Loaded pretrained weights for efficientnet-b2
---------- Fold: 1 ----------
0:05:06 | Epoch: 1/15 | Loss: 121.9 | Train Acc: 0.961 | Valid Acc: 0.912 | ROC: 0.913
0:05:01 | Epoch: 2/15 | Loss: 94.7 | Train Acc: 0.971 | Valid Acc: 0.938 | ROC: 0.974
0:04:59 | Epoch: 3/15 | Loss: 88.2 | Train Acc: 0.972 | Valid Acc: 0.939 | ROC: 0.968
0:04:57 | Epoch: 4/15 | Loss: 82.25 | Train Acc: 0.974 | Valid Acc: 0.952 | ROC: 0.981
0:04:56 | Epoch: 5/15 | Loss: 78.46 | Train Acc: 0.975 | Valid Acc: 0.924 | ROC: 0.95
0:04:55 | Epoch: 6/15 | Loss: 75.07 | Train Acc: 0.976 | Valid Acc: 0.939 | ROC: 0.975
Epoch     6: reducing learning rate of group 0 to 2.0000e-04.
0:04:55 | Epoch: 7/15 | Loss: 62.55 | Train Acc: 0.98 | Valid Acc: 0.946 | ROC: 0.981
Early stopping (no improvement since 3 models) | Best ROC: 0.9812042692939246
Loaded pretrained weights for efficientnet-b2
---------- Fold: 2 ----------
0:05:00 | Epoch: 1/15 | Loss: 88.39 | Train Acc: 0.971 | Valid Acc: 0.97 | ROC: 0.973
0:04:57 | Epoch: 2/15 | Loss: 81.09 | Train Acc: 0.974 | Valid Acc: 0.966 | ROC: 0.976
0:04:55 | Epoch: 3/15 | Loss: 76.29 | Train Acc: 0.976 | Valid Acc: 0.979 | ROC: 0.981
0:04:54 | Epoch: 4/15 | Loss: 74.83 | Train Acc: 0.976 | Valid Acc: 0.955 | ROC: 0.97
0:04:54 | Epoch: 5/15 | Loss: 69.22 | Train Acc: 0.977 | Valid Acc: 0.976 | ROC: 0.975
Epoch     5: reducing learning rate of group 0 to 2.0000e-04.
0:04:54 | Epoch: 6/15 | Loss: 57.89 | Train Acc: 0.981 | Valid Acc: 0.975 | ROC: 0.979
Early stopping (no improvement since 3 models) | Best ROC: 0.9811217376743738
Loaded pretrained weights for efficientnet-b2
---------- Fold: 3 ----------
0:04:54 | Epoch: 1/15 | Loss: 75.44 | Train Acc: 0.975 | Valid Acc: 0.967 | ROC: 0.982
0:04:54 | Epoch: 2/15 | Loss: 73.25 | Train Acc: 0.977 | Valid Acc: 0.971 | ROC: 0.981
0:04:54 | Epoch: 3/15 | Loss: 69.9 | Train Acc: 0.978 | Valid Acc: 0.978 | ROC: 0.983
0:04:53 | Epoch: 4/15 | Loss: 66.87 | Train Acc: 0.978 | Valid Acc: 0.979 | ROC: 0.984
0:04:52 | Epoch: 5/15 | Loss: 65.13 | Train Acc: 0.978 | Valid Acc: 0.978 | ROC: 0.987
0:04:53 | Epoch: 6/15 | Loss: 63.17 | Train Acc: 0.979 | Valid Acc: 0.979 | ROC: 0.984
0:04:51 | Epoch: 7/15 | Loss: 60.53 | Train Acc: 0.98 | Valid Acc: 0.978 | ROC: 0.984
Epoch     7: reducing learning rate of group 0 to 2.0000e-04.
