亲宝软件园·资讯

展开

Python 爬取二手房数据 Python爬虫入门案例之爬取二手房源数据

松鼠爱吃饼干 人气:0
想了解Python爬虫入门案例之爬取二手房源数据的相关内容吗,松鼠爱吃饼干在本文为您仔细讲解Python 爬取二手房数据的相关知识和一些Code实例,欢迎阅读和指正,我们先划重点:Python,爬取二手房数据,Python,案例,下面大家一起来学习吧。

本文重点

环境介绍

#模块使用

【付费VIP完整版】只要看了就能学会的教程,80集Python基础入门视频教学

点这里即可免费在线观看

爬虫代码实现步骤: 发送请求 >>> 获取数据 >>> 解析数据 >>> 保存数据

导入模块

import requests # 数据请求模块 第三方模块 pip install requests
import parsel # 数据解析模块
import re
import csv

发送请求, 对于房源列表页发送请求

url = 'https://bj.lianjia.com/ershoufang/pg1/'
# 需要携带上 请求头: 把python代码伪装成浏览器 对于服务器发送请求
# User-Agent 浏览器的基本信息
headers = {
    'User-Agent': 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/94.0.4606.61 Safari/537.36'
}
response = requests.get(url=url, headers=headers)

获取数据

print(response.text)

解析数据

selector_1 = parsel.Selector(response.text)
# 把获取到response.text 数据内容转成 selector 对象
href = selector_1.css('div.leftContent li div.title a::attr(href)').getall()
for link in href:
    html_data = requests.get(url=link, headers=headers).text
    selector = parsel.Selector(html_data)
    # css选择器 语法
    # try:
    title = selector.css('.title h1::text').get() # 标题
    area = selector.css('.areaName .info a:nth-child(1)::text').get()  # 区域
    community_name = selector.css('.communityName .info::text').get()  # 小区
    room = selector.css('.room .mainInfo::text').get()  # 户型
    room_type = selector.css('.type .mainInfo::text').get()  # 朝向
    height = selector.css('.room .subInfo::text').get().split('/')[-1]  # 楼层
    # 中楼层/共5层 split('/') 进行字符串分割  ['中楼层', '共5层'] [-1]
    # ['中楼层', '共5层'][-1] 列表索引位置取值 取列表中最后一个元素  共5层
    # re.findall('共(\d+)层', 共5层) >>>  [5][0] >>> 5
    height = re.findall('共(\d+)层', height)[0]
    sub_info = selector.css('.type .subInfo::text').get().split('/')[-1]  # 装修
    Elevator = selector.css('.content li:nth-child(12)::text').get()  # 电梯
    # if Elevator == '暂无数据电梯' or Elevator == None:
    #     Elevator = '无电梯'
    house_area = selector.css('.content li:nth-child(3)::text').get().replace('㎡', '')  # 面积
    price = selector.css('.price .total::text').get()  # 价格(万元)
    date = selector.css('.area .subInfo::text').get().replace('年建', '')  # 年份
    dit = {
        '标题': title,
        '市区': area,
        '小区': community_name,
        '户型': room,
        '朝向': room_type,
        '楼层': height,
        '装修情况': sub_info,
        '电梯': Elevator,
        '面积(㎡)': house_area,
        '价格(万元)': price,
        '年份': date,
    }
    csv_writer.writerow(dit)
    print(title, area, community_name, room, room_type, height, sub_info, Elevator, house_area, price, date,
          sep='|')

保存数据

f = open('二手房数据.csv', mode='a', encoding='utf-8', newline='')
csv_writer = csv.DictWriter(f, fieldnames=[
    '标题',
    '市区',
    '小区',
    '户型',
    '朝向',
    '楼层',
    '装修情况',
    '电梯',
    '面积(㎡)',
    '价格(万元)',
    '年份',
])
csv_writer.writeheader()

数据可视化

导入所需模块

import pandas as pd
from pyecharts.charts import Map
from pyecharts.charts import Bar
from pyecharts.charts import Line
from pyecharts.charts import Grid
from pyecharts.charts import Pie
from pyecharts.charts import Scatter
from pyecharts import options as opts

读取数据

df = pd.read_csv('链家.csv', encoding = 'utf-8')
df.head()

各城区二手房数量北京市地图

new = [x + '区' for x in region]
m = (
        Map()
        .add('', [list(z) for z in zip(new, count)], '北京')
        .set_global_opts(
            title_opts=opts.TitleOpts(title='北京市二手房各区分布'),
            visualmap_opts=opts.VisualMapOpts(max_=3000),
        )
    )
m.render_notebook()

