This article mainly introduces how to convert an html table into a CSV file using python, and related skills related to how to operate a csv file using Python, for more information about how to convert an html table to a CSV file using python, see the following example. Share it with you for your reference. The details are as follows:
Usage: python html2csv. py *. html
This code uses the HTMLParser module.
#! /Usr/bin/python #-*-coding: ISO-8859-1-*-# Hello, this program is written in Python-http://python.orgprogramname = 'html2csv-version 2002-09-20-http://sebsauvage.net 'import sys, getopt, OS. path, glob, HTMLParser, retry: import psyco; psyco. jit () # If present, use psyco to accelerate the programspont: passdef usage (progname): ''' Display program usage. ''' progname = OS. path. split (progname ) [1] if OS. path. splitext (progname) [1] in ['. py ','. pyc']: progname = 'Python' + progname return ''' % sA coarse HTML tables to CSV (Comma-Separated Values) converter. syntax: % s source.html Arguments: source.html is the HTML file you want to convert to CSV. by default, the file will be converted to csv with the same name and the csv extension (source.html-> source.csv) You can use * and ?. Examples: % s mypage.html: % s *. htmlThis program is public domain. Author: Sebastien SAUVAGE
Http://sebsauvage.net ''' % (programname, progname) class html2csv (HTMLParser. HTMLParser): ''' A basic parser which converts HTML tables into CSV. feed HTML with feed (). get CSV with getCSV (). (See example below .) all tables in HTML will be converted to CSV (in the order they occur in the HTML file ). you can process very large HTML files by feeding this class with chunks of html while getting chunks of CSV by calling getCSV (). shold handle badly formated html (missing,,, Extraneous,...). This parser uses HTMLParser from the HTMLParser module, not HTMLParser from the htmllib module. example: parser = html2csv () parser. feed (open('mypage.html ', 'RB '). read () open('mytables.csv ', 'W + B '). write (parser. getCSV () This class is public domain. author: Sé bastien SAUVAGE
Http://sebsauvage.net Versions: 2002-09-19:-First version 2002-09-20:-now uses HTMLParser. HTMLParser instead of htmllib. HTMLParser.-now parses command-line. To do:-handle
tags - convert html entities (&name; and &#ref;) to Ascii. ''' def __init__(self): HTMLParser.HTMLParser.__init__(self) self.CSV = '' # The CSV data self.CSVrow = '' # The current CSV row beeing constructed from HTML self.inTD = 0 # Used to track if we are inside or outside a ... tag. self.inTR = 0 # Used to track if we are inside or outside a ... tag. self.re_multiplespaces = re.compile('\s+') # regular expression used to remove spaces in excess self.rowCount = 0 # CSV output line counter. def handle_starttag(self, tag, attrs): if tag == 'tr': self.start_tr() elif tag == 'td': self.start_td() def handle_endtag(self, tag): if tag == 'tr': self.end_tr() elif tag == 'td': self.end_td() def start_tr(self): if self.inTR: self.end_tr() # implies self.inTR = 1 def end_tr(self): if self.inTD: self.end_td() # implies self.inTR = 0 if len(self.CSVrow) > 0: self.CSV += self.CSVrow[:-1] self.CSVrow = '' self.CSV += '\n' self.rowCount += 1 def start_td(self): if not self.inTR: self.start_tr() # implies self.CSVrow += '"' self.inTD = 1 def end_td(self): if self.inTD: self.CSVrow += '",' self.inTD = 0 def handle_data(self, data): if self.inTD: self.CSVrow += self.re_multiplespaces.sub(' ',data.replace('\t',' ').replace('\n','').replace('\r','').replace('"','""')) def getCSV(self,purge=False): ''' Get output CSV. If purge is true, getCSV() will return all remaining data, even if or are not properly closed. (You would typically call getCSV with purge=True when you do not have any more HTML to feed and you suspect dirty HTML (unclosed tags). ''' if purge and self.inTR: self.end_tr() # This will also end_td and append last CSV row to output CSV. dataout = self.CSV[:] self.CSV = '' return dataoutif __name__ == "__main__": try: # Put getopt in place for future usage. opts, args = getopt.getopt(sys.argv[1:],None) except getopt.GetoptError: print usage(sys.argv[0]) # print help information and exit: sys.exit(2) if len(args) == 0: print usage(sys.argv[0]) # print help information and exit: sys.exit(2) print programname html_files = glob.glob(args[0]) for htmlfilename in html_files: outputfilename = os.path.splitext(htmlfilename)[0]+'.csv' parser = html2csv() print 'Reading %s, writing %s...' % (htmlfilename, outputfilename) try: htmlfile = open(htmlfilename, 'rb') csvfile = open( outputfilename, 'w+b') data = htmlfile.read(8192) while data: parser.feed( data ) csvfile.write( parser.getCSV() ) sys.stdout.write('%d CSV rows written.\r' % parser.rowCount) data = htmlfile.read(8192) csvfile.write( parser.getCSV(True) ) csvfile.close() htmlfile.close() except: print 'Error converting %s ' % htmlfilename try: htmlfile.close() except: pass try: csvfile.close() except: pass print 'All done. '
I hope this article will help you with Python programming.