Tensorflow2（预课程）---4.1、逻辑回归实例-层方式
一、总结
一句话总结：
可以看到，相比于mse损失函数，cross entropy函数无论是收敛速度，还是最后的测试集的准确率都更加优秀
# 构建容器
model = tf.keras.Sequential()
# 输出层
model.add(tf.keras.Input(shape=(15,)))
# 中间层
model.add(tf.keras.layers.Dense(10,activation='relu'))
model.add(tf.keras.layers.Dense(10,activation='relu'))
# 输出层
model.add(tf.keras.layers.Dense(1,activation='sigmoid'))
# 配置优化函数和损失器
model.compile(optimizer='adam',loss='binary_crossentropy',metrics=['acc'])
# 开始训练
history = model.fit(train_x,train_y,epochs=5000,validation_data=(test_x,test_y))
二、逻辑回归实例-层方式
博客对应课程的视频位置：
步骤
1、读取数据集
2、拆分数据集（拆分成训练数据集和测试数据集）
3、构建模型
4、训练模型
5、检验模型
需求
对credit数据集进行预测，也就是根据已有数据，来预测后面的情况
Out[2]:
| | 0 | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 |
|---|
| 0 |
0 |
30.83 |
0.000 |
0 |
0 |
9 |
0 |
1.25 |
0 |
0 |
1 |
1 |
0 |
202 |
0.0 |
-1 |
|---|
| 1 |
1 |
58.67 |
4.460 |
0 |
0 |
8 |
1 |
3.04 |
0 |
0 |
6 |
1 |
0 |
43 |
560.0 |
-1 |
|---|
| 2 |
1 |
24.50 |
0.500 |
0 |
0 |
8 |
1 |
1.50 |
0 |
1 |
0 |
1 |
0 |
280 |
824.0 |
-1 |
|---|
| 3 |
0 |
27.83 |
1.540 |
0 |
0 |
9 |
0 |
3.75 |
0 |
0 |
5 |
0 |
0 |
100 |
3.0 |
-1 |
|---|
| 4 |
0 |
20.17 |
5.625 |
0 |
0 |
9 |
0 |
1.71 |
0 |
1 |
0 |
1 |
2 |
120 |
0.0 |
-1 |
|---|
| ... |
... |
... |
... |
... |
... |
... |
... |
... |
... |
... |
... |
... |
... |
... |
... |
... |
|---|
| 648 |
0 |
21.08 |
10.085 |
1 |
1 |
11 |
1 |
1.25 |
1 |
1 |
0 |
1 |
0 |
260 |
0.0 |
1 |
|---|
| 649 |
1 |
22.67 |
0.750 |
0 |
0 |
0 |
0 |
2.00 |
1 |
0 |
2 |
0 |
0 |
200 |
394.0 |
1 |
|---|
| 650 |
1 |
25.25 |
13.500 |
1 |
1 |
13 |
7 |
2.00 |
1 |
0 |
1 |
0 |
0 |
200 |
1.0 |
1 |
|---|
| 651 |
0 |
17.92 |
0.205 |
0 |
0 |
12 |
0 |
0.04 |
1 |
1 |
0 |
1 |
0 |
280 |
750.0 |
1 |
|---|
| 652 |
0 |
35.00 |
3.375 |
0 |
0 |
0 |
1 |
8.29 |
1 |
1 |
0 |
0 |
0 |
0 |
0.0 |
1 |
|---|
653 rows × 16 columns
这种-1和1的二分类问题，完全不需要one_hot编码
直接输出层激活函数sigmoid即可，因为sigmoid的y的取值是0-1
Out[3]:
| | 0 | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 |
|---|
| 0 |
1 |
17.83 |
11.000 |
0 |
0 |
10 |
1 |
1.000 |
0 |
0 |
11 |
1 |
0 |
0 |
3000.0 |
-1 |
|---|
| 1 |
1 |
22.50 |
11.000 |
1 |
1 |
8 |
0 |
3.000 |
0 |
1 |
0 |
0 |
0 |
268 |
0.0 |
1 |
|---|
| 2 |
0 |
47.33 |
6.500 |
0 |
0 |
0 |
0 |
1.000 |
1 |
1 |
0 |
0 |
0 |
0 |
228.0 |
1 |
|---|
| 3 |
0 |
31.08 |
1.500 |
1 |
1 |
9 |
0 |
0.040 |
1 |
1 |
0 |
1 |
2 |
160 |
0.0 |
1 |
|---|
| 4 |
0 |
38.17 |
10.125 |
0 |
0 |
10 |
0 |
2.500 |
0 |
0 |
6 |
1 |
0 |
520 |
196.0 |
-1 |
|---|
| ... |
... |
... |
... |
... |
... |
... |
... |
... |
... |
... |
... |
... |
... |
... |
... |
... |