各城区二手房数量-平均价格柱状图

df_price.values.tolist()
price = [round(x,2) for x in df_price.values.tolist()]
bar = (
    Bar()
    .add_xaxis(region)
    .add_yaxis('数量', count,
              label_opts=opts.LabelOpts(is_show=True))
    .extend_axis(
        yaxis=opts.AxisOpts(
            name="价格(万元)",
            type_="value",
            min_=200,
            max_=900,
            interval=100,
            axislabel_opts=opts.LabelOpts(formatter="{value}"),
        )
    )
    .set_global_opts(
        title_opts=opts.TitleOpts(title='各城区二手房数量-平均价格柱状图'),
        tooltip_opts=opts.TooltipOpts(
            is_show=True, trigger="axis", axis_pointer_type="cross"
        ),
        xaxis_opts=opts.AxisOpts(
            type_="category",
            axispointer_opts=opts.AxisPointerOpts(is_show=True, type_="shadow"),
        ),
        yaxis_opts=opts.AxisOpts(name='数量',
            axistick_opts=opts.AxisTickOpts(is_show=True),
            splitline_opts=opts.SplitLineOpts(is_show=False),)
    )
)

line2 = (
    Line()
    .add_xaxis(xaxis_data=region)
    .add_yaxis(
        
        series_name="价格",
        yaxis_index=1,
        y_axis=price,
        label_opts=opts.LabelOpts(is_show=True),
        z=10
        )
)

bar.overlap(line2)
grid = Grid()
grid.add(bar, opts.GridOpts(pos_left="5%", pos_right="20%"), is_control_axis_index=True)
grid.render_notebook()

area0 = top_price['小区'].values.tolist()
count = top_price['价格(万元)'].values.tolist()

bar = (
    Bar()
    .add_xaxis(area0)
    .add_yaxis('数量', count,category_gap = '50%')
    .set_global_opts(
        yaxis_opts=opts.AxisOpts(name='价格(万元)'),
        xaxis_opts=opts.AxisOpts(name='数量'),
    )
)
bar.render_notebook()

散点图

s = (
    Scatter()
    .add_xaxis(df['面积(㎡)'].values.tolist())
    .add_yaxis('',df['价格(万元)'].values.tolist())
    .set_global_opts(xaxis_opts=opts.AxisOpts(type_='value'))
)
s.render_notebook()

房屋朝向占比

directions = df_direction.index.tolist()
count = df_direction.values.tolist()

c1 = (
    Pie(init_opts=opts.InitOpts(
            width='800px', height='600px',
            )
       )
        .add(
        '',
        [list(z) for z in zip(directions, count)],
        radius=['20%', '60%'],
        center=['40%', '50%'],
#         rosetype="radius",
        label_opts=opts.LabelOpts(is_show=True),
        )    
        .set_global_opts(title_opts=opts.TitleOpts(title='房屋朝向占比',pos_left='33%',pos_top="5%"),
                        legend_opts=opts.LegendOpts(type_="scroll", pos_left="80%",pos_top="25%",orient="vertical")
                        )
        .set_series_opts(label_opts=opts.LabelOpts(formatter='{b}:{c} ({d}%)'),position="outside")
    )
c1.render_notebook()

装修情况/有无电梯玫瑰图(组合图)

fitment = df_fitment.index.tolist()
count1 = df_fitment.values.tolist()

directions = df_direction.index.tolist()
count2 = df_direction.values.tolist()

bar = (
    Bar()
    .add_xaxis(fitment)
    .add_yaxis('', count1, category_gap = '50%')
    .reversal_axis()
    .set_series_opts(label_opts=opts.LabelOpts(position='right'))    
    .set_global_opts(
        xaxis_opts=opts.AxisOpts(name='数量'),
        title_opts=opts.TitleOpts(title='装修情况/有无电梯玫瑰图(组合图)',pos_left='33%',pos_top="5%"),
        legend_opts=opts.LegendOpts(type_="scroll", pos_left="90%",pos_top="58%",orient="vertical")
    )
)

c2 = (
    Pie(init_opts=opts.InitOpts(
            width='800px', height='600px',
            )
       )
        .add(
        '',
        [list(z) for z in zip(directions, count2)],
        radius=['10%', '30%'],
        center=['75%', '65%'],
        rosetype="radius",
        label_opts=opts.LabelOpts(is_show=True),
        )    
        .set_global_opts(title_opts=opts.TitleOpts(title='有/无电梯',pos_left='33%',pos_top="5%"),
                        legend_opts=opts.LegendOpts(type_="scroll", pos_left="90%",pos_top="15%",orient="vertical")
                        )
        .set_series_opts(label_opts=opts.LabelOpts(formatter='{b}:{c} \n ({d}%)'),position="outside")
    )

bar.overlap(c2)
bar.render_notebook()

二手房楼层分布柱状缩放图

floor = df_floor.index.tolist()
count = df_floor.values.tolist()
bar = (
    Bar()
    .add_xaxis(floor)
    .add_yaxis('数量', count)
    .set_global_opts(
        title_opts=opts.TitleOpts(title='二手房楼层分布柱状缩放图'),
        yaxis_opts=opts.AxisOpts(name='数量'),
        xaxis_opts=opts.AxisOpts(name='楼层'),
        datazoom_opts=opts.DataZoomOpts(type_='slider')
    )
)
bar.render_notebook()

房屋面积分布纵向柱状图

area = df_area.index.tolist()
count = df_area.values.tolist()

bar = (
    Bar()
    .add_xaxis(area)
    .add_yaxis('数量', count)
    .reversal_axis()
    .set_series_opts(label_opts=opts.LabelOpts(position="right"))
    .set_global_opts(
        title_opts=opts.TitleOpts(title='房屋面积分布纵向柱状图'),
        yaxis_opts=opts.AxisOpts(name='面积(㎡)'),
        xaxis_opts=opts.AxisOpts(name='数量'),
    )
)
bar.render_notebook()

加载全部内容

相关教程
猜你喜欢
用户评论