tensorflow2知识总结---4、逻辑回归实例

tensorflow2知识总结---4、逻辑回归实例

一、总结

一句话总结：

也就是将损失函数设置为binary_crossentropy即可：model.compile(optimizer='adam',loss='binary_crossentropy',metrics=['acc'])

1、把tensorflow2的history对象的epoch和损失用图的方式展示出来？

plt.plot(history.epoch,history.history.get('loss'))

# 把epoch当横左边，把loss当纵坐标

二、逻辑回归实例

博客对应课程的视频位置：

In [1]:

import tensorflow as tf
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
%matplotlib inline

In [2]:

# 表示没有表头，如果不设置header=None，则第一行数据会变成表头
data = pd.read_csv('dataset/credit-a.csv',header=None)
data

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

0-14列是特征，第15列是目标值

In [3]:

# 前5行
data.head()

Out[3]:

	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

In [4]:

data.iloc[:,-1].value_counts()

Out[4]:

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

In [5]:

# 逗号前面表示取所有行，逗号后面表示取开头到-1列
x = data.iloc[:,:-1]
# 把-1用0来表示
y = data.iloc[:,-1].replace(-1,0)
# 1和-1适合支持向量机算法

In [6]:

x

Out[6]:

	0	1	2	3	4	5	6	7	8	9	10	11	12	13	14
0	0	30.83	0.000	0	0	9	0	1.25	0	0	1	1	0	202	0.0
1	1	58.67	4.460	0	0	8	1	3.04	0	0	6	1	0	43	560.0
2	1	24.50	0.500	0	0	8	1	1.50	0	1	0	1	0	280	824.0
3	0	27.83	1.540	0	0	9	0	3.75	0	0	5	0	0	100	3.0
4	0	20.17	5.625	0	0	9	0	1.71	0	1	0	1	2	120	0.0
...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
648	0	21.08	10.085	1	1	11	1	1.25	1	1	0	1	0	260	0.0
649	1	22.67	0.750	0	0	0	0	2.00	1	0	2	0	0	200	394.0
650	1	25.25	13.500	1	1	13	7	2.00	1	0	1	0	0	200	1.0
651	0	17.92	0.205	0	0	12	0	0.04	1	1	0	1	0	280	750.0
652	0	35.00	3.375	0	0	0	1	8.29	1	1	0	0	0	0	0.0

653 rows × 15 columns

In [7]:

y

Out[7]:

0      0
1      0
2      0
3      0
4      0
      ..
648    1
649    1
650    1
651    1
652    1
Name: 15, Length: 653, dtype: int64

In [8]:

model = tf.keras.Sequential()
# 中间4个神经元
model.add(tf.keras.layers.Dense(4,input_shape=(15,),activation='relu'))
# 第二层不需要告诉input_shape，他会自己推断
# 第一个隐藏层
model.add(tf.keras.layers.Dense(4,activation='relu'))
# 输出层
model.add(tf.keras.layers.Dense(1,activation='sigmoid'))

model.summary()

Model: "sequential"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
dense (Dense)                (None, 4)                 64        
_________________________________________________________________
dense_1 (Dense)              (None, 4)                 20        
_________________________________________________________________
dense_2 (Dense)              (None, 1)                 5         
=================================================================
Total params: 89
Trainable params: 89
Non-trainable params: 0
_________________________________________________________________

第一层64个节点 等于 15*4+4

第二层 20=4*4+4

In [9]:

# metrics=['acc'] 每次运算完毕，计算正确率
model.compile(optimizer='adam',loss='binary_crossentropy',metrics=['acc'])

history = model.fit(x,y,epochs=100) #epochs表示训练的次数

Epoch 1/100
21/21 [==============================] - 0s 712us/step - loss: 125.1340 - acc: 0.5283
Epoch 2/100
21/21 [==============================] - 0s 665us/step - loss: 73.5050 - acc: 0.5345
Epoch 3/100
21/21 [==============================] - 0s 714us/step - loss: 23.0695 - acc: 0.5865
Epoch 4/100
21/21 [==============================] - 0s 760us/step - loss: 3.7062 - acc: 0.6799
Epoch 5/100
21/21 [==============================] - 0s 807us/step - loss: 3.6635 - acc: 0.6738
Epoch 6/100
21/21 [==============================] - 0s 950us/step - loss: 3.1660 - acc: 0.6662
Epoch 7/100
21/21 [==============================] - 0s 855us/step - loss: 2.6206 - acc: 0.6662
......
Epoch 95/100
21/21 [==============================] - 0s 760us/step - loss: 0.4830 - acc: 0.7948
Epoch 96/100
21/21 [==============================] - 0s 807us/step - loss: 0.6317 - acc: 0.7887
Epoch 97/100
21/21 [==============================] - 0s 712us/step - loss: 0.4358 - acc: 0.8086
Epoch 98/100
21/21 [==============================] - 0s 760us/step - loss: 0.4004 - acc: 0.8300
Epoch 99/100
21/21 [==============================] - 0s 712us/step - loss: 0.3916 - acc: 0.8423
Epoch 100/100
21/21 [==============================] - 0s 665us/step - loss: 0.3911 - acc: 0.8438

In [10]:

history.history.keys()

Out[10]:

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

In [11]:

# 把epoch当横左边，把loss当纵坐标
plt.plot(history.epoch,history.history.get('loss'))

Out[11]:

[<matplotlib.lines.Line2D at 0x210dcf57908>]

In [12]:

plt.plot(history.epoch,history.history.get('acc'))

Out[12]:

[<matplotlib.lines.Line2D at 0x210de790b48>]

In [ ]: