1-2梯度下降法&多层感知器（Tensorflow学习笔记）

#寻找损失函数的极值点的算法
#局部最优
#为什么一定会找到最小值？
#随机初始化值的概率
#学习速率是一种超参数
#有哪些优化算法，以及他们的学习速率怎么定
#学习速率过高过低，会影响找到极值点的效率，如果太大，会让损失函数的极值在极值点附近来回跳动

感知器的历程

#多层感知器（神经网络）
#单个神经元
#多个神经元（多分类）

• 无法解决异或问题
• 单层神经元的缺陷：神经元要求数据必须是线性可分得，异或无法找到一条直线分割两个类
• 0 1 ：1；1 0 ：0；0 0： 0；1 1 ：1

多层感知器

- 输入层，隐含层，输出层，输出
- 激活函数：
- 	relu __/
- 	sigmoid,映射到（-1,1）
- 	tanh (-1,1)
-	leak relu 负小正大  

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

import os
os.getcwd()

path = 'D:\\xxx\\pythonPycharm_2018.3.5 Build 183.5912.18\\xxx
'
#修改当前工作目录
os.chdir(path)

#查看修改后的工作目录
print("目录修改成功 %s" %os.getcwd())

目录修改成功 D:\xxx\pythonPycharm_2018.3.5 Build 183.5912.18\xxx

data = pd.read_csv('./advertising.csv')

data.head()

	TV	radio	newspaper	sales
0	230.1	37.8	69.2	22.1
1	44.5	39.3	45.1	10.4
2	17.2	45.9	69.3	9.3
3	151.5	41.3	58.5	18.5
4	180.8	10.8	58.4	12.9

plt.scatter(data.TV,data.sales)

plt.scatter(data.radio,data.sales)

plt.scatter(data.newspaper,data.sales)

x = data.iloc[:,0:-1]
y = data.iloc[:,-1]
import tensorflow as tf 

print(y)

0      22.1
1      10.4
2       9.3
3      18.5
4      12.9
5       7.2
6      11.8
7      13.2
8       4.8
9      10.6
10      8.6
11     17.4
12      9.2
13      9.7
14     19.0
15     22.4
16     12.5
17     24.4
18     11.3
19     14.6
20     18.0
21     12.5
22      5.6
23     15.5
24      9.7
25     12.0
26     15.0
27     15.9
28     18.9
29     10.5
       ... 
170     8.4
171    14.5
172     7.6
173    11.7
174    11.5
175    27.0
176    20.2
177    11.7
178    11.8
179    12.6
180    10.5
181    12.2
182     8.7
183    26.2
184    17.6
185    22.6
186    10.3
187    17.3
188    15.9
189     6.7
190    10.8
191     9.9
192     5.9
193    19.6
194    17.3
195     7.6
196     9.7
197    12.8
198    25.5
199    13.4
Name: sales, Length: 200, dtype: float64

model = tf.keras.Sequential([tf.keras.layers.Dense(10,input_shape = (3,),activation = 'relu')
                            ,tf.keras.layers.Dense(1)]
                            )
#隐含层10
#3*10 
#输入层 10*3 + 10*1 bias  40个参数，隐含层10个值
#输出层 1个值，11个参数

model.summary()

Model: "sequential"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
dense (Dense)                (None, 10)                40        
_________________________________________________________________
dense_1 (Dense)              (None, 1)                 11        
=================================================================
Total params: 51
Trainable params: 51
Non-trainable params: 0
_________________________________________________________________

model.compile(optimizer = 'adam',
             loss = 'mse'
             )

model.fit(x,y,epochs = 100)

