I know there's some discussion about "what's appropriate" for the Wikipedia API, and I'd just like to share my recent experience.
I was trying to download the Wikipedia entries for people, of which I found about 800,000. I had a scanner already written that could do the download, so I got started.
After running for about I day, I estimated that it would take about 20 days to bring all of the pages down through the API (running single-threaded.) At that point I gave up, downloaded the data dump (3 hours) and wrote a script to extract the pages -- it then took about an hour to the extraction, gzip compressing the text and inserting into a mysql database.
Don't be intimidated by working with the data dumps. If you've got an XML API that does streaming processing (I used .NET's XmlReader) and use the old unix trick of piping the output of bunzip2 into your program, it's really pretty easy.
2010/10/13 Paul Houle paul@ontology2.com
Don't be intimidated by working with the data dumps. If you've gotan XML API that does streaming processing (I used .NET's XmlReader) and use the old unix trick of piping the output of bunzip2 into your program, it's really pretty easy.
When I worked into it.source (a small dump! something like 300Mby unzipped), I used a simple do-it-yourself string python search routine and I found it really faster then python xml routines. I presume that my scripts are really too rough to deserve sharing, but I encourage programmers to write a "simple dump reader" using speed of string search. My personal trick was to build an "index", t.i. a list of pointers to articles and name of articles into xml file, so that it was simple and fast to recover their content. I used it mainly because I didn't understand API at all. ;-)
Alex
On 14 October 2010 09:37, Alex Brollo alex.brollo@gmail.com wrote:
2010/10/13 Paul Houle paul@ontology2.com
Don't be intimidated by working with the data dumps. If you've got an XML API that does streaming processing (I used .NET's XmlReader) and use the old unix trick of piping the output of bunzip2 into your program, it's really pretty easy.
When I worked into it.source (a small dump! something like 300Mby unzipped), I used a simple do-it-yourself string python search routine and I found it really faster then python xml routines. I presume that my scripts are really too rough to deserve sharing, but I encourage programmers to write a "simple dump reader" using speed of string search. My personal trick was to build an "index", t.i. a list of pointers to articles and name of articles into xml file, so that it was simple and fast to recover their content. I used it mainly because I didn't understand API at all. ;-)
Alex
Hi Alex. I have been doing something similar in Perl for a few years for the English Wiktionary. I've never been sure on the best way to store all the index files I create especially in code to share with other people like I would like to happen. If you'd like to collaborate or anyone else for that matter it would be pretty cool.
You'll find my stuff on the Toolserver: https://fisheye.toolserver.org/browse/enwikt
Andrew Dunbar (hippietrail)
2010/11/7 Andrew Dunbar hippytrail@gmail.com
On 14 October 2010 09:37, Alex Brollo alex.brollo@gmail.com wrote:
Hi Alex. I have been doing something similar in Perl for a few years for the English Wiktionary. I've never been sure on the best way to store all the index files I create especially in code to share with other people like I would like to happen. If you'd like to collaborate or anyone else for that matter it would be pretty cool.
You'll find my stuff on the Toolserver: https://fisheye.toolserver.org/browse/enwikt
Thanks Andrew. I just got a toolserver account, but don't seach for any
contribution by me... I'm very worried about the whole stuff and needed skills. :-(
Alex
wikitech-l@lists.wikimedia.org