Hi Dylan,
The pagecounts-ez format isn't new, it's been there for years. More importantly the merge happens only once each day once all hourly files for that day are available (and then daily files are merged into a separate monthly file later on). So it's benefits is: one file instead of 24 or 31*24, with all hourly data preserved (in sparse arrays), and much less space (each title occurs once instead of up to 720 times). It's not intended for sites that need hourly updates. It's for archiving and longer term trend analysis.
Fortunately your migration task is much easier: just redirect your download script to https://dumps.wikimedia.org/other/pageviews/
You'll find the same hourly files you used up till now, in a downwards compatible scheme.
Main differences are: counts in new hourly files are cleaner (bot requests filtered out) and for each wiki requests from mobile devices are now included (desktop and mobile counts on separate lines).
I hope this helps,
Cheers,
Erik Zachte
From: Analytics [mailto:analytics-bounces@lists.wikimedia.org] On Behalf Of Jaime Crespo Sent: Wednesday, August 17, 2016 16:02 To: A mailing list for the Analytics Team at WMF and everybody who has an interest in Wikipedia and analytics. Subject: Re: [Analytics] Urgent Data Issue
we will be taking up more Wikimedia bandwidth
Please, note that from the operations side of things (Disclaimer: I am *not* a netops), I have the understanding that pure bandwidth usage is currently a non-issue (it is mostly a fixed cost, rather than a variable one). Hitting repeatedly a server is way more "costly" (all things considered, such as server purchase and maintenance) that a 1-time dump download. All dump users; use as much as you need (without wasting it) to meet your goals and do not worry too much about bandwidth.
Also want to say that we're very thankful for the work you all are doing publishing this dataset, it's enormously useful for entity popularity in our search engine for publishers https://graphiq.com/search .
My personal opinion is that, indeed, Analytics' work is very important for our mission (free knowledge spreading) and they are doing it great. I do not know if that is said enough.
On Tue, Aug 16, 2016 at 11:06 PM, Dylan Wenzlau dylan@graphiq.com wrote:
Thank you for the update. No one from our team is on the mailing list, and we have not viewed the /other/analytics page before (only the pagecounts-all-sites page https://wikitech.wikimedia.org/wiki/Analytics/Data/Pagecounts-all-sites and pages linked from there), which explains why we didn't know about this. I do see you recently added a link to Phabricator issue though, which is helpful!
I am currently rewriting our scripts to utilize the new pagecounts-ez format, although I think that this new format means that we will be taking up more Wikimedia bandwidth than we did previously, since we will have to re-downoad this merged daily file once per hour in order to utilize the hourly stats. Previously, we only had to download ~100MB per hour, and now it seems we'll be downloading ~350MB per hour. Please correct me if I'm missing something obvious here!
Also want to say that we're very thankful for the work you all are doing publishing this dataset, it's enormously useful for entity popularity in our search engine for publishers https://graphiq.com/search .
On Tue, Aug 16, 2016 at 1:48 PM, Dan Andreescu dandreescu@wikimedia.org wrote:
Dylan, there's also been a deprecation message on the page that links to these datasets, since last winter: https://dumps.wikimedia.org/other/analytics/
If you know of other places that these datasets are referenced, I'd be happy to update the docs and add links to the email threads. We usually publish information about this kind of deprecation on this list well in advance, but are open to reaching out in other ways.
On Tue, Aug 16, 2016 at 4:13 PM, Nuria Ruiz nuria@wikimedia.org wrote:
Dylan,
(cc-ing analytics@ public list)
Please see announcement about deprecation of datasets:
https://lists.wikimedia.org/pipermail/analytics/2016-August/005339.html
Thanks,
Nuria
On Tue, Aug 16, 2016 at 12:53 PM, Dylan Wenzlau dylan@graphiq.com wrote:
It seems the pagecounts-all-sites dumps have completely stopped updating, and I don't see any warning or message about why this is the case or whether it's currently being resolved. Our company relies pretty heavily on this data, as I imagine other projects & companies do as well, so I think it would be useful to at least display a big warning message on the documentation pages explaining why these are no longer updating.
Thanks,