search

TakeruEndo / kaggle_Cassava

0 stars 1 forks source link

2. 有用なDataAUgmentation #2

Open TakeruEndo opened 3 years ago

TakeruEndo commented 3 years ago

https://albumentations.ai/docs/api_reference/augmentations/transforms/

TakeruEndo commented 3 years ago

Mixup?? https://qiita.com/karaage0703/items/9b56a68a5dba4ad330d9

TakeruEndo commented 3 years ago

Try Mixup Without Hesitation

TakeruEndo commented 3 years ago

Fmix + Mixupでスコア低下

Revised trainingは有用ではない！？！？

========== fold: 0 training ==========
Epoch 3 - Accuracy: 0.8782710280373832
[2021-01-19 14:32:36,548][utils][INFO] - Epoch 3 - Accuracy: 0.8782710280373832
Epoch 3 - Save Best Score: 0.8783 Model
[2021-01-19 14:32:36,549][utils][INFO] - Epoch 3 - Save Best Score: 0.8783 Model
Epoch 4 - avg_train_loss: 0.6643  avg_val_loss: 0.3848
[2021-01-19 15:02:40,592][utils][INFO] - Epoch 4 - avg_train_loss: 0.6643  avg_val_loss: 0.3848
Epoch 4 - Accuracy: 0.8817757009345795
[2021-01-19 15:02:40,593][utils][INFO] - Epoch 4 - Accuracy: 0.8817757009345795
Epoch 4 - Save Best Score: 0.8818 Model
[2021-01-19 15:02:40,593][utils][INFO] - Epoch 4 - Save Best Score: 0.8818 Model
Epoch 5 - avg_train_loss: 0.6305  avg_val_loss: 0.3556
[2021-01-19 15:32:33,357][utils][INFO] - Epoch 5 - avg_train_loss: 0.6305  avg_val_loss: 0.3556
Epoch 5 - Accuracy: 0.8880841121495328
[2021-01-19 15:32:33,358][utils][INFO] - Epoch 5 - Accuracy: 0.8880841121495328
Epoch 5 - Save Best Score: 0.8881 Model
[2021-01-19 15:32:33,358][utils][INFO] - Epoch 5 - Save Best Score: 0.8881 Model
Epoch 6 - avg_train_loss: 0.6195  avg_val_loss: 0.3543
[2021-01-19 16:02:08,498][utils][INFO] - Epoch 6 - avg_train_loss: 0.6195  avg_val_loss: 0.3543
Epoch 6 - Accuracy: 0.8864485981308411
[2021-01-19 16:02:08,499][utils][INFO] - Epoch 6 - Accuracy: 0.8864485981308411
Epoch 7 - avg_train_loss: 0.6092  avg_val_loss: 0.3483
[2021-01-19 16:31:45,754][utils][INFO] - Epoch 7 - avg_train_loss: 0.6092  avg_val_loss: 0.3483
Epoch 7 - Accuracy: 0.8911214953271028
[2021-01-19 16:31:45,754][utils][INFO] - Epoch 7 - Accuracy: 0.8911214953271028
Epoch 7 - Save Best Score: 0.8911 Model
[2021-01-19 16:31:45,755][utils][INFO] - Epoch 7 - Save Best Score: 0.8911 Model
Epoch 8 - avg_train_loss: 0.5894  avg_val_loss: 0.3425
[2021-01-19 17:01:29,666][utils][INFO] - Epoch 8 - avg_train_loss: 0.5894  avg_val_loss: 0.3425
Epoch 8 - Accuracy: 0.8922897196261682
[2021-01-19 17:01:29,667][utils][INFO] - Epoch 8 - Accuracy: 0.8922897196261682
Epoch 8 - Save Best Score: 0.8923 Model
[2021-01-19 17:01:29,667][utils][INFO] - Epoch 8 - Save Best Score: 0.8923 Model
Epoch 9 - avg_train_loss: 0.5817  avg_val_loss: 0.3346
[2021-01-19 17:31:10,332][utils][INFO] - Epoch 9 - avg_train_loss: 0.5817  avg_val_loss: 0.3346
Epoch 9 - Accuracy: 0.8932242990654206
[2021-01-19 17:31:10,333][utils][INFO] - Epoch 9 - Accuracy: 0.8932242990654206
Epoch 9 - Save Best Score: 0.8932 Model
[2021-01-19 17:31:10,333][utils][INFO] - Epoch 9 - Save Best Score: 0.8932 Model
Epoch 10 - avg_train_loss: 0.5737  avg_val_loss: 0.3354
[2021-01-19 18:00:47,430][utils][INFO] - Epoch 10 - avg_train_loss: 0.5737  avg_val_loss: 0.3354
Epoch 10 - Accuracy: 0.8936915887850467
[2021-01-19 18:00:47,431][utils][INFO] - Epoch 10 - Accuracy: 0.8936915887850467
Epoch 10 - Save Best Score: 0.8937 Model
[2021-01-19 18:00:47,432][utils][INFO] - Epoch 10 - Save Best Score: 0.8937 Model
========== fold: 0 result ==========
Score - Accuracy: 0.8936915887850467
[2021-01-19 18:00:50,651][utils][INFO] - Score - Accuracy: 0.8936915887850467
========== fold: 0 Revised training ==========
Epoch 1 - avg_train_loss: 0.3655  avg_val_loss: 0.3893
[2021-01-19 18:16:04,887][utils][INFO] - Epoch 1 - avg_train_loss: 0.3655  avg_val_loss: 0.3893
Epoch 1 - Accuracy: 0.8689252336448599
[2021-01-19 18:16:04,888][utils][INFO] - Epoch 1 - Accuracy: 0.8689252336448599
Epoch 2 - avg_train_loss: 0.3351  avg_val_loss: 0.3554
[2021-01-19 18:31:12,166][utils][INFO] - Epoch 2 - avg_train_loss: 0.3351  avg_val_loss: 0.3554
Epoch 2 - Accuracy: 0.8817757009345795
[2021-01-19 18:31:12,166][utils][INFO] - Epoch 2 - Accuracy: 0.8817757009345795
Epoch 3 - avg_train_loss: 0.3132  avg_val_loss: 0.3596
[2021-01-19 18:46:20,449][utils][INFO] - Epoch 3 - avg_train_loss: 0.3132  avg_val_loss: 0.3596
Epoch 3 - Accuracy: 0.8850467289719626
[2021-01-19 18:46:20,450][utils][INFO] - Epoch 3 - Accuracy: 0.8850467289719626