Table of contents
Task-solving ideas and experience gains
I'm Zheng Yin looking forward to your attention
Hello everyone, this is Zheng Yin. Today I will teach you a relatively criminal case of reptiles
Python crawler crawls and downloads data from American scientific research sites
It is not easy to make a free follow and support it
Mission Introduction
-
Target website: https://app.powerbigov.us/view?r=eyJrIjoiYWEx…
-
Target data: Download tabular data for 2009-2013 and store as CSV file
The target website is the scientific research data of Beautiful Country and the webpage data implemented by PowerBI. The content cannot be copied with Ctrl+C. Therefore, we are asked to crawl for help.
Task-solving ideas and experience gains
First, the task can be disassembled into two parts: one is to crawl data from the website to the local, and the other is to parse the data and output the CSV file :
-
Crawl the data part:
-
Parse the web page and find the actual request address and parameters for asynchronous loading of data
-
Write crawler code to get all data
-
-
Parse the data part
This is the main difficulty of this task. The difficulty lies in: in the returned data list, the elements are not a fixed number, but only have different values from the previous row, and which column is different and which column is the same, is to use a "R" "value indicates that the normal solution is to reverse the JS to find the function that parses the R relationship, and complete the parsing . However, because the JS of the web page is very complex, and many function names are shorthand, it is very difficult to read, and there has been no reverse success.
In order to solve this problem, the first version was manually queried and summarized, and the first version was completed. Unexpectedly, when writing this sharing article, the idea suddenly opened up, the method of requesting data was changed, and the steps of analyzing the R relationship were bypassed:
-
Option 1: Parse data using R relation according to normal request
After downloading the complete data and analyzing it, among the required data from 2009 to 2013, there are a total of 124 R relationships, with a minimum value of 0 and a maximum value of 4083. By manually querying these 124 relationships, a dictionary is made to complete the analysis. The relationship summarized is as follows (manual query for 5 hours, tired):
-
Option 2: Exchange time for space, and request only one row of data at a time, bypassing the problem of parsing the relationship between
R , then there is no possibility of the same situation as the previous line, in which case the difficulty of parsing the R relationship can be bypassed. After the test, the solution is feasible, but the following issues need to be considered:-
Turn on multi-thread acceleration to shorten the time, but even if multi-threading is turned on, you can only open 12 threads according to 12 years, and the year with the largest number of lines is about 20,000 lines, and the crawler needs to run for about 5 to 6 hours
-
Breakpoints continue to climb to avoid the need to start from the beginning after the program is abnormally interrupted;
-
Specific steps
-
target site analysis
The first step is of course to analyze the target website and find the correct request address for the data. This is very easy. Open Chrome's developer mode, drag the scroll bar down, and see the new request, which is the real address. Go directly to the result:
Then look at the parameters of the POST request
request parameters
Looking at the Headers again, I was surprised to find that there was no anti-climbing! No back climbing! No back climbing! Well, Pretty Country's website is the atmosphere.
- Analysis summary:
# 完整参数就略过，关键参数以下三项： # 1.筛选年份的参数，示例： param['queries'][0]['Query']['Commands'][0]['SemanticQueryDataShapeCommand']['Query']['Where'][0]['Condition']['In']['Values'][0][0]['Literal']['Value'] = '2009L' # 2.请求下一批数据(请求首批数据时无需传入该参数)，示例： param['queries'][0]['Query']['Commands'][0]['SemanticQueryDataShapeCommand']['Binding']['DataReduction']['Primary']['Window']['RestartTokens'] = [["'Augusto E Caballero-Robles'","'Physician'","'159984'","'Daiichi Sankyo, Inc.'","'CC0131'","'Basking Ridge'","'NJ'","'Compensation for Bona Fide Services'","2009L","'4753'"]] # 注：以上"RestartTokens"的值在前一批数据的response中，为上一批数据的返回字典值，示例res['results'][0]['result']['data']['dsr']['DS'][0]['RT'] # 3.请求页面的行数(浏览器访问默认是500行/页,但爬虫访问的话...你懂的)，示例： param['queries'][0]['Query']['Commands'][0]['SemanticQueryDataShapeCommand']['Binding']['DataReduction']['Primary']['Window']['Count'] = 500 # 参数还有很多，例如排序的参数ordby，各种筛选项等-
URL to request data:
https://wabi-us-gov-virginia-api.analysis.usgovcloudapi.net/public/reports/querydata?synchronous=true -
Key parameters of POST:
-
-
The main steps and code of the crawler
-
To get all the data, a while True infinite loop is used. If there is an "RT" keyword in the return value of each request, the POST parameters are modified and the next request is initiated until there is no "RT" keyword in the return value, which means All data crawling ends (see code for details)
-
Exceptions in the request process need to be caught, and the next step is determined according to the type of the exception. Since the website has no reverse crawling, only exceptions such as timeout or connection error, you only need to restart the request, so you can continue crawling regardless of breakpoints
-
The above steps, the detailed code is as follows:
-
There is no anti-crawling of the website. After finding the correct path and parameters, the implementation of the crawler code is relatively simple. You can directly initiate a post request. The code is implemented through the PageSpider class (detailed code is attached)
-
In the resuming of the breakpoint, the process is solved, each line of data is stored in the TXT file, the file name records the year and the number of lines, first read the crawled records, find the result of the last request, and then initiate subsequent requests.
