tensorflow学习笔记（一）

import tensorflow as tf import numpy as np  #检查tensorflow版本  tf.__version__  2.8.0  x=[[1.]] m=tf.matmul(x,x)#将矩阵 a 乘以矩阵 b,生成a * b print(m)  tf.Tensor([[1.]], shape=(1, 1), dtype=float32)  x=tf.constant([[1,9],[3,6]]) tf.Tensor( [[1 9]  [3 6]], shape=(2, 2), dtype=int32)  x=tf.add(x,1) tf.Tensor( [[ 2 10]  [ 4  7]], shape=(2, 2), dtype=int32) 

x.numpy()
[[ 2 10]
 [ 4  7]]

x=tf.cast(x,tf.float32)#格式转换为浮点
tf.Tensor(
[[ 2. 10.]
 [ 4.  7.]], shape=(2, 2), dtype=float32)

在tensorflow2中将大量使用keras的简介建模方法

import numpy as np
import pandas as pd 
import matplotlib.pyplot as plt
import tensorflow as tf
from tensorflow.python.keras import layers
import tensorflow.python.keras
import os
os.environ['TF_CPP_MIN_LOG_LEVEL']='2'

使用的是一个温度的数据集来进行实验

features = pd.read_csv('temps0.csv')
print(features.head())

   year  month  day  week  temp_2  temp_1  average  friend
0  2016      1    1   Fri      45      45     45.6      29
1  2016      1    2   Sat      44      45     45.7      61
2  2016      1    3   Sun      45      44     45.8      56
3  2016      1    4   Mon      44      41     45.9      53
4  2016      1    5  Tues      41      40     46.0      41


print("数据维度",features.shape)
数据维度 (348, 8)

处理时间数据

import datetime
years=features['year']
months=features['month']
days=features['day']
#datatime格式
dates=[str(int(year))+'-'+str(int(month))+'-'+str(int(day)) for year ,month ,day in zip(years,months,days)]
dates=[datetime.datetime.strptime(date,'%Y-%m-%d')for date in dates]
print(dates[:5])

[datetime.datetime(2016, 1, 1, 0, 0), datetime.datetime(2016, 1, 2, 0, 0), datetime.datetime(2016, 1, 3, 0, 0), datetime.datetime(2016, 1, 4, 0, 0), datetime.datetime(2016, 1, 5, 0, 0)]

准备画图,再次之间要进行数据处理，删除异常数据，清洗数据等。

一些常用的参数

model=tf.keras.Sequential()
model.add(layers.Dense(16))#全连接层
model.add(layers.Dense(32))
model.add(layers.Dense(1))

compile相当于对网络进行配置，指定好优化器和损失函数等

model.compile(optimizer=tf.keras.optimizers.SGD(0.001),loss='mean_squared_error')
model.fit(input_features,labels,validation_split=0.25,epochs=10,batch_size=64)

Epoch 1/10
5/5 [==============================] - 2s 44ms/step - loss: 4202.6343 - val_loss: 3279.2751
Epoch 2/10
5/5 [==============================] - 0s 10ms/step - loss: 1709.3342 - val_loss: 4615.9727
Epoch 3/10
5/5 [==============================] - 0s 9ms/step - loss: 225.1389 - val_loss: 3343.1311
Epoch 4/10
5/5 [==============================] - 0s 9ms/step - loss: 105.3543 - val_loss: 2275.7815
Epoch 5/10
5/5 [==============================] - 0s 9ms/step - loss: 101.7793 - val_loss: 1845.8273
Epoch 6/10
5/5 [==============================] - 0s 9ms/step - loss: 75.1766 - val_loss: 1353.8687
Epoch 7/10
5/5 [==============================] - 0s 9ms/step - loss: 113.6695 - val_loss: 1052.7235
Epoch 8/10
5/5 [==============================] - 0s 10ms/step - loss: 63.4713 - val_loss: 779.7868
Epoch 9/10
5/5 [==============================] - 0s 10ms/step - loss: 57.4578 - val_loss: 649.0630
Epoch 10/10
5/5 [==============================] - 0s 10ms/step - loss: 39.8692 - val_loss: 638.9788

model.summary()
Model: "sequential"
_________________________________________________________________
 Layer (type)                Output Shape              Param # 
=================================================================
 module_wrapper (ModuleWrapp  (None, 16)               240       
 er)                                                             
                                                                 
 module_wrapper_1 (ModuleWra  (None, 32)               544       
 pper)                                                           
                                                                 
 module_wrapper_2 (ModuleWra  (None, 1)                33        
 pper)                                                           
                                                                 
=================================================================
Total params: 817
Trainable params: 817
Non-trainable params: 0

观察发现有点过拟合,更改初始化方法后

model.add(layers.Dense(16,kernel_initializer='random_normal'))
model.add(layers.Dense(32,kernel_initializer='random_normal'))
model.add(layers.Dense(1,kernel_initializer='random_normal'))

Epoch 1/10
5/5 [==============================] - 1s 39ms/step - loss: 4390.7549 - val_loss: 2869.1125
Epoch 2/10
5/5 [==============================] - 0s 10ms/step - loss: 4295.9175 - val_loss: 2789.7224
Epoch 3/10
5/5 [==============================] - 0s 11ms/step - loss: 4186.6431 - val_loss: 2676.1052
Epoch 4/10
5/5 [==============================] - 0s 10ms/step - loss: 3986.0823 - val_loss: 2471.7280
Epoch 5/10
5/5 [==============================] - 0s 11ms/step - loss: 3223.1704 - val_loss: 2771.4094
Epoch 6/10
5/5 [==============================] - 0s 10ms/step - loss: 387.2784 - val_loss: 1830.3267
Epoch 7/10
5/5 [==============================] - 0s 11ms/step - loss: 65.9947 - val_loss: 1299.6189
Epoch 8/10
5/5 [==============================] - 0s 11ms/step - loss: 67.4968 - val_loss: 1037.1642
Epoch 9/10
5/5 [==============================] - 0s 10ms/step - loss: 44.5196 - val_loss: 702.2187
Epoch 10/10
5/5 [==============================] - 0s 10ms/step - loss: 45.0128 - val_loss: 681.6292

当我们加入惩罚项