用python+sklearn(机器学习)实现天气预报数据 数据

项目地址

github项目:PYWeatherReport

系列教程

机器学习参考篇: python+sklearn+kaggle机器学习
用python+sklearn(机器学习)实现天气预报数据 数据
用python+sklearn(机器学习)实现天气预报 准备
用python+sklearn(机器学习)实现天气预报 模型和使用

勘误表

1. 感谢"Gbilibili"的提醒，下面url代码生成片段
   应从

# 爬取数据链接
   url = "http://www.meteomanz.com/sy2?l=1&cou=2250&ind=59287&d1=" + 
   str(week_ago.day).zfill(2) + 
   "&m1=" + str(week_ago.month).zfill(2) + 
   "&y1=" + str(week_ago.month) + 
   "&d2=" + str(week_pre.day - years[0]).zfill(2) + 
   "&m2=" + str(week_pre.month).zfill(2) + 
   "&y2=" + str(week_pre.year - years[1])

改成

 # 爬取数据链接
url = "http://www.meteomanz.com/sy2?l=1&cou=2250&ind=59287&d1=" + 
    str(week_ago.day).zfill(2) +
    "&m1=" + str( week_ago.month).zfill(2) +
   "&y1=" + str(week_ago.year - years[0]) +
	"&d2=" + str(week_pre.day).zfill(2) +
	"&m2=" + str(week_pre.month).zfill(2) +
    "&y2=" + str(week_pre.year - years[1])

2. 感谢@L-Zzxnn的提醒，代码已添加负温度支持，最新代码请以github上为准

0.前言

在上一篇教程里我们已经知道了数据来源网页的规则，所以这一篇就讲数据如何用爬虫获取和机器学习的数据预处理阶段

1.爬虫

爬虫这方面可以参考我之前的一篇文章

a.确认要被爬取的网页网址

首先我们主要要爬取去年今日的半个月前到去年今日，而根据上一篇我们得出的网址规则，我们可以得到(PS:真正的链接里是没有换行的)

http://www.meteomanz.com/sy2?l=1&cou=2250&ind=59287
&d1=去年今日的半个月前的日
&m1=去年今日的半个月前的月份
&y1=去年年份
&d2=今天的日期的日
&m2=今天的日期的月份
&y2=今年年份

而为什么是取去年和时间要半个月呢？因为去年的天气环境相比于前年或者更久之前是和我们现在的天气条件更相似的，可以减少误差，半个月而不是一个星期是因为使用多的数据量可以减少误差，不是一个月而是因为网站的限制，而且在实验中也会增加少量的误差。所以最终取用了去年和半个月的时间。

如果我们是只测今天这一次上面的网址就可以人工填写，但是如果我们要做不用人工填就要用datetime这个python库
如下:

import datetime as DT

# 取现在日期
today = DT.datetime.now()
# 取b[0]天前日期
week_ago = (today - DT.timedelta(days=b[0])).date()
 # b[1]天后
 week_pre = (today + DT.timedelta(days=b[1])).date()

我们传入b = [-15 0]，就可以获取上个半月的日期在week_ago里，今天的日期在week_pre里
所以，可以用这一行构建需要的网址

 # 爬取数据链接
 url = "http://www.meteomanz.com/sy2?l=1&cou=2250&ind=59287&d1=" + 
	 str(week_ago.day).zfill(2) +
	 "&m1=" + str( week_ago.month).zfill(2) +
	"&y1=" + str(week_ago.year - years[0]) +
 	"&d2=" + str(week_pre.day).zfill(2) +
 	"&m2=" + str(week_pre.month).zfill(2) +
	 "&y2=" + str(week_pre.year - years[1])

其中.zfill(2)是指填充2位，比如如果是1就返回01，如果是12就返回12
有了网址，接下来就是爬虫爬取网页然后分析网页元素取出里面的数据

b.爬虫部分

首先先写爬虫部分，这部分很简单，写了个GetData class

# -*- coding: utf-8 -*-
# @Time: 2020/12/16
# @Author: Eritque arcus
# @File: GetData.py
# 功能: 爬取数据
import urllib3


class GetData:
    url = ""
    headers = ""

    def __init__(self, url, header=""):
        """
        :param url: 获取的网址
        :param header: 请求头，默认已内置
        """
        self.url = url
        if header == "":
            self.headers = {
                'Connection': 'Keep-Alive',
                'Accept': 'text/html,application/xhtml+xml,application/xml;q=0.9,image/avif,image/webp,image/apng,'
                          '*/*;q=0.8,application/signed-exchange;v=b3;q=0.9',
                'Accept-Language': 'zh-CN,zh;q=0.9,en;q=0.8',
                'Accept-Encoding': 'gzip, deflate',
                'User-Agent': 'Mozilla/5.0 (Linux; Android 6.0; Nexus 5 Build/MRA58N) AppleWebKit/537.36 (KHTML, '
                              'like Gecko) Chrome/87.0.4280.66 Mobile Safari/537.36 ',
                'Host': 'www.meteomanz.com'
            }
        else:
            self.headers = header

    def Get(self):
        """
        :return: 网址对应的网页内容
        """
        http = urllib3.PoolManager()
        return http.request('GET', self.url, headers=self.headers).data

