我们知道，影响训练效果的主要是：
****train_ratio** 、**crop_fix_size**、**crop_ratio**、**lr**
、**batch** 、**epoch****
*****
因此测试一下：
### 一、当**仅改变train_ratio时**：
*****
#### ①**0.8**
train_ratio=0.8
crop_fix_size=(30,30)
crop_ratio=0.6
lr=0.05
batch=10
epoch=20
~~~
Epoch 20/20
369/369 [==============================] - 12s 31ms/step - loss: 0.7828 - acc: 0.6842 - val_loss: 0.8361 - val_acc: 0.6707
~~~
*****
②**0.85**
train_ratio=0.85
crop_fix_size=(30,30)
crop_ratio=0.6
lr=0.05
batch=10
epoch=20
~~~
Epoch 20/20
392/392 [==============================] - 10s 26ms/step - loss: 0.7734 - acc: 0.6806 - val_loss: 0.8888 - val_acc: 0.6739
~~~
~~~
The loss: 0.901434063911438, Test accuracy: 0.6681159138679504
~~~
*****
③**0.9**
train_ratio=0.9
crop_fix_size=(30,30)
crop_ratio=0.6
lr=0.05
batch=10
epoch=20
~~~
Epoch 20/20
293/293 [==============================] - 15s 51ms/step - loss: 0.8614 - acc: 0.6695 - val_loss: 0.8607 - val_acc: 0.6740
~~~
~~~
The loss: 0.8515320420265198, Test accuracy: 0.6739726066589355
~~~
*****
### 二、当**固定ratio：0.8，仅改变lr时**：
*****
#### ①**0.03**
train_ratio=0.8
crop_fix_size=(30,30)
crop_ratio=0.6
lr=0.03
batch=10
epoch=20
~~~
Epoch 20/20
293/293 [==============================] - 58s 198ms/step - loss: 1.0009 - acc: 0.5930 - val_loss: 1.0493 - val_acc: 0.5649
~~~
~~~
The loss: 1.063227849872145, Test accuracy: 0.568493127822876
~~~
#### ②**0.05**
train_ratio=0.8
crop_fix_size=(30,30)
crop_ratio=0.6
lr=0.05
batch=10
epoch=20
~~~
Epoch 20/20
293/293 [==============================] - 58s 198ms/step - loss: 0.9075 - acc: 0.6487 - val_loss: 0.9102 - val_acc: 0.6423
~~~
~~~
The loss: 0.9140146644148108, Test accuracy: 0.6547945141792297
~~~
#### ③**0.1**
train_ratio=0.8
crop_fix_size=(30,30)
crop_ratio=0.6
lr=0.1
batch=10
epoch=20
~~~
Epoch 20/20
293/293 [==============================] - 12s 43ms/step - loss: 0.8428 - acc: 0.6856 - val_loss: 0.8485 - val_acc: 0.6795
~~~
~~~
The loss: 0.8455102443695068, Test accuracy: 0.682191789150238
~~~
### 三、当**仅改变epoch时**：
*****
#### ①**10**
train_ratio=0.8
crop_fix_size=(30,30)
crop_ratio=0.6
lr=0.05
batch=10
epoch=10
~~~
Epoch 10/10
293/293 [==============================] - 12s 42ms/step - loss: 1.1422 - acc: 0.5048 - val_loss: 1.1555 - val_acc: 0.5027
~~~
~~~
The loss: 1.1439930200576782, Test accuracy: 0.4890410900115967
~~~
#### ②**30**
train_ratio=0.8
crop_fix_size=(30,30)
crop_ratio=0.6
lr=0.05
batch=10
epoch=30
~~~
Epoch 30/30
293/293 [==============================] - 12s 43ms/step - loss: 0.7440 - acc: 0.7119 - val_loss: 0.8707 - val_acc: 0.6575
~~~
~~~
The loss: 0.8830211758613586, Test accuracy: 0.6602739691734314
~~~
#### ③**40**
~~~
Epoch 28/40
293/293 [==============================] - 12s 43ms/step - loss: 0.7773 - acc: 0.7078 - val_loss: 0.8821 - val_acc: 0.6671
~~~
~~~
The loss: 0.882400631904602, Test accuracy: 0.6767123341560364
~~~
### 四、当**仅改变crop_ratio时**：
*****
#### ①**0.5**
train_ratio=0.8
crop_fix_size=(30,30)
crop_ratio=0.5
lr=0.05
batch=10
epoch=20
~~~
Epoch 20/20
293/293 [==============================] - 13s 44ms/step - loss: 0.8767 - acc: 0.6500 - val_loss: 0.9435 - val_acc: 0.6219
~~~
~~~
The loss: 0.9300558567047119, Test accuracy: 0.6164383292198181
~~~
#### ②**0.8**
train_ratio=0.8
crop_fix_size=(30,30)
crop_ratio=0.8
lr=0.05
batch=10
epoch=20
~~~
Epoch 20/20
293/293 [==============================] - 12s 43ms/step - loss: 0.9282 - acc: 0.6452 - val_loss: 0.8859 - val_acc: 0.6548
~~~
~~~
The loss: 0.8911147117614746, Test accuracy: 0.6424657702445984
~~~
