文的文字及图片来源于网络,仅供学习、交流使用,不具有任何商业用途,版权归原作者所有,如有问题请及时联系我们以作处理。
爬虫的网站:万邦国际集团。其成立于2010年,总部位于河南省郑州市,以“立足三农、保障民生、服务全国”为宗旨,业务涵盖综合性农产品冷链物流、高效生态农业开发、生鲜连锁超市、跨境电子商务、进出口贸易等农业全产业链。荣获重点龙头企业、全国农产品“综合十强市场”、“星创天地”、全国“万企帮万村”精准扶贫先进民营企业等荣誉称号。目前,集团在中牟县建设运营的万邦农产品物流园区,已累计完成投资100亿元,占地5000亩,建筑面积达350万平方米。拥有固定商户6000多家,2017年各类农副产品交易额913亿元,交易量1720万吨,位居全国前列,实现农产品“买全球、卖全国”。
其价格信息查询为get请求,网页比较规范,且短期内不会有大的变动,很容易分析,故选择之。
# _*_ coding:utf-8 _*_
# 开发人员:未央
# 开发时间:2020/4/12 16:03
# 文件名:Scrapy_lab1.py
# 开发工具:PyCharm
import csv
import codecs
import requests # 导入requests包
from bs4 import BeautifulSoup # 导入bs4包
from datetime import datetime
class Produce:
price_data = [] # 农产品的价格数据列表
item_name = "" # 农产品的类别名
def __init__(self, category):
self.item_name = category
self.price_data = []
# 读取某一页的数据,默认是第一页
def get_price_page_data(self, page_index=1):
url = 'http://www.wbncp.com/PriceQuery.aspx?PageNo=' + str(
page_index) + '&ItemName=' + self.item_name + '&DateStart=2017/10/1&DateEnd=2020/3/31 '
strhtml = requests.get(url) # GET方式,获取网页数据
# print(strhtml.text)
soup = BeautifulSoup(strhtml.text, 'html.parser') # 解析网页文档
# print(soup)
table_node = soup.find_all('table')
# number = 0
# for table in table_node:
# number += 1
# print(number, table)
all_price_table = table_node[21] # 获取含有农产品价钱的table的数据
# print(all_price_table)
for tr in all_price_table.find_all('tr'):
number = 0
price_line = []
for td in tr.find_all('td'):
number += 1
# print(number, td)
if number == 1:
price_line.append(td.get_text().split()) # 获取品名
elif number == 2:
price_line.append(td.get_text().split()) # 获取产地
elif number == 3:
price_line.append(td.get_text().split()) # 获取规格
elif number == 4:
price_line.append(td.get_text().split()) # 获取单位
elif number == 5:
price_line.append(td.get_text().split()) # 获取最高价
elif number == 6:
price_line.append(td.get_text().split()) # 获取最低价
elif number == 7:
price_line.append(td.get_text().split()) # 获取均价
elif number == 8:
price_line.append(datetime.strptime(td.get_text().replace('/', '-'), '%Y-%m-%d')) # 获取日期
self.price_data.append(price_line)
return
# 获取全部页面的数据
def get_price_data(self):
for i in range(33):
self.get_price_page_data(str(i))
return
# 讲爬虫的数据写入到CSV文件,路径为:D:\Data_pytorch\名字.csv
def data_write_csv(self): # file_address为写入CSV文件的路径,self.price_data为要写入数据列表
self.get_price_data()
file_address = "D:\Data_pytorch\\" + self.item_name.__str__() + ".csv"
file_csv = codecs.open(file_address, 'w+', 'utf-8') # 追加
writer = csv.writer(file_csv, delimiter=' ', quotechar=' ', quoting=csv.QUOTE_MINIMAL)
for temp_data in self.price_data:
writer.writerow(temp_data)
print(self.item_name + "爬虫数据保存到文件成功!")
# 以字典类型读取csv文件,读取路径为:D:\Data_pytorch\名字.csv
def data_reader_csv(self):
file_address = "D:\Data_pytorch\\" + self.item_name.__str__() + ".csv"
with open(file_address, 'r', encoding='utf8')as fp:
# 使用列表推导式,将读取到的数据装进列表
data_list = [i for i in csv.DictReader(fp, fieldnames=None)] # csv.DictReader 读取到的数据是list类型
print(self.item_name + "数据如下:")
print(data_list)
return data_list
list = ["白菜", "包菜", "土豆", "菠菜", "蒜苔"]
for temp_name in list:
produce = Produce(temp_name)
produce.data_write_csv()
data = produce.data_reader_csv()
运行之后,文件显示内容如下:
二、使用Scrapy爬取数据
类似之前的学习案例,这里不再一步一步的介绍,直接上代码:
items.py代码如下:
# -*- coding: utf-8 -*-
# Define here the models for your scraped items
#
# See documentation in:
# https://doc.scrapy.org/en/latest/topics/items.html
import scrapy
from scrapy.loader import ItemLoader
from scrapy.loader.processors import TakeFirst
class PriceSpiderItemLoader(ItemLoader):
# 自定义itemloader,用于存储爬虫所抓取的字段内容的
default_output_processor = TakeFirst()
class PriceSpiderItem(scrapy.Item):
# define the fields for your item here like:
# name = scrapy.Field()
name = scrapy.Field() # 品名
address = scrapy.Field() # 产地
norms = scrapy.Field() # 规格
unit = scrapy.Field() # 单位
high = scrapy.Field() # 最高价
low = scrapy.Field() # 最低价
price_ave = scrapy.Field() # 均价
price_date = scrapy.Field() # 日期
setting.py代码如下:
# -*- coding: utf-8 -*-
# Scrapy settings for price_spider project
#
# For simplicity, this file contains only settings considered important or
# commonly used. You can find more settings consulting the documentation:
#
# https://doc.scrapy.org/en/latest/topics/settings.html
# https://doc.scrapy.org/en/latest/topics/downloader-middleware.html
# https://doc.scrapy.org/en/latest/topics/spider-middleware.html
from scrapy.exporters import JsonLinesItemExporter
# 默认显示的中文是阅读性较差的Unicode字符
# 需要定义子类显示出原来的字符集(将父类的ensure_ascii属性设置为False即可)
class CustomJsonLinesItemExporter(JsonLinesItemExporter):
def __init__(self, file, **kwargs):
super(CustomJsonLinesItemExporter, self).__init__(file, ensure_ascii=False, **kwargs)
# 启用新定义的Exporter类
FEED_EXPORTERS = {
'json': 'price_spider.settings.CustomJsonLinesItemExporter',
}
BOT_NAME = 'price_spider'
SPIDER_MODULES = ['price_spider.spiders']
NEWSPIDER_MODULE = 'price_spider.spiders'
# Crawl responsibly by identifying yourself (and your website) on the user-agent
# USER_AGENT = 'price_spider (+http://www.yourdomain.com)'
# Obey robots.txt rules
ROBOTSTXT_OBEY = False
# Configure maximum concurrent requests performed by Scrapy (default: 16)
# CONCURRENT_REQUESTS = 32
# Configure a delay for requests for the same website (default: 0)
# See https://doc.scrapy.org/en/latest/topics/settings.html#download-delay
# See also autothrottle settings and docs
DOWNLOAD_DELAY = 3
爬虫逻辑(spider.py)代码如下:
# _*_ coding:utf-8 _*_
# 开发人员:未央
# 开发时间:2020/4/16 14:55
# 文件名:spider.py
# 开发工具:PyCharm
import scrapy
from price_spider.items import PriceSpiderItemLoader, PriceSpiderItem
class SpiderSpider(scrapy.Spider):
name = 'spider'
allowed_domains = ['www.wbncp.com']
start_urls = ['http://www.wbncp.com/PriceQuery.aspx?PageNo=1&ItemName=%e7%99%bd%e8%8f%9c&DateStart=2017/10/1'
'&DateEnd=2020/3/31', 'http://www.wbncp.com/PriceQuery.aspx?PageNo=1&ItemName=土豆&DateStart=2017/10/1'
'&DateEnd=2020/3/31', 'http://www.wbncp.com/PriceQuery.aspx?PageNo=1&ItemName'
'=芹菜&DateStart=2017/10/1 &DateEnd=2020/3/31']
def parse(self, response):
item_nodes = response.xpath("//tr[@class='Center' or @class='Center Gray']")
for item_node in item_nodes:
item_loader = PriceSpiderItemLoader(item=PriceSpiderItem(), selector=item_node)
item_loader.add_css("name", "td:nth-child(1) ::text")
item_loader.add_css("address", "td:nth-child(2) ::text")
item_loader.add_css("norms", "td:nth-child(3) ::text")
item_loader.add_css("unit", "td:nth-child(4) ::text")
item_loader.add_css("high", "td:nth-child(5) ::text")
item_loader.add_css("low", "td:nth-child(6) ::text")
item_loader.add_css("price_ave", "td:nth-child(7)::text")
item_loader.add_css("price_date", "td:nth-child(8)::text")
price_item = item_loader.load_item()
yield price_item
next_page = response.xpath("//*[@id='cphRight_lblPage']/div/a[10]/@href").extract_first()
if next_page is not None:
next_page = response.urljoin(next_page)
yield scrapy.Request(next_page, callback=self.parse)
替代运行命令(price_scrapy_main.py)的代码如下:
# _*_ coding:utf-8 _*_
# 开发人员:未央
# 开发时间:2020/4/16 14:55
# 文件名:price_scrapy_main.py
# 开发工具:PyCharm
from scrapy.cmdline import execute
execute(["scrapy", "crawl", "spider", "-o", "price_data.csv"])
运作后,将csv数据导入excel中,结果如下:
三、经验总结:
1.使用request确实比较灵活,但是如果爬取数据多很不方便,代码也会很长,还是使用scrapy方便。特别是爬取多个页面,scrapy 的横向和纵向爬取,超级腻害!
2.Scrapy主要是设置文件(setting.py)的各种设置以及爬虫文件(本文是spider.py)的爬虫逻辑,其中主要是选择器部分比较麻烦
如果你处于想学Python或者正在学习Python,Python的教程不少了吧,但是是最新的吗?说不定你学了可能是两年前人家就学过的内容,在这小编分享一波2020最新的Python教程。获取方式,私信小编 “ 资料 ”,即可免费获取哦!