本处用了urllib3库和GET方式，其中headers是申请头，这部分可以在按F12调出开发者工具，在Network那一栏，点击任意一个事件，往下滑就有了，可以用我的也可以。请求头主要是http协议里的东西，想要了解可以自行搜索。

c.网页内容匹配取出部分

本处使用了BeautifulSoup库

		g = GetData(url).Get()
        # beautifulsoup解析网页
        soup = BeautifulSoup(g, "html5lib")
        # 取内容
        tb = soup.find(name="tbody")
        # 取tr内容
        past_tr = tb.find_all(name="tr")
        for tr in past_tr:
            # 取tr内每个td的内容
            text = tr.find_all(name="td")
            flag = False
            for i in range(0, len(text)):
                if i == 0:
                    text[i] = text[i].a.string
                    # 网站bug，跨月请求的话会给每个月第0天的数据，但是里面是全空的因为日期不存在，比如 00/11/2020(日/月/年),所以要手动drop掉这个数据
                    if "00/" in text[i]:
                        flag = True
                elif i == 8:
                    # 把/8去掉，网页显示的格式问题
                    text[i] = text[i].string.replace("/8", "")
                elif i == 5:
                    # 去掉Hpa单位
                    text[i] = text[i].string.replace(" Hpa", "")
                elif i == 6:
                    # 用正则去掉风力里括号内的内容
                    text[i] = re.sub(u"[º(.*?|N|W|E|S)]", "", text[i].string)
                else:
                    # 取每个元素的内容
                    text[i] = text[i].string
                # 丢失数据都取2(简陋做法)
                # 这么做 MAE=3.6021
                text[i] = text[i].replace("-", "2")
                text[i] = text[i].replace("Tr", "2")

如果有什么不清楚的评论里答复。

d.写入csv文件格式化

	import csv
    # 创建文件对象
    f = open(c, 'w', encoding='utf-8', newline='')

    # 基于文件对象构建 csv写入对象
    csv_writer = csv.writer(f)
    # 写入内容,text数组
 	csv_writer.writerow(text)
    # 关闭文件
    f.close()

e.封装成类

Write.py

# -*- coding: utf-8 -*-
# @Time: 2020/12/16
# @Author: Eritque arcus
# @File: Write.py
import re
from bs4 import BeautifulSoup
from GetData import GetData
import datetime as DT
import csv

def a(t):
    return t.replace(" - ", "0")

# 功能: 写csv
def write(years, b, c):
    """
    :param years: [开始日期距离现在的年份, 结束日期距离现在的年份]
    :param b: [开始日期距离现在日期的天数, 结束日期距离现在日期的天数]
    :param c: csv文件名
    :return: None
    """
    # 1. 创建文件对象
    f = open(c, 'w', encoding='utf-8', newline='')