|---|
| 648 |
0 |
16.50 |
0.125 |
0 |
0 |
0 |
0 |
0.165 |
1 |
1 |
0 |
1 |
0 |
132 |
0.0 |
1 |
|---|
| 649 |
1 |
41.00 |
2.040 |
1 |
1 |
8 |
1 |
0.125 |
0 |
0 |
23 |
0 |
0 |
455 |
1236.0 |
-1 |
|---|
| 650 |
0 |
23.92 |
0.665 |
0 |
0 |
0 |
0 |
0.165 |
1 |
1 |
0 |
1 |
0 |
100 |
0.0 |
-1 |
|---|
| 651 |
0 |
31.08 |
3.085 |
0 |
0 |
0 |
0 |
2.500 |
1 |
0 |
2 |
0 |
0 |
160 |
41.0 |
1 |
|---|
| 652 |
1 |
33.67 |
0.375 |
0 |
0 |
2 |
0 |
0.375 |
1 |
1 |
0 |
1 |
0 |
300 |
44.0 |
-1 |
|---|
653 rows × 16 columns
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14
0 1 17.83 11.000 0 0 10 1 1.000 0 0 11 1 0 0 3000.0
1 1 22.50 11.000 1 1 8 0 3.000 0 1 0 0 0 268 0.0
2 0 47.33 6.500 0 0 0 0 1.000 1 1 0 0 0 0 228.0
3 0 31.08 1.500 1 1 9 0 0.040 1 1 0 1 2 160 0.0
4 0 38.17 10.125 0 0 10 0 2.500 0 0 6 1 0 520 196.0
.. .. ... ... .. .. .. .. ... .. .. .. .. .. ... ...
595 1 64.08 0.165 0 0 13 7 0.000 0 0 1 1 0 232 100.0
596 0 29.83 2.040 1 1 10 1 0.040 1 1 0 1 0 128 1.0
597 0 34.50 4.040 1 1 3 2 8.500 0 0 7 0 0 195 0.0
598 0 36.33 3.790 0 0 9 0 1.165 0 1 0 0 0 200 0.0
599 0 22.67 1.585 1 1 9 0 3.085 0 0 6 1 0 80 0.0
[600 rows x 15 columns]
0 0
1 1
2 1
3 1
4 0
..
595 0
596 1
597 0
598 1
599 0
Name: 15, Length: 600, dtype: int64
输入是15维，输出1维，输出二分类问题，不需要one_hot
输出换成0和1，这样输出层激活函数指定sigmoid，就不需要做其它操作了
sigmoid输出的本身就是0-1之间的概率，softmax也是表示的概率
所以模型应该是15->n->1
Model: "sequential"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
dense (Dense) (None, 10) 160
_________________________________________________________________
dense_1 (Dense) (None, 10) 110
_________________________________________________________________
dense_2 (Dense) (None, 1) 11
=================================================================
Total params: 281
Trainable params: 281
Non-trainable params: 0
_________________________________________________________________
对于特征比较多的神经网络，第一层的神经元最好也多一点，这样效果要好
当第一层神经元的个数为4的时候，5000次的epoch的成功率只有0.5+，
而第一层神经元的个数为10的时候，5000次的epoch的成功率有8.3+，
但是也不能太多
Epoch 1/5000
19/19 [==============================] - 0s 9ms/step - loss: 17.5357 - acc: 0.4467 - val_loss: 9.5537 - val_acc: 0.4717
Epoch 2/5000
19/19 [==============================] - 0s 3ms/step - loss: 8.2702 - acc: 0.5667 - val_loss: 3.3894 - val_acc: 0.8113
Epoch 3/5000
19/19 [==============================] - 0s 3ms/step - loss: 2.7807 - acc: 0.6450 - val_loss: 1.2214 - val_acc: 0.7925
Epoch 4/5000
19/19 [==============================] - 0s 3ms/step - loss: 1.8072 - acc: 0.6667 - val_loss: 0.8503 - val_acc: 0.7358
Epoch 5/5000