WARNING:tensorflow:Falling back from v2 loop because of error: Failed to find data adapter that can handle input: <class 'pandas.core.frame.DataFrame'>, <class 'NoneType'>
Train on 200 samples
Epoch 1/100
200/200 [==============================] - 2s 8ms/sample - loss: 552.0370
Epoch 2/100
200/200 [==============================] - 0s 274us/sample - loss: 416.8464
Epoch 3/100
200/200 [==============================] - 0s 303us/sample - loss: 317.0524
Epoch 4/100
200/200 [==============================] - 0s 214us/sample - loss: 238.6739
Epoch 5/100
200/200 [==============================] - 0s 319us/sample - loss: 192.6979
Epoch 6/100
200/200 [==============================] - 0s 304us/sample - loss: 156.6224
Epoch 7/100
200/200 [==============================] - 0s 279us/sample - loss: 137.7957
Epoch 8/100
200/200 [==============================] - 0s 249us/sample - loss: 123.7041
Epoch 9/100
200/200 [==============================] - 0s 230us/sample - loss: 116.2142
Epoch 10/100
200/200 [==============================] - 0s 304us/sample - loss: 109.5697
Epoch 11/100
200/200 [==============================] - 0s 299us/sample - loss: 104.0680
Epoch 12/100
200/200 [==============================] - 0s 329us/sample - loss: 98.7045
Epoch 13/100
200/200 [==============================] - 0s 299us/sample - loss: 93.6163
Epoch 14/100
200/200 [==============================] - 0s 255us/sample - loss: 88.7300
Epoch 15/100
200/200 [==============================] - 0s 274us/sample - loss: 84.1461
Epoch 16/100
200/200 [==============================] - 0s 449us/sample - loss: 79.4312
Epoch 17/100
200/200 [==============================] - 0s 261us/sample - loss: 75.0235
Epoch 18/100
200/200 [==============================] - 0s 274us/sample - loss: 71.0549
Epoch 19/100
200/200 [==============================] - 0s 239us/sample - loss: 66.6472
Epoch 20/100
200/200 [==============================] - 0s 304us/sample - loss: 62.6407
Epoch 21/100
200/200 [==============================] - 0s 304us/sample - loss: 58.9893
Epoch 22/100
200/200 [==============================] - 0s 284us/sample - loss: 55.3490
Epoch 23/100
200/200 [==============================] - 0s 280us/sample - loss: 51.8215
Epoch 24/100
200/200 [==============================] - ETA: 0s - loss: 30.16 - 0s 319us/sample - loss: 48.5518
Epoch 25/100
200/200 [==============================] - 0s 434us/sample - loss: 45.5142
Epoch 26/100
200/200 [==============================] - 0s 300us/sample - loss: 42.7557
Epoch 27/100
200/200 [==============================] - 0s 249us/sample - loss: 40.1837
Epoch 28/100
200/200 [==============================] - 0s 262us/sample - loss: 37.6466
Epoch 29/100
200/200 [==============================] - 0s 299us/sample - loss: 35.3706
Epoch 30/100
200/200 [==============================] - 0s 349us/sample - loss: 33.1281
Epoch 31/100
200/200 [==============================] - 0s 244us/sample - loss: 31.0810
Epoch 32/100
200/200 [==============================] - 0s 244us/sample - loss: 28.9801
Epoch 33/100
200/200 [==============================] - 0s 240us/sample - loss: 27.1998
Epoch 34/100
200/200 [==============================] - 0s 304us/sample - loss: 25.4241
Epoch 35/100
200/200 [==============================] - 0s 249us/sample - loss: 23.3864
Epoch 36/100
200/200 [==============================] - 0s 284us/sample - loss: 21.3694
Epoch 37/100
200/200 [==============================] - 0s 324us/sample - loss: 19.3851
Epoch 38/100
200/200 [==============================] - 0s 254us/sample - loss: 17.2471
Epoch 39/100
200/200 [==============================] - 0s 200us/sample - loss: 15.2541
Epoch 40/100
200/200 [==============================] - 0s 254us/sample - loss: 13.8191
Epoch 41/100
200/200 [==============================] - 0s 273us/sample - loss: 12.3512
Epoch 42/100
200/200 [==============================] - 0s 259us/sample - loss: 11.1168
Epoch 43/100
200/200 [==============================] - 0s 269us/sample - loss: 10.2072
Epoch 44/100
200/200 [==============================] - 0s 269us/sample - loss: 9.4599
Epoch 45/100
200/200 [==============================] - 0s 319us/sample - loss: 8.9538
Epoch 46/100
200/200 [==============================] - 0s 319us/sample - loss: 8.4765
Epoch 47/100
200/200 [==============================] - 0s 238us/sample - loss: 8.0536
Epoch 48/100
200/200 [==============================] - 0s 249us/sample - loss: 7.7974
Epoch 49/100
200/200 [==============================] - 0s 284us/sample - loss: 7.4901
Epoch 50/100
200/200 [==============================] - 0s 394us/sample - loss: 7.2274
Epoch 51/100
200/200 [==============================] - 0s 299us/sample - loss: 7.0059
Epoch 52/100
200/200 [==============================] - 0s 244us/sample - loss: 6.8145
Epoch 53/100
200/200 [==============================] - 0s 269us/sample - loss: 6.5581