-
""" 爬取页面数据的爬虫 """ import pathlib as pl import requests import json import time import threading import urllib3 def get_cost_time(start: time.time, end: time.time = None): """ 计算间隔时长的方法 :param start: 起始时间 :param end: 结束时间，默认为空，按最新时间计算 :return: 时分秒格式 """ if not end: end = time.time() cost = end - start days = int(cost / 86400) hours = int(cost % 86400 / 3600) mins = int(cost % 3600 / 60) secs = round(cost % 60, 4) text = '' if days: text = f'{text}{days}天' if hours: text = f'{text}{hours}小时' if mins: text = f'{text}{mins}分钟' if secs: text = f'{text}{secs}秒' return text class PageSpider: def __init__(self, year: int, nrows: int = 500, timeout: int = 30): """ 初始化爬虫的参数 :param year: 下载数据的年份，默认空，不筛选年份，取得全量数据 :param nrows: 每次请求获取的数据行数，默认500，最大30000(服务器自动限制，超过无效) :param timeout: 超时等待时长 """ self.year = year if year else 'all' self.timeout = timeout # 请求数据的地址 self.url = 'https://wabi-us-gov-virginia-api.analysis.usgovcloudapi.net/public/reports/querydata?synchronous=true' # 请求头 self.headers = { # 太长省略，自行在浏览器中复制 } # 默认参数 self.params = { # 太长省略，自行在浏览器中复制 } # 修改默认参数中的每次请求的行数 self.params['queries'][0]['Query']['Commands'][0]['SemanticQueryDataShapeCommand']['Binding']['DataReduction'][ 'Primary']['Window']['Count'] = nrows # 修改默认参数中请求的年份 if self.year != 'all': self.params['queries'][0]['Query']['Commands'][0]['SemanticQueryDataShapeCommand']['Query']['Where'][0][ 'Condition']['In']['Values'][0][0]['Literal']['Value'] = f'{year}L' @classmethod def read_json(cls, file_path: pl.Path): with open(file_path, 'r', encoding='utf-8') as fin: res = json.loads(fin.read()) return res def get_idx_and_rt(self): """ 获取已经爬取过的信息，最大的idx以及请求下一页的参数 """ single = True tmp_path = pl.Path('./tmp/') if not tmp_path.is_dir(): tmp_path.mkdir() files = list(tmp_path.glob(f'{self.year}_part*.txt')) if files: idx = max([int(filename.stem.replace(f'{self.year}_part', '')) for filename in files]) res = self.read_json(tmp_path / f'{self.year}_part{idx}.txt') key = res['results'][0]['result']['data']['dsr']['DS'][0].get('RT') if not key: single = False else: idx = 0 key = None return idx, key, single def make_params(self, key: list = None) -> dict: """ 制作请求体中的参数 :param key: 下一页的关键字RestartTokens，默认空，第一次请求时无需传入该参数 :return: dict """ params = self.params.copy() if key: params['queries'][0]['Query']['Commands'][0]['SemanticQueryDataShapeCommand']['Binding']['DataReduction'][ 'Primary']['Window']['RestartTokens'] = key return params def crawl_pages(self, idx: int = 1, key: list = None): """ 爬取页面并输出TXT文件的方法， :param idx: 爬取的索引值，默认为1，在每行爬取时，代表行数 :param key: 下一页的关键字RestartTokens，默认空，第一次请求时无需传入该参数 :return: None """ start = time.time() while True: # 创建死循环爬取直至结束 try: res = requests.post(url=self.url, headers=self.headers, json=self.make_params(key), timeout=self.timeout) except ( requests.exceptions.ConnectTimeout, requests.exceptions.ConnectionError, urllib3.exceptions.ConnectionError, urllib3.exceptions.ConnectTimeoutError ): # 捕获超时异常 或 连接异常 print(f'{self.year}_part{idx}: timeout, wait 5 seconds retry') time.sleep(5) # 休息5秒后再次请求 continue # 跳过后续步骤 except Exception as e: # 其他异常，打印一下异常信息 print(f'{self.year}_part{idx} Error: {e}') time.sleep(5) # 休息5秒后再次请求 continue # 跳过后续步骤 if res.status_code == 200: with open(f'./tmp/{self.year}_part{idx}.txt', 'w', encoding='utf-8') as fout: fout.write(res.text) if