    # 2. 基于文件对象构建 csv写入对象
    csv_writer = csv.writer(f)

    # 3. 构建列表头
    # , "negAve", "negMax", "negMin"
    csv_writer.writerow(["Time", "Ave_t", "Max_t", "Min_t", "Prec", "SLpress", "Winddir", "Windsp", "Cloud"])
    # 取现在日期
    today = DT.datetime.now()
    # 取20天前日期
    week_ago = (today - DT.timedelta(days=b[0])).date()
    # 20天后
    week_pre = (today + DT.timedelta(days=b[1])).date()
    # 城市id 广州59287 青岛 54857
    id = "59287"
    # 爬取数据链接
    url = "http://www.meteomanz.com/sy2?l=1&cou=2250&ind=" + id + "&d1=" + str(week_ago.day).zfill(2) + "&m1=" + str(
        week_ago.month).zfill(2) + "&y1=" + str(week_ago.year - years[0]) + "&d2=" + str(week_pre.day).zfill(
        2) + "&m2=" + str(week_pre.month).zfill(2) + "&y2=" + str(week_pre.year - years[1])
    # 显示获取数据集的网址
    print(url)
    g = GetData(url).Get()
    # beautifulsoup解析网页
    soup = BeautifulSoup(g, "html5lib")
    # 取内容
    tb = soup.find(name="tbody")
    # 取tr内容
    past_tr = tb.find_all(name="tr")
    for tr in past_tr:
        # 取tr内每个td的内容
        text = tr.find_all(name="td")
        flag = False
        negA = negMax = negMin = False
        for i in range(0, len(text)):
            if i == 0:
                text[i] = text[i].a.string
                # 网站bug，会给每个月第0天，比如 00/11/2020,所以要drop掉
                if "00/" in text[i]:
                    flag = True
            elif i == 8:
                # 把/8去掉，网页显示的格式
                text[i] = text[i].string.replace("/8", "")
            elif i == 5:
                # 去掉单位
                text[i] = text[i].string.replace(" Hpa", "")
            elif i == 6:
                # 去掉风力里括号内的内容
                text[i] = re.sub(u"[º(.*?|N|W|E|S)]", "", text[i].string)
            else:
                # 取每个元素的内容
                text[i] = text[i].string
            # 丢失数据都取2(简陋做法)
            # 这么做 MAE=3.6021
            text[i] = "2" if text[i] == "-" else text[i]
            text[i] = "2" if text[i] == "Tr" else text[i]
        text = text[0:9]
        # ext += [str(int(negA)), str(int(negMax)), str(int(negMin))]
        # 4. 写入csv文件内容
        if not flag:
            csv_writer.writerow(text)
    # 5. 关闭文件
    f.close()

GetData.py

# -*- coding: utf-8 -*-
# @Time: 2020/12/16
# @Author: Eritque arcus
# @File: GetData.py
# 功能: 爬取数据
import urllib3


class GetData:
    url = ""
    headers = ""

    def __init__(self, url, header=""):
        """
        :param url: 获取的网址
        :param header: 请求头，默认已内置
        """
        self.url = url
        if header == "":
            self.headers = {
                'Connection': 'Keep-Alive',
                'Accept': 'text/html,application/xhtml+xml,application/xml;q=0.9,image/avif,image/webp,image/apng,'
                          '*/*;q=0.8,application/signed-exchange;v=b3;q=0.9',
                'Accept-Language': 'zh-CN,zh;q=0.9,en;q=0.8',
                'Accept-Encoding': 'gzip, deflate',
                'User-Agent': 'Mozilla/5.0 (Linux; Android 6.0; Nexus 5 Build/MRA58N) AppleWebKit/537.36 (KHTML, '
                              'like Gecko) Chrome/87.0.4280.66 Mobile Safari/537.36 ',
                'Host': 'www.meteomanz.com'
            }
        else:
            self.headers = header

    def Get(self):
        """
        :return: 网址对应的网页内容
        """
        http = urllib3.PoolManager()
        return http.request('GET', self.url, headers=self.headers).data

到时候就可以直接用一行命令取得天气数据了，如下面是取去年今日的20天到去年今日的天气数据

	# 用近几年的数据做训练集
    # 如 [1,1], [20, 0]就是用2019年的今天的20天前到2019年的今天数据做训练集
    # 写入csv
    Write([1, 1], [20, 0], "weather_train_train.csv")

结果如下
weather_train_train.csv

Time,Ave_t,Max_t,Min_t,Prec,SLpress,Winddir,Windsp,Cloud

07/12/2019,14.8,20.8,8.8,0.0,1026.3,331,11,0
06/12/2019,15.2,19.8,10.7,0.0,1026.6,344,15,0
05/12/2019,14.5,20.4,8.6,2,1026.2,346,13,8
04/12/2019,13.8,20.4,7.1,0.0,1024.7,335,16,2
03/12/2019,13.0,18.9,7.1,0.0,1024.8,330,10,0
02/12/2019,18.2,24.9,11.5,0.0,1024.8,347,18,3
01/12/2019,18.1,24.9,11.4,0.0,1020.9,332,16,1
30/11/2019,17.5,23.6,11.4,0.0,1020.5,352,8,3
29/11/2019,15.8,20.1,11.5,0.0,1023.6,349,11,4
28/11/2019,20.4,27.1,13.8,0.0,1024.5,337,19,3
27/11/2019,21.9,27.1,16.6,0.0,1021.3,336,12,0
26/11/2019,22.2,28.4,16.1,0.0,1021.1,356,6,6
25/11/2019,22.2,29.3,15.2,0.0,1020.8,344,13,3
24/11/2019,21.4,29.3,13.6,0.0,1018.5,346,5,0
23/11/2019,20.7,28.4,13.0,0.0,1017.2,352,5,1
22/11/2019,19.6,27.6,11.6,0.0,1017.3,331,6,0
21/11/2019,18.4,25.1,11.6,0.0,1019.1,323,9,1
20/11/2019,18.3,24.2,12.4,0.0,1020.3,338,7,0
19/11/2019,19.1,25.4,12.8,0.0,1020.5,342,11,0
18/11/2019,22.2,28.8,15.7,0.0,1018.8,342,17,0
17/11/2019,22.2,28.8,15.7,0.0,1015.2,358,7,3