19/19 [==============================] - 0s 3ms/step - loss: 1.2854 - acc: 0.6333 - val_loss: 0.8329 - val_acc: 0.7736
Epoch 6/5000
19/19 [==============================] - 0s 3ms/step - loss: 1.0678 - acc: 0.6250 - val_loss: 0.7051 - val_acc: 0.7358
Epoch 7/5000
19/19 [==============================] - 0s 3ms/step - loss: 0.8744 - acc: 0.6350 - val_loss: 0.7378 - val_acc: 0.6981
Epoch 8/5000
19/19 [==============================] - 0s 3ms/step - loss: 0.7846 - acc: 0.6433 - val_loss: 0.6449 - val_acc: 0.7170
Epoch 9/5000
19/19 [==============================] - 0s 3ms/step - loss: 0.7270 - acc: 0.6633 - val_loss: 0.5953 - val_acc: 0.7170
Epoch 10/5000
19/19 [==============================] - 0s 3ms/step - loss: 0.7340 - acc: 0.6750 - val_loss: 0.7433 - val_acc: 0.8113
Epoch 11/5000
19/19 [==============================] - 0s 3ms/step - loss: 0.8346 - acc: 0.6850 - val_loss: 0.6981 - val_acc: 0.8113
Epoch 12/5000
19/19 [==============================] - 0s 3ms/step - loss: 0.7294 - acc: 0.6717 - val_loss: 0.6034 - val_acc: 0.8113
Epoch 13/5000
19/19 [==============================] - 0s 3ms/step - loss: 0.6561 - acc: 0.6917 - val_loss: 0.5421 - val_acc: 0.7170
Epoch 14/5000
19/19 [==============================] - 0s 3ms/step - loss: 0.6122 - acc: 0.6833 - val_loss: 0.4921 - val_acc: 0.7736
Epoch 15/5000
19/19 [==============================] - 0s 3ms/step - loss: 0.6230 - acc: 0.7033 - val_loss: 0.4789 - val_acc: 0.8113
Epoch 16/5000
19/19 [==============================] - 0s 3ms/step - loss: 0.5656 - acc: 0.7100 - val_loss: 0.4925 - val_acc: 0.8113
Epoch 17/5000
19/19 [==============================] - 0s 3ms/step - loss: 0.7055 - acc: 0.6950 - val_loss: 0.6540 - val_acc: 0.8113
Epoch 18/5000
19/19 [==============================] - 0s 3ms/step - loss: 0.6596 - acc: 0.7083 - val_loss: 0.4843 - val_acc: 0.7736
Epoch 19/5000
19/19 [==============================] - 0s 3ms/step - loss: 0.6000 - acc: 0.7067 - val_loss: 0.4473 - val_acc: 0.7925
Epoch 20/5000
19/19 [==============================] - 0s 3ms/step - loss: 0.6648 - acc: 0.7067 - val_loss: 0.4772 - val_acc: 0.7358
Epoch 21/5000
19/19 [==============================] - 0s 3ms/step - loss: 2.0961 - acc: 0.6333 - val_loss: 2.5110 - val_acc: 0.7925
Epoch 22/5000
19/19 [==============================] - 0s 3ms/step - loss: 2.3280 - acc: 0.7033 - val_loss: 0.8230 - val_acc: 0.7358
Epoch 23/5000
19/19 [==============================] - 0s 3ms/step - loss: 1.0403 - acc: 0.6950 - val_loss: 0.8077 - val_acc: 0.7925
..........................