idx % 100 == 0: print(f'{self.year}的第{idx}行数据写入完成，已用时: {get_cost_time(start)}') key = json.loads(res.text)['results'][0]['result']['data']['dsr']['DS'][0].get('RT', None) if not key: # 如果没有RT值，说明已经全部爬取完毕了，打印一下信息退出 print(f'{self.year} completed max_idx is {idx}') return idx += 1 else: # 打印一下信息重新请求 print(f'{self.year}_part{idx} not 200，check please', res.text) continue def mul_crawl(year: int, nrows: int = 2): """ 多线程爬取的方法，注按行爬取 :param year: 需要爬取的年份 :param nrows: 每份爬取的行数，若每次仅爬取1行数据，nrows参数需要为2，才会有下一行，否则都是第一行 """ # 定义爬虫对象 spider = PageSpider(year, nrows=nrows) # 获取爬取对象已爬取的idx，key和是否完成爬取的信号single idx, key, single = spider.get_idx_and_rt() if not single: print(f'{year}年的共{idx}行数据已经全部下载，无需爬取') return print(f'{year}年的爬虫任务启动, 从{idx+1}行开始爬取') spider.crawl_pages(idx+1, key) # 特别注意，已经爬取了idx行，重启时，下一行需要+1，否则重启后，会覆盖一行数据 if __name__ == '__main__': pools = [] for y in range(2009, 2021): pool = threading.Thread( target=mul_crawl, args=(y, 2), name=f'{y}_thread' # 按行爬取，nrows参数需要为2 ) pool.start() pools.append(pool) for pool in pools: pool.join() print('任务全部完成')
-
Code running example:
Exchange time for space, request only one line at a time, bypassing R relational parsing
-
Analytical data
-
Option One
-
The difficult part of parsing the data is to find out the relationship between R. This part is solved by manual query. Let’s go directly to the code:
class ParseData:
"""
解析数据的对象
"""
def __init__(self, file_path: pl.Path = None):
"""
初始化对象
:param file_path: TXT数据存放的路径，默认自身目录下的tmp文件夹
"""
self.file_path = pl.Path('./tmp') if not file_path else file_path
self.files = list(self.file_path.glob('2*.txt'))
self.cols_dict = None
self.colname_dict = {
'D0': 'License Type',
'D1': 'License Number',
'D2': 'Manufacturer Full Name',
'D3': 'Manufacturer ID',
'D4': 'City',
'D5': 'State',
'D6': 'Full Name',
'D7': 'Payment Category',
'D8': 'Covered Recipient ID'
}
self.colname_dict_T = {v: k for k, v in self.colname_dict.items()}
def make_excels(self):
"""
将每个数据文件单独转换为excel数据表用于分析每份数据
:return:
"""
for file in self.files:
with open(file, 'r') as fin:
res = json.loads(fin.read())
dfx = pd.DataFrame(res['results'][0]['result']['data']['dsr']['DS'][0]['PH'][0]['DM0'])
dfx['filename'] = file.stem
dfx[['year', 'part']] = dfx['filename'].str.split('_', expand=True)
dfx['C_count'] = dfx['C'].map(len)
writer = pd.ExcelWriter(self.file_path / f'{file.stem}.xlsx')
dfx.to_excel(writer, sheet_name='data')
for k, v in res['results'][0]['result']['data']['dsr']['DS'][0]['ValueDicts'].items():
dfx = pd.Series(v).to_frame()
dfx.to_excel(writer, sheet_name=k)
writer.save()
print('所有数据均已转为Excel')
def make_single_excel(self):
"""
将所有数据生成一份excel文件，不包含字典
:return:
"""
# 合并成整个文件
df = pd.DataFrame()
for file in self.files:
with open(file, 'r') as fin:
res = json.loads(fin.read())
dfx = pd.DataFrame(res['results'][0]['result']['data']['dsr']['DS'][0]['PH'][0]['DM0'])
dfx['filename'] = file.stem
dfx[['year', 'part']] = dfx['filename'].str.split('_', expand=True)
dfx['C_count'] = dfx['C'].map(len)
df = pd.concat([df, dfx])
return df
def get_cols_dict(self):
"""
读取列关系的字典
:return:
"""
# 读取列字典表
self.cols_dict = pd.read_excel(self.file_path.parent / 'cols_dict.xlsx')
self.cols_dict.set_index('R', inplace=True)
self.cols_dict = self.cols_dict.dropna()
self.cols_dict.drop(columns=['C_count', ], inplace=True)
