Send AI mailing list submissions to
ai@lists.wikimedia.org
To subscribe or unsubscribe via the World Wide Web, visit
https://lists.wikimedia.org/mailman/listinfo/ai
or, via email, send a message with subject or body 'help' to
ai-request@lists.wikimedia.org
You can reach the person managing the list at
ai-owner@lists.wikimedia.org
When replying, please edit your Subject line so it is more specific
than "Re: Contents of AI digest..."
Today's Topics:
1. Wikilabels incident: Reversed diffs (Amir Tafreshi)
2. More/better model information and "threshold optimizations
(zppix e)
3. Re: More/better model information and "threshold
optimizations (zppix e)
4. Re: More/better model information and "threshold
optimizations (Aaron Halfaker)
------------------------------------------------------------ ----------
Message: 1
Date: Thu, 31 Aug 2017 18:06:19 +0400
From: Amir Tafreshi <amir.tafreshi_ext@wikimedia.de >
To: Application of Artificial Intelligence and other advanced
computing strategies to Wikimedia Projects <ai@lists.wikimedia.org>,
Wikimedia developers <Wikitech-l@lists.wikimedia.org >
Subject: [AI] Wikilabels incident: Reversed diffs
Message-ID:
<CAGe7S=apMa8h3d0=UeJOfYtPhqL-3P-640c-gdgM7z90J4v33w@mail. >gmail.com
Content-Type: text/plain; charset="utf-8"
Hello,
Today, we discovered a major regression in Wikilabels. We've patched the
issue and made an emergency deployment. We also deleted some labels that
were saved while the system was compromised.
Here is the details of of what happened:
https://phabricator.wikimedia.org/phame/post/view/69/ wikilabels_incident_reversed_ diffs/
We're very sorry for this inconvenience,
Yours truly
--
Amir Sarabadani Tafreshi
Software Engineer (contractor)
-------------------------------------
Wikimedia Deutschland e.V. | Tempelhofer Ufer 23-24 | 10963 Berlin
http://wikimedia.de
Wikimedia Deutschland - Gesellschaft zur Förderung Freien Wissens e.V.
Eingetragen im Vereinsregister des Amtsgerichts Berlin-Charlottenburg unter
der Nummer 23855 B. Als gemeinnützig anerkannt durch das Finanzamt für
Körperschaften I Berlin, Steuernummer 27/681/51985.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.wikimedia.org/pipermail/ai/attachments/ >20170831/fb0ad77c/attachment- 0001.html
------------------------------
Message: 2
Date: Thu, 31 Aug 2017 16:18:02 -0500
From: zppix e <megadev44s.mail@gmail.com>
To: AI@lists.wikimedia.org
Subject: [AI] More/better model information and "threshold
optimizations
Message-ID: <C80586B8-5BD2-419D-976B-E2E56C404D30@gmail.com >
Content-Type: text/plain; charset="utf-8"
We'd like to let you know about an upcoming change to ORES that might break your code. We've made some improvements to the way that "model_info" is reported from ORES. Rather than us needing to specify thresholds when we train a model (e.g. "filter_rate_at_recall(min_recall=0.9)"), the system now allows arbitrary queries (e.g. "filter_rate @ recall >= 0.9"). This if more flexible but it may cause problems for anyone who was using the machine-readable thresholds to support a UI in the past. Scores themselves will remain unchanged.
The model info and threshold systems will stay in the current state until we are ready to deploy the newer version to production (probably around Sept. 20th). If you have any questions, please feel free to contact the Scoring Platform Team, via email (ai@lists.wikimedia.net) or IRC (chat.freenode.net:6668 #wikimedia-ai)
--
Zppix
Volunteer developer for Wikimedia
Sent from my iPhone
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.wikimedia.org/pipermail/ai/attachments/ >20170831/409ff8fb/attachment- 0001.html
------------------------------
Message: 3
Date: Thu, 31 Aug 2017 16:20:22 -0500
From: zppix e <megadev44s.mail@gmail.com>
To: AI@lists.wikimedia.org
Subject: Re: [AI] More/better model information and "threshold
optimizations
Message-ID: <BC3D0A6F-EDBC-41E9-B540-D9054736AB68@gmail.com >
Content-Type: text/plain; charset="utf-8"
Sorry I ommited some info.
