斯坦福在线机器学习(Machine Learning)课程终于在昨天(10月10号)开课了，网址在 http://www.ml-class.org

和AI课程的视频放在youtube上不同，ML的在线视频提供下载，但是下载的版本没有英文字幕，故此放出自产自用字幕下载脚本，仿照以前的豆瓣电台歌曲信息脚本，也是从firefox获得cookie，然后爬得xml的字幕，再转换为srt格式的，这样就可以离线或者在iPad上看了！^_^

第一个是偷firefox的cookie爬网页，下载xml版的脚本，注意修改其中的firefox的cookie文件的目录，需要安装pysqlite包：

# coding: utf-8

import cookielib,urllib2
from cStringIO import StringIO
from pysqlite2 import dbapi2 as sqlite
import re,os

# a useful function from others
def sqlite2cookie(filename,host):
    con = sqlite.connect(filename)
    con.text_factory = str
    cur = con.cursor()
    cur.execute("select host, path, isSecure, expiry, name, value from moz_cookies where host like ?"
            ,['%%%s%%' % host])
    ftstr = ["FALSE","TRUE"]
    s = StringIO()
    s.write("""\
# Netscape HTTP Cookie File
# http://www.netscape.com/newsref/std/cookie_spec.html
# This is a generated file! Do not edit.
""")
    for item in cur.fetchall():
        s.write("%s\t%s\t%s\t%s\t%s\t%s\t%s\n" % (
            item[0], ftstr[item[0].startswith('.')], item[1],
            ftstr[item[2]], item[3], item[4], item[5]))
    s.seek(0)
    #print "cookie:",s.read() ;s.seek(0)
    cookie_jar = cookielib.MozillaCookieJar()
    cookie_jar._really_load(s, '', True, True)
    return cookie_jar

# get cookie
cookiejar = sqlite2cookie(r'C:\Users\lenovo\AppData\Roaming\Mozilla\Firefox\Profiles\osfuexqh.default\cookies.sqlite','ml-class.org')
opener = urllib2.build_opener(urllib2.HTTPCookieProcessor(cookiejar))
urllib2.install_opener(opener)

# post cookie
print 'Getting page, if it takes too long time, kill me!'
content=urllib2.urlopen('http://www.ml-class.org/course/video/list?mode=view').read()
#print content
content=''.join(content.split('\n'))
print "Page got!"

# get file name
namefinder=re.compile(r"file: ([^,]*),")
found=namefinder.findall(content)
print 'Number of files:',len(found)
found=map(lambda s: s[1:-1],found)
#print found

# down xml
baseurl='http://s3.amazonaws.com/stanford_videos/cs229/subtitles/%s-subtitles.xml'
for i,fn in enumerate(found):
    fn=fn.replace(r'\'','\'')
    outfile=fn+'.xml'
    if os.path.exists(outfile):
        continue
    print 'Getting:',fn
    fo=file(outfile,'w')
    fo.write(urllib2.urlopen(baseurl % fn).read())
    fo.close()
    print 'Done:',i,outfile

第二个是批量把xml脚本转化为srt字幕文件的脚本，各种字符串处理比较挫，见笑了：

# coding: utf-8
import os,sys,re,codecs

limit=[60,60,60,1000]
def xml2srt(fi,fo):
    data=''.join((fi.read().split('\n')[9:-4])).strip().split('</p>')
    for i in range(0,len(data)-1):
        #print i,data[i]
        if data[i]:
            st_st=data[i].index('"')
            st_ed=data[i].index('"',st_st+1)
            if i+1<len(data)-1:
                nx_st=data[i+1].index('"')
                nx_ed=data[i+1].index('"',nx_st+1)
            fo.write(str(i+1)+' \n')
            stamps=[data[i][st_st+1:st_ed],
                    data[i+1][nx_st+1:nx_ed] if i+1<len(data)-1 else "99:59:59.999"]
            word=data[i][data[i].index('>')+1:].replace('\n',' ')+' \n\n\n'
            for i,stamp in enumerate(stamps):
                stamp=stamp.split('.')
                stamps[i]=map(int,stamp[0].split(':'))
                stamps[i].append(int(stamp[1]))
            stamps=map(lambda s:"%02d:%02d:%02d,%03d" % tuple(s),stamps)
            fo.write("%s --> %s \n" % tuple(stamps))
            fo.write(word)
    #print 'OK!'

if __name__=='__main__':
    for fn in os.listdir('.'):
        if fn[-4:]!='.xml':
            continue
        print 'Converting:',fn[:-4]
        fi=file(fn,'r')
        if os.path.exists(fn[:-4]+'.srt'):
            continue
        fo=file(fn[:-4]+'.srt','w')
        xml2srt(fi,fo)
        fo.close()
        print 'Done'

感谢wr大牛(@isnowfy)的支持，附上wr大牛的单个字幕下载的简化版本，还有exe可执行文件哦！