self.cols_dict.columns = [col.split(':')[-1] for col in self.cols_dict.columns]
self.cols_dict = self.cols_dict.astype('int')
def make_dataframe(self, filename):
"""
读取TXT文件，转换成dataframe
:param filename: 需要转换的文件
:return:
"""
with open(filename, 'r') as fin:
res = json.loads(fin.read())
df0 = pd.DataFrame(res['results'][0]['result']['data']['dsr']['DS'][0]['PH'][0]['DM0'])
df0['R'] = df0['R'].fillna(0)
df0['R'] = df0['R'].map(int)
values_dict = res['results'][0]['result']['data']['dsr']['DS'][0]['ValueDicts']
dfx = []
for idx in df0.index:
row_value = df0.loc[idx, 'C'].copy()
cols = self.cols_dict.loc[int(df0.loc[idx, 'R'])].to_dict()
row = {}
for col in ['License Type', 'License Number', 'Manufacturer Full Name', 'Manufacturer ID', 'City', 'State',
'Full Name', 'Payment Category', 'Disclosure Year', 'Covered Recipient ID', 'Amount of Payment',
'Number of Events Reflected']:
v = cols.get(col)
if v:
value = row_value.pop(0)
if col in self.colname_dict.values():
if not isinstance(value, str):
value_list = values_dict.get(self.colname_dict_T.get(col), [])
value = value_list[value]
row[col] = value
else:
row[col] = None
row['R'] = int(df0.loc[idx, 'R'])
dfx.append(row)
dfx = pd.DataFrame(dfx)
dfx = dfx.fillna(method='ffill')
dfx[['Disclosure Year', 'Number of Events Reflected']] = dfx[
['Disclosure Year', 'Number of Events Reflected']].astype('int')
dfx = dfx[['Covered Recipient ID', 'Full Name', 'License Type', 'License Number', 'Manufacturer ID',
'Manufacturer Full Name', 'City',
'State', 'Payment Category', 'Amount of Payment', 'Number of Events Reflected', 'Disclosure Year',
'R']]
return dfx
def parse_data(self, out_name: str = None):
"""
解析合并数据
:param out_name: 输出的文件名
:return:
"""
df = pd.DataFrame()
for n, f in enumerate(self.files):
dfx = self.make_dataframe(f)
df = pd.concat([df, dfx])
print(f'完成第{n + 1}个文件，剩余{len(self.files) - n - 1}个，共{len(self.files)}个')
df.drop(columns='R').to_csv(self.file_path / f'{out_name}.csv', index=False)
return df
-
Option II
When using scheme 2 to process data, after performing data posteriori, it is found that there are still two detailed problems to be solved:
one is that a new keyword "Ø" appears in the return value. There is a case where the value itself is null. After traversing the data, it is found that there are only 3 different values (60, 128, 2048). Therefore, col_dict is manually made (see the code for details). \
class ParseDatav2:
"""
解析数据的对象第二版，将按行爬取的的json文件，转换成dataframe，增量写入csv文件，
因每次请求一行，首行数据不存在与上一行相同情形，因此，除个别本身无数据情况，绝大多数均为完整的12列数据,
"""
def __init__(self):
"""
初始化
"""
# 初始化一行的dataframe，
self.row = pd.DataFrame([
'Covered Recipient ID', 'Full Name', 'License Type', 'License Number', 'Manufacturer ID',
'Manufacturer Full Name', 'City', 'State', 'Payment Category', 'Amount of Payment',
'Number of Events Reflected', 'Disclosure Year'
]).set_index(0)
self.row[0] = None
self.row = self.row.T
self.row['idx'] = None
# 根据 Ø 值的不同选择不同的列，目前仅三种不同的Ø值，注0为默认值，指包含所有列
self.col_dict = {
# 完整的12列
0: ['License Type', 'License Number', 'Manufacturer Full Name', 'Manufacturer ID', 'City', 'State',
'Full Name', 'Payment Category', 'Disclosure Year', 'Covered Recipient ID', 'Amount of Payment',
'Number of Events Reflected'],
# 有4列是空值，分别是 'Manufacturer Full Name', 'Manufacturer ID', 'City', 'State'
60: ['License Type', 'License Number',