Epoch 54/100
200/200 [==============================] - ETA: 0s - loss: 6.477 - 0s 309us/sample - loss: 6.3564
Epoch 55/100
200/200 [==============================] - 0s 295us/sample - loss: 6.1689
Epoch 56/100
200/200 [==============================] - 0s 234us/sample - loss: 6.0046
Epoch 57/100
200/200 [==============================] - 0s 289us/sample - loss: 5.8701
Epoch 58/100
200/200 [==============================] - 0s 264us/sample - loss: 5.7062
Epoch 59/100
200/200 [==============================] - 0s 304us/sample - loss: 5.5620
Epoch 60/100
200/200 [==============================] - 0s 284us/sample - loss: 5.4335
Epoch 61/100
200/200 [==============================] - 0s 234us/sample - loss: 5.3382
Epoch 62/100
200/200 [==============================] - 0s 224us/sample - loss: 5.2391
Epoch 63/100
200/200 [==============================] - 0s 249us/sample - loss: 5.1235
Epoch 64/100
200/200 [==============================] - 0s 194us/sample - loss: 5.0437
Epoch 65/100
200/200 [==============================] - 0s 219us/sample - loss: 4.9524
Epoch 66/100
200/200 [==============================] - 0s 225us/sample - loss: 4.8507
Epoch 67/100
200/200 [==============================] - 0s 219us/sample - loss: 4.7549
Epoch 68/100
200/200 [==============================] - 0s 189us/sample - loss: 4.6943
Epoch 69/100
200/200 [==============================] - 0s 284us/sample - loss: 4.5864
Epoch 70/100
200/200 [==============================] - 0s 265us/sample - loss: 4.5284
Epoch 71/100
200/200 [==============================] - 0s 294us/sample - loss: 4.4632
Epoch 72/100
200/200 [==============================] - 0s 315us/sample - loss: 4.3811
Epoch 73/100
200/200 [==============================] - 0s 314us/sample - loss: 4.3129
Epoch 74/100
200/200 [==============================] - 0s 249us/sample - loss: 4.2554
Epoch 75/100
200/200 [==============================] - 0s 254us/sample - loss: 4.2137
Epoch 76/100
200/200 [==============================] - 0s 224us/sample - loss: 4.1420
Epoch 77/100
200/200 [==============================] - 0s 209us/sample - loss: 4.0912
Epoch 78/100
200/200 [==============================] - 0s 214us/sample - loss: 4.0383
Epoch 79/100
200/200 [==============================] - 0s 259us/sample - loss: 3.9995
Epoch 80/100
200/200 [==============================] - 0s 219us/sample - loss: 3.9455
Epoch 81/100
200/200 [==============================] - 0s 224us/sample - loss: 3.9012
Epoch 82/100
200/200 [==============================] - 0s 194us/sample - loss: 3.8656
Epoch 83/100
200/200 [==============================] - 0s 234us/sample - loss: 3.8233
Epoch 84/100
200/200 [==============================] - 0s 184us/sample - loss: 3.7835
Epoch 85/100
200/200 [==============================] - 0s 180us/sample - loss: 3.7444
Epoch 86/100
200/200 [==============================] - 0s 189us/sample - loss: 3.7232
Epoch 87/100
200/200 [==============================] - 0s 209us/sample - loss: 3.6805
Epoch 88/100
200/200 [==============================] - 0s 224us/sample - loss: 3.6476
Epoch 89/100
200/200 [==============================] - 0s 159us/sample - loss: 3.6209
Epoch 90/100
200/200 [==============================] - 0s 234us/sample - loss: 3.5872
Epoch 91/100
200/200 [==============================] - 0s 254us/sample - loss: 3.5596
Epoch 92/100
200/200 [==============================] - 0s 219us/sample - loss: 3.5362
Epoch 93/100
200/200 [==============================] - 0s 185us/sample - loss: 3.5044
Epoch 94/100
200/200 [==============================] - 0s 434us/sample - loss: 3.5039
Epoch 95/100
200/200 [==============================] - 0s 344us/sample - loss: 3.4594
Epoch 96/100
200/200 [==============================] - 0s 279us/sample - loss: 3.4415
Epoch 97/100
200/200 [==============================] - 0s 299us/sample - loss: 3.4206
Epoch 98/100
200/200 [==============================] - 0s 224us/sample - loss: 3.3926
Epoch 99/100
200/200 [==============================] - 0s 345us/sample - loss: 3.3714
Epoch 100/100
200/200 [==============================] - 0s 319us/sample - loss: 3.3478





<tensorflow.python.keras.callbacks.History at 0x23c87e55080>

test = data.iloc[:10,0:-1]

model.predict(test)

WARNING:tensorflow:Falling back from v2 loop because of error: Failed to find data adapter that can handle input: <class 'pandas.core.frame.DataFrame'>, <class 'NoneType'>





array([[22.526861 ],
       [ 6.9270716],
       [ 8.889206 ],
       [19.793991 ],
       [12.460696 ],
       [ 9.313395 ],
       [11.779992 ],
       [11.498346 ],
       [ 1.3081465],
       [10.4266   ]], dtype=float32)

test = data.iloc[:10,-1]
print(test)

0    22.1
1    10.4
2     9.3
3    18.5
4    12.9
5     7.2
6    11.8
7    13.2
8     4.8
9    10.6
Name: sales, dtype: float64

来源：CSDN

作者：Z pz

链接：https://blog.csdn.net/wenniewennie/article/details/104413635