# Model Train and Test
###### tags: `鍾`
## Dataset
DatasetNT
### Train
Cataract: 3562
Normal: 4074
### Test
Cataract: 902
Normal: 1030

---
### -dropout (w:without dropout)
**patince=10**
e50p10 1st : modele50p10w.h5
acc: 74%
e50p10 2nd : modele50p10we.h5
acc: 71%

e100p10 : modele100p10w.h5
acc: 77%

e300p10: modele300p10w.h5
acc: 73%

e500p10: modele500p10w.h5
acc: 66%

e1000p10: modele1000p10w.h5
acc: 71%

比對3/28模型訓練1
**patince=5**
e50p5 (i guess maybe this is the before item) : modele50p5w.h5
acc:71%

e100p5 : modele100p5w.h5
acc: 82%

e300p5 : modele300p5w.h5
acc: 75%

e500p5 : modele500p5w.h5
acc: 79%

e1000p5 : modele1000p5w.h5
acc: 79%

**Q:有一些model測normal會全對，但有些測cataract會全對，可能要找一個對兩者都較好的參數設定。**
**Q:是否patience小會比較好呢？**
- +dropout patience=5 ...

---
#### Test: Patience at 10 would it stop at the same epoch?
**patience = 10** (+dropout=0.5)
##### test 1st : Earlye50p10.h5
Epoch: 50
Patience : **10**
stop at 34
correct: 52
error: 22
total: 74
acc: **70.27%**

##### test 2nd : Earlye100p10.h5
Epoch: 100
Patience: 10
stop at 81
correct: 37
error: 37
total: 74
acc: **50%**

##### test 3rd : Earlye300p10.h5
Epoch: 300
Patience: 10
stop at 57
correct: 58
error: 16
total: 74
acc: **78.37%**

##### test 4th : Earlye500p10.h5
Epoch: 500
Patience: 10
stop at 67
correct: 41
error: 33
total: 74
acc: **55.4%**

##### test 5th : Earlye1000p10.h5
Epoch: 1000
Patience: 10
stop at 46
correct: 41
error: 33
total: 74
acc: **55.4%**

---
**patience = 5** (+dropout)
e50p5: modele50p5.h5
acc: 74%

e100p5: modele100p5.h5
acc: 52%

e300p5: modele300p5.h5
acc: 70%

e500p5: modele500p5.h5
acc: 78%

e1000p5: modele1000p5.h5
acc: 67%

---
每個test的停頓點我們最初理想是會都停在一樣的點，但現在測完發現會停頓在不同地方？原因？

#### Test: Would there deeper valley when there longer epoch?
+dropout
**checkpoint**
##### test 1st : best_modele50.h5 & modele50.h5
Epoch: 50
Patience: 50
min val_loss:0.04178 at 37
C: 46
E: 28
total: 74
acc: 0.62 & 59%

##### test 2nd : best_modele100.h5 & modele100.h5
Epoch: 100
Patience: 100
min val_loss:0.03032 at 96
C: 39
E: 35
total: 74
acc: 0.52 & 56%

##### test 3rd : best_modele300.h5 & modele300.h5
Epoch: 300
Patience: 300
min val_loss:0.02123 at 257
C: 51
E: 23
total: 74
acc: 0.69 & 67%

##### test 4th : best_modele500.h5 & modele500.h5
Epoch: 500
Patience: 500
min val_loss:0.01703 at 388
C: 45
E: 29
total:
acc: 0.60 & 62%

##### test 5th : best_modele1000.h5 & modele1000.h5
Epoch: 1000
Patience: 1000
min val_loss:0.01716 at N/A
C: 51
E: 23
total:
acc: 0.69 & 56%

---
-dropout
**checkpoint**
e50:
min val_loss: 0.06698 at 38
acc: 82%

e100:
acc:71%

e300:
acc:79%

e500:
acc: 82%
min val_loss: 0.06605

e1000:
acc:85%
min val_loss: 0.06806

**e1000w:** best_modele1000w.h5
acc:
min val_loss: 0.04738

test data2
total: 74
cataract correct:23(tp) error:14(fn)
normal correct:32(tn) error:5(fp)
acc:74%

test data1 100
total: 100
cataract correct:37(tp) error:13(fn)
normal correct:41(tn) error:9(fp)
acc:78%

test data2
total: 89 (all)
cataract correct:27(tp) error:16(fn)
normal correct:40(tn) error:6(fp)
acc:75%

e10002
acc:83%
test data2
total: 74
cataract correct:33(tp) error:4(fn)
normal correct:29(tn) error:8(fp)