'Full Name', 'Payment Category', 'Disclosure Year', 'Covered Recipient ID', 'Amount of Payment',
'Number of Events Reflected'],
# 有1列是空值，是 'Payment Category'
128: ['License Type', 'License Number', 'Manufacturer Full Name', 'Manufacturer ID', 'City', 'State',
'Full Name', 'Disclosure Year', 'Covered Recipient ID', 'Amount of Payment',
'Number of Events Reflected'],
# 有1列是空值，是 'Number of Events Reflected'
2048: ['License Type', 'License Number', 'Manufacturer Full Name', 'Manufacturer ID', 'City', 'State',
'Full Name', 'Payment Category', 'Disclosure Year', 'Covered Recipient ID', 'Amount of Payment'],
}
# 列名转换字典
self.colname_dict = {
'License Type': 'D0',
'License Number': 'D1',
'Manufacturer Full Name': 'D2',
'Manufacturer ID': 'D3',
'City': 'D4',
'State': 'D5',
'Full Name': 'D6',
'Payment Category': 'D7',
'Covered Recipient ID': 'D8'
}
# 储存爬取的json文件的路径
self.data_path = pl.Path('./tmp')
# 获取json文件的迭代器
self.files = self.data_path.glob('*.txt')
# 初始化输出文件的名称及路径
self.file_name = self.data_path.parent / 'data.csv'
def create_csv(self):
"""
先输出一个CSV文件头用于增量写入数据
:return:
"""
self.row.drop(0, axis=0).to_csv(self.file_name, index=False)
def parse_data(self, filename: pl.Path):
"""
读取按1行数据请求获取的json文件，一行数据
:param filename: json文件的路径
:return: None
"""
row = self.row.copy() # 复制一行dataframe用于后续修改
res = PageSpider.read_json(filename)
# 获取数据中的valuedicts
valuedicts = res['results'][0]['result']['data']['dsr']['DS'][0]['ValueDicts']
# 获取数据中每行的数据
row_values = res['results'][0]['result']['data']['dsr']['DS'][0]['PH'][0]['DM0'][0]['C']
# 获取数据中的'Ø'值(若有)，该值代表输出的行中，存在空白部分，用于确定数据列
cols = ic(self.col_dict.get(
res['results'][0]['result']['data']['dsr']['DS'][0]['PH'][0]['DM0'][0].get('Ø', 0)
))
# 遍历每行数据，修改row这个dataframe的值
for col, value in zip(cols, row_values):
ic(col, value)
colname = self.colname_dict.get(col) # colname转换，D0~D8
if colname: # 如果非空，则需要转换值
value = valuedicts.get(self.colname_dict.get(col))[0]
# 修改dataframe数据
row.loc[0, col] = value
# 写入索引值
row['idx'] = int(filename.stem.split('_')[-1].replace('part', ''))
return row
def run(self):
"""
运行写入程序
"""
self.create_csv()
for idx, filename in enumerate(self.files):
row = self.parse_data(filename)
row.to_csv(self.file_name, mode='a', header=None, index=False)
print(f'第{idx + 1}个文件{filename.stem}写入表格成功')
print('全部文件写入完成')
Second, for each row of data requests, nrows needs to be set to 2, and the last row of data cannot be obtained in this way. Therefore, the last row of data needs to be parsed from the last returned json data (see the LastRow class for details)
class LastRow:
"""
获取并写入最后一行数据的类
由于每次请求一行数据的方式，存在缺陷，无法获取到最后一行数据，
本方法是对最后一个能够获取的json(倒数第二行)进行解析，取得最后一行数据，
本方法存在缺陷，即默认最后一行“Amount of Payment”列值一定与倒数第二行不同，
目前2009年至2020年共12年的数据中，均满足上述条件，没有出错。
除本方法外，还可以通过逆转排序请求的方式，获取最后一行数据
"""
def __init__(self):
"""
初始化
"""
self.file_path = pl.Path('./tmp') # 存储爬取json数据的路径
self.files_df = pd.DataFrame() # 初始化最后一份请求的dataframe
# 列名对应的字典
self.colname_dict = {
'D0': 'License Type',
'D1': 'License Number',
'D2': 'Manufacturer Full Name',
'D3': 'Manufacturer ID',
'D4': 'City',
'D5': 'State',
'D6': 'Full Name',
'D7': 'Payment Category',
'year': 'Disclosure Year',
'D8': 'Covered Recipient ID',
'M0': 'Amount of Payment',
'M1': 'Number of Events Reflected'
}
self.data = pd.DataFrame() # 初始化最后一行数据data
def get_last_file(self):