0:04:53 | Epoch: 8/15 | Loss: 47.86 | Train Acc: 0.983 | Valid Acc: 0.982 | ROC: 0.984
Early stopping (no improvement since 3 models) | Best ROC: 0.9868701392202659
Loaded pretrained weights for efficientnet-b2
---------- Fold: 4 ----------
0:04:55 | Epoch: 1/15 | Loss: 61.75 | Train Acc: 0.98 | Valid Acc: 0.974 | ROC: 0.984
0:04:52 | Epoch: 2/15 | Loss: 59.32 | Train Acc: 0.98 | Valid Acc: 0.974 | ROC: 0.986
0:04:54 | Epoch: 3/15 | Loss: 55.97 | Train Acc: 0.981 | Valid Acc: 0.974 | ROC: 0.983
0:04:55 | Epoch: 4/15 | Loss: 55.27 | Train Acc: 0.981 | Valid Acc: 0.97 | ROC: 0.977
Epoch     4: reducing learning rate of group 0 to 2.0000e-04.
0:04:55 | Epoch: 5/15 | Loss: 41.34 | Train Acc: 0.986 | Valid Acc: 0.976 | ROC: 0.983
Early stopping (no improvement since 3 models) | Best ROC: 0.9862235439534969
Loaded pretrained weights for efficientnet-b2
---------- Fold: 5 ----------
0:04:54 | Epoch: 1/15 | Loss: 63.11 | Train Acc: 0.979 | Valid Acc: 0.979 | ROC: 0.992
0:04:53 | Epoch: 2/15 | Loss: 60.21 | Train Acc: 0.98 | Valid Acc: 0.98 | ROC: 0.989
0:04:55 | Epoch: 3/15 | Loss: 57.91 | Train Acc: 0.98 | Valid Acc: 0.98 | ROC: 0.991
Epoch     3: reducing learning rate of group 0 to 2.0000e-04.
0:04:56 | Epoch: 4/15 | Loss: 45.1 | Train Acc: 0.984 | Valid Acc: 0.981 | ROC: 0.992
0:04:56 | Epoch: 5/15 | Loss: 41.61 | Train Acc: 0.985 | Valid Acc: 0.979 | ROC: 0.991
0:04:56 | Epoch: 6/15 | Loss: 36.48 | Train Acc: 0.988 | Valid Acc: 0.982 | ROC: 0.992
Epoch     6: reducing learning rate of group 0 to 8.0000e-05.
0:04:57 | Epoch: 7/15 | Loss: 31.25 | Train Acc: 0.989 | Valid Acc: 0.982 | ROC: 0.992
Early stopping (no improvement since 3 models) | Best ROC: 0.9923051292463533
Loaded pretrained weights for efficientnet-b2
---------- Fold: 6 ----------
0:04:56 | Epoch: 1/15 | Loss: 62.28 | Train Acc: 0.98 | Valid Acc: 0.986 | ROC: 0.992
0:04:57 | Epoch: 2/15 | Loss: 57.42 | Train Acc: 0.981 | Valid Acc: 0.982 | ROC: 0.989
0:04:57 | Epoch: 3/15 | Loss: 59.15 | Train Acc: 0.979 | Valid Acc: 0.984 | ROC: 0.99
Epoch     3: reducing learning rate of group 0 to 2.0000e-04.
0:04:57 | Epoch: 4/15 | Loss: 44.86 | Train Acc: 0.984 | Valid Acc: 0.986 | ROC: 0.992
0:04:56 | Epoch: 5/15 | Loss: 39.38 | Train Acc: 0.986 | Valid Acc: 0.987 | ROC: 0.991
Epoch     5: reducing learning rate of group 0 to 8.0000e-05.
0:04:56 | Epoch: 6/15 | Loss: 34.3 | Train Acc: 0.988 | Valid Acc: 0.987 | ROC: 0.991
0:04:56 | Epoch: 7/15 | Loss: 30.06 | Train Acc: 0.99 | Valid Acc: 0.985 | ROC: 0.992
0:04:55 | Epoch: 8/15 | Loss: 27.28 | Train Acc: 0.991 | Valid Acc: 0.986 | ROC: 0.992
0:04:55 | Epoch: 9/15 | Loss: 26.81 | Train Acc: 0.99 | Valid Acc: 0.987 | ROC: 0.991
Epoch     9: reducing learning rate of group 0 to 3.2000e-05.
0:04:52 | Epoch: 10/15 | Loss: 24.69 | Train Acc: 0.991 | Valid Acc: 0.987 | ROC: 0.99
Early stopping (no improvement since 3 models) | Best ROC: 0.9922027185133513
Loaded pretrained weights for efficientnet-b2