2.数据预处理

如果在把上面的数据作为数据集训练，我们还需要做些数据的预处理，因为有些情况下我们得到的数据会有残缺，这种情况我们就要选择抛弃那一列或者用方差或其他什么的方法填充缺少的数据。

因为在我已经决定把数据里丢失的项全部取2了，所以下面我会列出可能的解决方法而不使用。

新建个ProcessData.py里建立ProcessData方法以获得数据

# -*- coding: utf-8 -*-
# @Time: 2020/12/16
# @Author: Eritque arcus
# @File: ProcessData.py
from Write import Write
import pandas as pd
from sklearn.model_selection import train_test_split
from sklearn.impute import SimpleImputer
import seaborn as sns
import matplotlib.pyplot as plt


# 功能: 数据预处理
def ProcessData():
    """
    :return:
        [X_train X训练数据集,
        X_valid X训练数据集的验证集,
        y_train Y训练数据集,
        y_valid Y训练数据集的验证集,
        imputed_X_test 预测数据集]
    """
    # 用近几年的数据做训练集
    # 如 [1,1], [20, 0]就是用2019年的今天的20天前到2019年的今天数据做训练集
    # 写入csv
    Write([1, 1], [20, 0], "weather_train_train.csv")
    Write([1, 1], [0, 20], "weather_train_valid.csv")
    Write([0, 0], [20, 0], "weather_test.csv")
    X_test = pd.read_csv("weather_test.csv", index_col="Time", parse_dates=True)
    # 读取测试集和验证集
    X = pd.read_csv("weather_train_train.csv", index_col="Time", parse_dates=True)
    y = pd.read_csv("weather_train_valid.csv", index_col="Time", parse_dates=True)
    # 把全部丢失的数据都drop，MAE=3.7又高了，所以去掉了
    # dxtcol = [col for col in X_test.columns
    #           if X_test[col].isnull().all()]
    # dxcol = [col for col in X.columns
    #          if X[col].isnull().all()]
    # dycol = [col for col in y.columns
    #          if y[col].isnull().all()]
    # for a1 in [dxtcol, dxcol, dycol]:
    #     for a2 in a1:
    #         if a2 in X_test.columns:
    #             X_test = X_test.drop(a2, axis=1)
    #         if a2 in X.columns:
    #             X = X.drop(a2, axis=1)
    #         if a2 in y.columns:
    #             y = y.drop(a2, axis=1)
    # 数据归一化和标准化，无法还原不用
    # scaler = preprocessing.StandardScaler()
    # pars = [cols for cols in X.columns if cols != "Time"]
    # for data in [X, y, X_test]:
    #     for par in pars:
    #         data[par] = scaler.fit_transform(data[par].values.reshape(-1, 1))
    #         # temp = scaler.fit(data[par].values.reshape(-1, 1))
    #         # data[par] = scaler.fit_transform(data[par].values.reshape(-1, 1), temp)

    # 填充缺少的数值用方差，不清楚效果如何
    my_imputer = SimpleImputer()
    X_train, X_valid, y_train, y_valid = train_test_split(X, y, train_size=0.8, test_size=0.2, random_state=0)
    imputed_X_train = pd.DataFrame(my_imputer.fit_transform(X_train))
    imputed_X_valid = pd.DataFrame(my_imputer.transform(X_valid))
    imputed_X_train.columns = X_train.columns
    imputed_X_valid.columns = X_valid.columns
    imputed_y_train = pd.DataFrame(my_imputer.fit_transform(y_train))
    imputed_y_valid = pd.DataFrame(my_imputer.transform(y_valid))
    imputed_y_train.columns = y_train.columns
    imputed_y_valid.columns = y_valid.columns
    imputed_X_test = pd.DataFrame(my_imputer.fit_transform(X_test))

    # 画折线图
    # sns.lineplot(data=X)
    # plt.show()
    # sns.lineplot(data=y)
    # plt.show()
    # sns.lineplot(data=X_test)
    # plt.show()
    # 返回分割后的数据集
    return [imputed_X_train, imputed_X_valid, imputed_y_train, imputed_y_valid, imputed_X_test]

下一篇 用python+sklearn(机器学习)实现天气预报 模型和使用