-
Notifications
You must be signed in to change notification settings - Fork 0
/
note
54 lines (49 loc) · 4.2 KB
/
note
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
During that challenge I did two different trainings. However, I submitted my best result which is my second training. So, I think that to review the whole results well, the following extra explanation will be better for reviewer.
My submitted code is building a model with respect to "EarlyStopping(monitor='val_loss', patience=3)". It means that after the third worst result based on 'val_loss', the training phase was terminated. The best model is M1 = [loss: 0.2716 - acc: 0.9175 - val_loss: 0.8367 - val_acc: 0.7611] I got in the first training which I forgot to forward to you the below. So, Here you can find my first training results. If you run my code once, you will get same results roughly since you don't have any initial calculated weights.
=================================== 1st training ===================================
Train on 29458 samples, validate on 3274 samples
Epoch 1/100
29458/29458 [==============================] - 134s - loss: 2.3073 - acc: 0.1970 - val_loss: 2.3132 - val_acc: 0.1735
Epoch 2/100
29458/29458 [==============================] - 136s - loss: 2.0656 - acc: 0.3058 - val_loss: 1.7405 - val_acc: 0.4291
Epoch 3/100
29458/29458 [==============================] - 163s - loss: 1.7875 - acc: 0.4097 - val_loss: 1.6554 - val_acc: 0.4404
Epoch 4/100
29458/29458 [==============================] - 149s - loss: 1.5521 - acc: 0.4861 - val_loss: 1.6295 - val_acc: 0.4908
Epoch 5/100
29458/29458 [==============================] - 151s - loss: 1.3739 - acc: 0.5496 - val_loss: 1.4229 - val_acc: 0.5516
Epoch 6/100
29458/29458 [==============================] - 147s - loss: 1.1621 - acc: 0.6194 - val_loss: 1.2519 - val_acc: 0.6045
Epoch 7/100
29458/29458 [==============================] - 164s - loss: 0.9966 - acc: 0.6749 - val_loss: 1.1696 - val_acc: 0.6222
Epoch 8/100
29458/29458 [==============================] - 153s - loss: 0.8049 - acc: 0.7390 - val_loss: 1.0118 - val_acc: 0.6848
Epoch 9/100
29458/29458 [==============================] - 151s - loss: 0.6382 - acc: 0.7954 - val_loss: 0.9054 - val_acc: 0.7208
Epoch 10/100
29458/29458 [==============================] - 151s - loss: 0.4903 - acc: 0.8453 - val_loss: 0.9261 - val_acc: 0.7202
Epoch 11/100
29458/29458 [==============================] - 154s - loss: 0.3685 - acc: 0.8857 - val_loss: 0.8590 - val_acc: 0.7459
Epoch 12/100
29458/29458 [==============================] - 156s - loss: 0.2716 - acc: 0.9175 - val_loss: 0.8367 - val_acc: 0.7611
Epoch 13/100
29458/29458 [==============================] - 153s - loss: 0.2017 - acc: 0.9411 - val_loss: 0.9164 - val_acc: 0.7569
Epoch 14/100
29458/29458 [==============================] - 152s - loss: 0.1575 - acc: 0.9546 - val_loss: 0.8574 - val_acc: 0.7804
Epoch 15/100
29458/29458 [==============================] - 152s - loss: 0.1194 - acc: 0.9644 - val_loss: 0.9226 - val_acc: 0.7670
Epoch 16/100
29458/29458 [==============================] - 154s - loss: 0.0987 - acc: 0.9731 - val_loss: 0.9343 - val_acc: 0.7795
You only have second training results, below, since during this training I used best model from first training, this second training I submitted in the notebook file is using the previous calculated weights (M1) from the first training. That is the reason why this second training is starting high 'val_acc' and low 'val_loss' values
According to this whole results, the test train file (ytest.txt) I submitted was labeled by the best model which is calculated in the second training (M2 = [loss: 0.2088 - acc: 0.9372 - val_loss: 0.8074 - val_acc: 0.7755]) on top of first training (M1).
=================================== 2nd training ===================================
Epoch 1/100
29458/29458 [==============================] - 243s - loss: 0.2088 - acc: 0.9372 - val_loss: 0.8074 - val_acc: 0.7755
Epoch 2/100
29458/29458 [==============================] - 139s - loss: 0.1673 - acc: 0.9509 - val_loss: 0.8821 - val_acc: 0.7688
Epoch 3/100
29458/29458 [==============================] - 129s - loss: 0.1385 - acc: 0.9594 - val_loss: 0.8535 - val_acc: 0.7783
Epoch 4/100
29458/29458 [==============================] - 131s - loss: 0.1155 - acc: 0.9666 - val_loss: 0.8926 - val_acc: 0.7776
Epoch 5/100
29458/29458 [==============================] - 130s - loss: 0.0961 - acc: 0.9734 - val_loss: 0.9319 - val_acc: 0.7770