tensorflow2_tf.keras实现逻辑回归

逻辑回归与线性回归的区别是在输出层加入 sigmoid函数。

使用的数据集是信用卡欺诈数据集credit-a

下载地址：链接：https://pan.baidu.com/s/1eGyb1bzT3x3BLakNmjI7tA 提取码：500l

处理过程

import tensorflow as tf
import pandas as pd
import matplotlib.pyplot as plt
#读取数据集
data = pd.read_csv('dataset/credit-a.csv')
data.head()

	0	30.83	0.1	0.2	0.3	9	0.4	1.25	0.5	0.6	1	1.1	0.7	202	0.8	-1
0	1	58.67	4.460	0	0	8	1	3.04	0	0	6	1	0	43	560.0	-1
1	1	24.50	0.500	0	0	8	1	1.50	0	1	0	1	0	280	824.0	-1
2	0	27.83	1.540	0	0	9	0	3.75	0	0	5	0	0	100	3.0	-1
3	0	20.17	5.625	0	0	9	0	1.71	0	1	0	1	2	120	0.0	-1
4	0	32.08	4.000	0	0	6	0	2.50	0	1	0	0	0	360	0.0	-1

#从上可以看出此数据集没有表头，把第一行数据当成了表头，重读一遍
data = pd.read_csv('dataset/credit-a.csv',header=None)
data.head()

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

#查看第15列结果有几类
data.iloc[:,-1].value_counts()

1 357 -1 296 Name: 15, dtype: int64

是个二分类问题，-1  1 用于SVM，我们要用逻辑回归对其进行处理，所以先把-1 全部替换成0

#构造x,y
x = data.iloc[:,:-1]
y = data.iloc[:,-1].replace(-1,0)
#构建一个 输入为 15 隐藏层为 10 10 输出层为1的神经网络，由于是逻辑回归，最后输出层的激活函数为sigmoid
model = tf.keras.Sequential([
    tf.keras.layers.Dense(10,input_shape=(15,),activation='relu'),
    tf.keras.layers.Dense(10,activation='relu'),
    tf.keras.layers.Dense(1,activation='sigmoid')
])
model.summary()

#设置优化器、损失函数
model.compile(
    optimizer = 'adam',  #优化器 
    loss='binary_crossentropy', #损失函数，交叉熵
    metrics=['acc']  #准确率
)
#训练80次
history = model.fit(x,y,epochs=80)

In [17]:

#训练80次 history = model.fit(x,y,epochs=80)

x

#训练80次

history = model.fit(x,y,epochs=80)