The blog post for more info can be found at:
https://phabricator.wikimedia.org/phame/post/view/68/more_ better_model_information_and_ threshold_optimizations/
> On Aug 31, 2017, at 4:18 PM, zppix e <megadev44s.mail@gmail.com> wrote:
>
> We'd like to let you know about an upcoming change to ORES that might break your code. We've made some improvements to the way that "model_info" is reported from ORES. Rather than us needing to specify thresholds when we train a model (e.g. "filter_rate_at_recall(min_recall=0.9)"), the system now allows arbitrary queries (e.g. "filter_rate @ recall >= 0.9"). This if more flexible but it may cause problems for anyone who was using the machine-readable thresholds to support a UI in the past. Scores themselves will remain unchanged.
>
> The model info and threshold systems will stay in the current state until we are ready to deploy the newer version to production (probably around Sept. 20th). If you have any questions, please feel free to contact the Scoring Platform Team, via email (ai@lists.wikimedia.net) or IRC (chat.freenode.net:6668 #wikimedia-ai)
>
>
> --
> Zppix
> Volunteer developer for Wikimedia
>
>
> Sent from my iPhone
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.wikimedia.org/pipermail/ai/attachments/ >20170831/7a087a9e/attachment- 0001.html
------------------------------
Message: 4
Date: Thu, 31 Aug 2017 16:21:30 -0500
From: Aaron Halfaker <aaron.halfaker@gmail.com>
To: Application of Artificial Intelligence and other advanced
computing strategies to Wikimedia Projects <ai@lists.wikimedia.org>
Subject: Re: [AI] More/better model information and "threshold
optimizations
Message-ID:
<CANQe2T8ZPkAw4W2vbSexCv-MUt_w_CLm1tUVZYiwR8aosR_grw@mail. >gmail.com
Content-Type: text/plain; charset="utf-8"
Just to be clear, the blog post contains detailed information about that
changes, a demo, and information about how you can test your tool against a
server running the updated code. https://ores-misc.wmflabs.org
On Thu, Aug 31, 2017 at 4:20 PM, zppix e <megadev44s.mail@gmail.com> wrote:
> Sorry I ommited some info.
> The blog post for more info can be found at:
> https://phabricator.wikimedia.org/phame/post/view/68/more_
> better_model_information_and_threshold_optimizations/
> On Aug 31, 2017, at 4:18 PM, zppix e <megadev44s.mail@gmail.com> wrote:
>
> We'd like to let you know about an upcoming change to ORES that might
> break your code. We've made some improvements to the way that "model_info"
> is reported from ORES. Rather than us needing to specify thresholds when
> we train a model (e.g. "filter_rate_at_recall(min_recall=0.9)"), the
> system now allows arbitrary queries (e.g. "filter_rate @ recall >= 0.9").
> This if more flexible but it may cause problems for anyone who was using
> the machine-readable thresholds to support a UI in the past. Scores
> themselves will remain unchanged.
>
> The model info and threshold systems will stay in the current state until
> we are ready to deploy the newer version to production (probably around
> Sept. 20th). If you have any questions, please feel free to contact the Scoring
> Platform Team, via email (ai@lists.wikimedia.net) or IRC (
> chat.freenode.net:6668 #wikimedia-ai)
>
>
> --
> Zppix
> Volunteer developer for Wikimedia
>
>
> Sent from my iPhone
>
>
> _______________________________________________
> AI mailing list
> AI@lists.wikimedia.org
> https://lists.wikimedia.org/mailman/listinfo/ai
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.wikimedia.org/pipermail/ai/attachments/ >20170831/385cd509/attachment. html
------------------------------
Subject: Digest Footer
_______________________________________________
AI mailing list
AI@lists.wikimedia.org
https://lists.wikimedia.org/mailman/listinfo/ai
------------------------------
End of AI Digest, Vol 21, Issue 4
*********************************