=5 5,320 558 Failure <5 587 461 ------------------------------------------------------------------ Normal correctly rejected: 90.51% Failures successfully detected: 43.99% Confusion matrix — fold 7 — drift-aware @rate=18.164% ------------------------------------------------------------------ Actual / Predicted Normal >=5 Failure <5 ------------------------------------------------------------------ Normal >=5 5,144 734 Failure <5 524 524 ------------------------------------------------------------------ Normal correctly rejected: 87.51% Failures successfully detected: 50.00% ------------------------------------------------------------------------------------------------------------ POOLED 12577 12.8% | 0.401 0.807 | 0.856 0.436 0.434 0.435 | 14.5% 0.847 0.416 0.471 0.441 Confusion matrix — POOLED — fixed development threshold ------------------------------------------------------------------ Actual / Predicted Normal >=5 Failure <5 ------------------------------------------------------------------ Normal >=5 10,063 903 Failure <5 912 699 ------------------------------------------------------------------ Normal correctly rejected: 91.77% Failures successfully detected: 43.39% Confusion matrix — POOLED — drift-aware warning policy ------------------------------------------------------------------ Actual / Predicted Normal >=5 Failure <5 ------------------------------------------------------------------ Normal >=5 9,901 1,065 Failure <5 853 758 ------------------------------------------------------------------ Normal correctly rejected: 90.29% Failures successfully detected: 47.05% ====================================================================================================