---

#### Test: Find the best of Dropout Rate!
e = 25
#### test 1st : best_drop01e25.h5 & drop01e25.h5
val_loss min at:17
val_loss=0.07503
acc:

test data2
total: 74
cataract correct:14(tp) error:23(fn)
normal correct:35(tn) error:2(fp)
acc:66%

test data1 100
total: 100
cataract correct:38(tp) error:12(fn)
normal correct:44(tn) error:6(fp)
acc:82%

#### test 2nd : best_drop02e25.h5 & drop02e25.h5
val_loss min at:25
val_loss:0.05583
acc:

test data2
total: 74
cataract correct:29(tp) error:8(fn)
normal correct:32(tn) error:5(fp)
**acc:82%**

test data1 100
total: 100
cataract correct:42(tp) error:8(fn)
normal correct:35(tn) error:15(fp)
acc:77%

#### test 3rd : best_drop03e25.h5 & drop03e25.h5
val_loss min at:
val_loss:
acc:

test data2
total: 74
cataract correct:15(tp) error:22(fn)
normal correct:34(tn) error:3(fp)
acc:66%

test data1 100
total: 100
cataract correct:31(tp) error:19(fn)
normal correct:40(tn) error:10(fp)
acc:71%

#### test 4 : best_drop04e25.h5 & drop04e25.h5
val_loss min at: 20
val_loss: 0.04860
acc:

test data2
total: 74
cataract correct:27(tp) error:10(fn)
normal correct:25(tn) error:12(fp)
acc:70%

test data1 100
total: 100
cataract correct:41(tp) error:9(fn)
normal correct:30(tn) error:20(fp)
acc:71%

#### test 5 : best_drop05e25.h5 & drop05e25.h5
val_loss min at:25
val_loss:0.5085
acc:

test data2
total: 74
cataract correct:2(tp) error:35(fn)
normal correct:33(tn) error:4(fp)
acc:47%

test data1 100
total: 100
cataract correct:29(tp) error:21(fn)
normal correct:46(tn) error:4(fp)
acc:75%

#### test 6 : best_drop06e25.h5 & drop06e25.h5
val_loss min at: 25
val_loss: 0.07014
acc:

test data2
total: 74
cataract correct: 24(tp) error:13(fn)
normal correct: 30(tn) error:7(fp)
acc: 58%

test data1 100
total: 100
cataract correct: 32(tp) error:18(fn)
normal correct: 35(tn) error:15(fp)
acc: 67%

#### test 7 : best_drop07e25.h5 & drop07e25.h5
val_loss min at: 23
val_loss: 0.09994
acc:

test data2
total: 74
cataract correct: 32(tp) error: 5(fn)
normal correct: 15(tn) error: 22(fp)
acc: 63.5%

test data1 100
total: 100
cataract correct: 37(tp) error: 13(fn)
normal correct: 14(tn) error: 36(fp)
acc: 51%

#### test 8 : best_drop08e25.h5 & drop08e25.h5
val_loss min at:22
val_loss:0.12484
acc:

test data2
total: 74
cataract correct:6(tp) error:31(fn)
normal correct:30(tn) error:7(fp)
acc:49%

test data1 100
total: 100
cataract correct:30(tp) error:20(fn)
normal correct:45(tn) error:5(fp)
acc:75%

#### test 9 : best_drop09e25.h5 & drop09e25.h5
val_loss min at:25
val_loss:0.20343
acc:

test data2
total: 74
cataract correct:2(tp) error:35(fn)
normal correct:36(tn) error:1(fp)
acc:51%

test data1 100
total: 100
cataract correct:28(tp) error:22(fn)
normal correct:47(tn) error:3(fp)
acc:75%

---
## lab 107 with GPU
### Device Detail
- GPU: NVIDIA GeForce GTX 1080
- CPU: Intel(R) Core(TM) i7-7700 CPU @ 3.6oGHz
- Memory: 32.0GB
- TensorFlow: 2.3.0
- Keras: 2.4.3
- h5py: 2.10.0
- matplotlib: 3.4.1
- numpy: 1.20.2

### checkpoint, without dropout, epochs=1000
#### 1st
- result

---
checkpoint dropout=0.2 epochs=1000
- best_modeld02e1000.h5