Train on 653 samples
Epoch 1/80
653/653 [==============================] - 0s 59us/sample - loss: 0.4691 - acc: 0.7963
Epoch 2/80
653/653 [==============================] - 0s 43us/sample - loss: 0.3971 - acc: 0.8438
Epoch 3/80
653/653 [==============================] - 0s 43us/sample - loss: 0.3794 - acc: 0.8484
Epoch 4/80
653/653 [==============================] - 0s 55us/sample - loss: 0.3665 - acc: 0.8545
Epoch 5/80
653/653 [==============================] - 0s 62us/sample - loss: 0.4101 - acc: 0.8300
Epoch 6/80
653/653 [==============================] - 0s 45us/sample - loss: 0.3759 - acc: 0.8423
Epoch 7/80
653/653 [==============================] - 0s 37us/sample - loss: 0.3732 - acc: 0.8423
Epoch 8/80
653/653 [==============================] - 0s 39us/sample - loss: 0.3724 - acc: 0.8530
Epoch 9/80
653/653 [==============================] - 0s 57us/sample - loss: 0.3982 - acc: 0.8453
Epoch 10/80
653/653 [==============================] - 0s 58us/sample - loss: 0.3757 - acc: 0.8438
Epoch 11/80
653/653 [==============================] - 0s 47us/sample - loss: 0.3658 - acc: 0.8576
Epoch 12/80
653/653 [==============================] - 0s 34us/sample - loss: 0.3795 - acc: 0.8499
Epoch 13/80
653/653 [==============================] - 0s 36us/sample - loss: 0.3788 - acc: 0.8484
Epoch 14/80
653/653 [==============================] - 0s 36us/sample - loss: 0.6189 - acc: 0.8116
Epoch 15/80
653/653 [==============================] - 0s 38us/sample - loss: 0.3756 - acc: 0.8392
Epoch 16/80
653/653 [==============================] - 0s 40us/sample - loss: 0.3777 - acc: 0.8576
Epoch 17/80
653/653 [==============================] - 0s 49us/sample - loss: 0.3665 - acc: 0.8530
Epoch 18/80
653/653 [==============================] - 0s 28us/sample - loss: 0.3702 - acc: 0.8484
Epoch 19/80
653/653 [==============================] - 0s 38us/sample - loss: 0.3616 - acc: 0.8484
Epoch 20/80
653/653 [==============================] - 0s 35us/sample - loss: 0.3621 - acc: 0.8560
Epoch 21/80
653/653 [==============================] - 0s 47us/sample - loss: 0.3624 - acc: 0.8545
Epoch 22/80
653/653 [==============================] - 0s 40us/sample - loss: 0.3686 - acc: 0.8484
Epoch 23/80
653/653 [==============================] - 0s 41us/sample - loss: 0.3514 - acc: 0.8698
Epoch 24/80
653/653 [==============================] - 0s 39us/sample - loss: 0.3626 - acc: 0.8622
Epoch 25/80
653/653 [==============================] - 0s 53us/sample - loss: 0.3973 - acc: 0.8530
Epoch 26/80
653/653 [==============================] - 0s 31us/sample - loss: 0.3505 - acc: 0.8652
Epoch 27/80
653/653 [==============================] - 0s 42us/sample - loss: 0.3480 - acc: 0.8668
Epoch 28/80
653/653 [==============================] - 0s 31us/sample - loss: 0.3453 - acc: 0.8683
Epoch 29/80
653/653 [==============================] - 0s 46us/sample - loss: 0.3333 - acc: 0.8698
Epoch 30/80
653/653 [==============================] - 0s 50us/sample - loss: 0.3608 - acc: 0.8576
Epoch 31/80
653/653 [==============================] - 0s 49us/sample - loss: 0.3616 - acc: 0.8637
Epoch 32/80
653/653 [==============================] - 0s 35us/sample - loss: 0.3609 - acc: 0.8515
Epoch 33/80
653/653 [==============================] - 0s 37us/sample - loss: 0.4148 - acc: 0.8484
Epoch 34/80
653/653 [==============================] - 0s 44us/sample - loss: 1.3700 - acc: 0.7825
Epoch 35/80
653/653 [==============================] - 0s 48us/sample - loss: 0.6737 - acc: 0.7933
Epoch 36/80
653/653 [==============================] - 0s 43us/sample - loss: 0.4980 - acc: 0.8025
Epoch 37/80
653/653 [==============================] - 0s 48us/sample - loss: 0.4042 - acc: 0.8423
Epoch 38/80
653/653 [==============================] - 0s 43us/sample - loss: 0.3452 - acc: 0.8698
Epoch 39/80
653/653 [==============================] - 0s 36us/sample - loss: 0.3442 - acc: 0.8683
Epoch 40/80
653/653 [==============================] - 0s 44us/sample - loss: 0.3456 - acc: 0.8683
Epoch 41/80
653/653 [==============================] - 0s 43us/sample - loss: 0.3456 - acc: 0.8683