"""
遍历文件夹，取得最后一份请求的dataframe
"""
self.files_df = pd.DataFrame(list(self.file_path.glob('*.txt')), columns=['filename'])
self.files_df[['year', 'idx']] = self.files_df['filename'].map(lambda x: x.stem).str.split('_', expand=True)
self.files_df['idx'] = self.files_df['idx'].str.replace('part', '')
self.files_df['idx'] = self.files_df['idx'].astype(int)
self.files_df.sort_values(by=['year', 'idx'], inplace=True)
self.files_df = self.files_df.drop_duplicates('year', keep='last')
def get_last_row(self, ser: pd.Series) -> pd.DataFrame:
"""
解析文件，获取最后一行的数据
:param ser: 一行文件信息的series
"""
# 读取json数据
res = PageSpider.read_json(ser['filename'])
# 获取values_dict
values_dict = res['results'][0]['result']['data']['dsr']['DS'][0]['ValueDicts']
# 获取文件中的第一行数据
row_values = res['results'][0]['result']['data']['dsr']['DS'][0]['PH'][0]['DM0'][0]['C']
# 获取文件中的下一行数据，因文件是倒数第二行的数据，因此下一行即为最后一行
next_row_values = res['results'][0]['result']['data']['dsr']['DS'][0]['PH'][0]['DM0'][1]['C']
# 初始化Series
row = pd.Series()
# 解析数据填充series
for k, col in self.colname_dict.items():
value = row_values.pop(0)
if k.startswith('D'): # 如果K值是D开头
values = values_dict[k]
if len(values) == 2:
value = next_row_values.pop(0)
value = values[-1]
elif k == 'year':
pass
else:
if next_row_values:
value = next_row_values.pop(0)
row[col] = value
row['idx'] = ser['idx'] + 1
row = row.to_frame().T
return row
def run(self):
"""
运行获取最后一行数据的方法
"""
self.get_last_file()
for i in self.files_df.index:
self.data = pd.concat([self.data, self.get_last_row(self.files_df.loc[i])])
self.data = self.data[[
'Covered Recipient ID', 'Full Name', 'License Type', 'License Number', 'Manufacturer ID',
'Manufacturer Full Name', 'City', 'State', 'Payment Category', 'Amount of Payment',
'Number of Events Reflected', 'Disclosure Year', 'idx'
]]
filename = self.file_path.parent / 'data.csv'
self.data.to_csv(filename, mode='a', index=False, header=None)
return self.data
-
Results display
year 2010
2015
end of 2020
-
Extended thinking
If you combine the above scheme 1 with scheme 2, sort out all the row samples of different R relations, use scheme 2 to crawl a small number of partial examples, then derive the complete R relation dictionary, and then use scheme 1 to crawl and parse , will save a lot of time. This method can be considered when the amount of data far exceeds the current amount.
Summarize
In the process of completing the whole project, I went through: dark and cool (completed some functions of the crawler in less than 1 hour) -> confused (JS failed in reverse, unable to summarize the relationship between R) -> anxiety and irritability (worry about not being able to complete the task, manually query the rules More than 5 hours) -> Kaiqiao (suddenly discovering new ideas during the review process) a series of processes. The final result was a relatively smooth completion of the entire task, and the biggest feeling was the development of ideas: when one road fails, maybe a different direction can solve the problem (Note: the count parameter 500 has been used from the beginning, but the number of requests has been increasing. the number of rows, and never thought that a small change such as reducing the number of requested rows could bring about a huge breakthrough).
I'm Zheng Yin looking forward to your attention