Epoch 4995/5000
19/19 [==============================] - 0s 3ms/step - loss: 0.1757 - acc: 0.9217 - val_loss: 2.0568 - val_acc: 0.8113
Epoch 4996/5000
19/19 [==============================] - 0s 3ms/step - loss: 0.1735 - acc: 0.9233 - val_loss: 2.0542 - val_acc: 0.8113
Epoch 4997/5000
19/19 [==============================] - 0s 3ms/step - loss: 0.1749 - acc: 0.9250 - val_loss: 2.1591 - val_acc: 0.8113
Epoch 4998/5000
19/19 [==============================] - 0s 3ms/step - loss: 0.1788 - acc: 0.9267 - val_loss: 2.2617 - val_acc: 0.8302
Epoch 4999/5000
19/19 [==============================] - 0s 3ms/step - loss: 0.1826 - acc: 0.9283 - val_loss: 2.3343 - val_acc: 0.7925
Epoch 5000/5000
19/19 [==============================] - 0s 3ms/step - loss: 0.1716 - acc: 0.9267 - val_loss: 2.3872 - val_acc: 0.8113
可以看到，相比于mse损失函数，cross entropy函数无论是收敛速度，还是最后的测试集的准确率都更加优秀
[[3.8453022e-21]
[8.5991585e-01]
[6.5455580e-10]
[1.0000000e+00]
[1.0000000e+00]
[1.0000000e+00]
[9.5815730e-01]
[6.3400120e-02]
[9.4178385e-01]
[9.8131865e-01]
[1.0000000e+00]
[2.5722989e-01]
[9.6723849e-01]
[3.5510752e-01]
[2.1926977e-03]
[3.4676839e-03]
[4.1911963e-02]
[4.1010199e-07]
[8.6172342e-01]
[7.0969802e-01]
[7.7412263e-02]
[7.5796050e-01]
[1.0000000e+00]
[0.0000000e+00]
[2.3604299e-01]
[1.7066821e-02]
[1.0000000e+00]
[9.6950209e-01]
[2.4396995e-01]
[2.1174717e-01]
[9.6668184e-01]
[9.6301866e-01]
[2.1353349e-02]
[1.0000000e+00]
[4.1701463e-01]
[1.0000000e+00]
[2.8559173e-04]
[7.4790055e-01]
[1.0000000e+00]
[8.7733787e-01]
[4.8705450e-01]
[2.5270924e-01]
[1.0000000e+00]
[9.7821814e-01]
[7.8488225e-01]
[1.0000000e+00]
[3.1834045e-01]
[1.4321352e-06]
[1.0000000e+00]
[1.1380038e-01]
[1.0000000e+00]
[9.3310910e-01]
[9.6864295e-01]]
600 0
601 1
602 0
603 1
604 1
605 1
606 1
607 0
608 1
609 1
610 1
611 1
612 1
613 1
614 0
615 0
616 0
617 0
618 1
619 1
620 0
621 1
622 1
623 0
624 0
625 0
626 1
627 1
628 1
629 0
630 1
631 1
632 0
633 0
634 1
635 1
636 0
637 0
638 0
639 1
640 0
641 0
642 1
643 1
644 1
645 1
646 1
647 0
648 1
649 0
650 0
651 1
652 0
Name: 15, dtype: int64
[0 1 0 1 1 1 1 0 1 1 1 1 1 1 0 0 0 0 1 1 0 1 1 0 0 0 1 1 1 0 1 1 0 0 1 1 0
0 0 1 0 0 1 1 1 1 1 0 1 0 0 1 0]
[3.8453022e-21 8.5991585e-01 6.5455580e-10 1.0000000e+00 1.0000000e+00
1.0000000e+00 9.5815730e-01 6.3400120e-02 9.4178385e-01 9.8131865e-01
1.0000000e+00 2.5722989e-01 9.6723849e-01 3.5510752e-01 2.1926977e-03
3.4676839e-03 4.1911963e-02 4.1010199e-07 8.6172342e-01 7.0969802e-01
7.7412263e-02 7.5796050e-01 1.0000000e+00 0.0000000e+00 2.3604299e-01
1.7066821e-02 1.0000000e+00 9.6950209e-01 2.4396995e-01 2.1174717e-01
9.6668184e-01 9.6301866e-01 2.1353349e-02 1.0000000e+00 4.1701463e-01
1.0000000e+00 2.8559173e-04 7.4790055e-01 1.0000000e+00 8.7733787e-01
4.8705450e-01 2.5270924e-01 1.0000000e+00 9.7821814e-01 7.8488225e-01
1.0000000e+00 3.1834045e-01 1.4321352e-06 1.0000000e+00 1.1380038e-01
1.0000000e+00 9.3310910e-01 9.6864295e-01]
需求：
tensorflow 大于某个值为1,小于某个值为0
tf.Tensor(
[0 1 0 1 1 1 1 0 1 1 1 0 1 0 0 0 0 0 1 1 0 1 1 0 0 0 1 1 0 0 1 1 0 1 0 1 0
1 1 1 0 0 1 1 1 1 0 0 1 0 1 1 1], shape=(53,), dtype=int32)
[0 1 0 1 1 1 1 0 1 1 1 1 1 1 0 0 0 0 1 1 0 1 1 0 0 0 1 1 1 0 1 1 0 0 1 1 0
0 0 1 0 0 1 1 1 1 1 0 1 0 0 1 0]