Epoch 42/80
653/653 [==============================] - 0s 44us/sample - loss: 0.3377 - acc: 0.8714
Epoch 43/80
653/653 [==============================] - 0s 44us/sample - loss: 0.3409 - acc: 0.8591
Epoch 44/80
653/653 [==============================] - 0s 44us/sample - loss: 0.3465 - acc: 0.8606
Epoch 45/80
653/653 [==============================] - 0s 52us/sample - loss: 0.3270 - acc: 0.8760
Epoch 46/80
653/653 [==============================] - 0s 45us/sample - loss: 0.3337 - acc: 0.8760
Epoch 47/80
653/653 [==============================] - 0s 41us/sample - loss: 0.3326 - acc: 0.8683
Epoch 48/80
653/653 [==============================] - 0s 41us/sample - loss: 0.3269 - acc: 0.8698
Epoch 49/80
653/653 [==============================] - 0s 49us/sample - loss: 0.3284 - acc: 0.8729
Epoch 50/80
653/653 [==============================] - 0s 28us/sample - loss: 0.3311 - acc: 0.8790
Epoch 51/80
653/653 [==============================] - 0s 43us/sample - loss: 0.3242 - acc: 0.8821
Epoch 52/80
653/653 [==============================] - 0s 39us/sample - loss: 0.3220 - acc: 0.8714
Epoch 53/80
653/653 [==============================] - 0s 40us/sample - loss: 0.3742 - acc: 0.8606
Epoch 54/80
653/653 [==============================] - 0s 55us/sample - loss: 0.3976 - acc: 0.8438
Epoch 55/80
653/653 [==============================] - 0s 31us/sample - loss: 0.3378 - acc: 0.8806
Epoch 56/80
653/653 [==============================] - 0s 38us/sample - loss: 0.4113 - acc: 0.8530
Epoch 57/80
653/653 [==============================] - 0s 42us/sample - loss: 0.5724 - acc: 0.8254
Epoch 58/80
653/653 [==============================] - 0s 25us/sample - loss: 0.4780 - acc: 0.8377
Epoch 59/80
653/653 [==============================] - 0s 33us/sample - loss: 0.4665 - acc: 0.8300
Epoch 60/80
653/653 [==============================] - 0s 44us/sample - loss: 0.3649 - acc: 0.8668
Epoch 61/80
653/653 [==============================] - 0s 44us/sample - loss: 0.3242 - acc: 0.8744
Epoch 62/80
653/653 [==============================] - 0s 45us/sample - loss: 0.3166 - acc: 0.8821
Epoch 63/80
653/653 [==============================] - 0s 44us/sample - loss: 0.3236 - acc: 0.8729
Epoch 64/80
653/653 [==============================] - 0s 29us/sample - loss: 0.3264 - acc: 0.8760
Epoch 65/80
653/653 [==============================] - 0s 45us/sample - loss: 0.3227 - acc: 0.8806
Epoch 66/80
653/653 [==============================] - 0s 40us/sample - loss: 0.3116 - acc: 0.8821
Epoch 67/80
653/653 [==============================] - 0s 50us/sample - loss: 0.3161 - acc: 0.8836
Epoch 68/80
653/653 [==============================] - 0s 41us/sample - loss: 0.3159 - acc: 0.8775
Epoch 69/80
653/653 [==============================] - 0s 38us/sample - loss: 0.3270 - acc: 0.8683
Epoch 70/80
653/653 [==============================] - 0s 32us/sample - loss: 0.3424 - acc: 0.8668
Epoch 71/80
653/653 [==============================] - 0s 51us/sample - loss: 0.3242 - acc: 0.8698
Epoch 72/80
653/653 [==============================] - 0s 45us/sample - loss: 0.3218 - acc: 0.8698
Epoch 73/80
653/653 [==============================] - 0s 45us/sample - loss: 0.3145 - acc: 0.8790
Epoch 74/80
653/653 [==============================] - 0s 31us/sample - loss: 0.3110 - acc: 0.8698
Epoch 75/80
653/653 [==============================] - 0s 51us/sample - loss: 0.3248 - acc: 0.8729
Epoch 76/80
653/653 [==============================] - 0s 43us/sample - loss: 0.3226 - acc: 0.8851
Epoch 77/80
653/653 [==============================] - 0s 33us/sample - loss: 0.3285 - acc: 0.8698
Epoch 78/80
653/653 [==============================] - 0s 35us/sample - loss: 0.3183 - acc: 0.8683
Epoch 79/80
653/653 [==============================] - 0s 25us/sample - loss: 0.3041 - acc: 0.8775
Epoch 80/80
653/653 [==============================] - 0s 36us/sample - loss: 0.3186 - acc: 0.8760

#查看history中都有些啥
history.history.keys()

dict_keys(['loss', 'acc'])

#绘制训练次数与loss的图像
plt.plot(history.epoch, history.history.get('loss'))

#绘制训练次数与准确率的图像
plt.plot(history.epoch, history.history.get('acc'))