last day (351 days later) » 

8:23 AM
I started a separate room about Deepsmoke (PR#1014) to keep this discussion from polluting CharcoalHQ and vice versa
2
I have been running with a couple of breaks since several hours now on all of Stack Overflow (including pt, ja, ru, are there others?) and the API is giving me a clean verdict for everything
this is still running on my laptop so it will break when I shut the lid or lose the network etc
I sent my SSH key to @Undo in response to his offer to host an experiment on Azure
 
I have invited @TajyMany here as well, in the hope that we could speed up communications compared to the slow trickle of comments to the PR ticket
if anybody from Charcoal would like to be added as a room owner here, just ping me (or one of the other room owners -- I added Undo and Art as owners already)
I would like to know what sort of timetable we can hope for for the Python-only implementation, and whether and if so when we can regard the API as stable
other pending questions are about the costs of running the thing on Bluemix (already covered in the ticket) and how to set up a host name for the current IP address (only discussed in chat)
I guess Art is the owner of charcoal-se.org so I'd like to ask him to add deepsmoke.charcoal-se.org for this service
 
Thanks for inviting me! Communicating here should be much more efficient.
 
9:19 AM
@TajyMany thanks for joining! good to see you here
@TajyMany so, any estimate for speeding up the stack with uwsgi or similar? where are you with that?
and how strict is the filter now, I have scanned on the order of 2,000 messages with no positive verdict
270 characters is not a lot for many of these posts, should it perhaps be longer?
I'm doing another run right now, can you say anything about the server-side load?
 
 
1 hour later…
10:44 AM
I have this up and running on a server now
not sure what sort of CPU/network allocation I have there, it's a shared server with a grumpy admin so I'm not entirely sure it will stay up
 
10:55 AM
... and it's got Python 3.4 which trips a bunch of random legacy errors, ugh
 
 
1 hour later…
12:20 PM
I have two instances running now ... will turn off the one on my laptop at least for now
I think I managed to fix the Python 3.4 legacy issue (different json internal namespace in Python 3.4) to the point where it has scanned 166 messages without falling over
 
12:52 PM
oh hi @Andy, I was going to add you as a room owner too but it says you already are (by way of being blue I suppose)
 
Lucky me :)
 
I have a cron job which puts the last 100 results in a static JSON file on a web server but I'm a bit hesitant to put that up for consumption
if I can get the Azure image up tomorrow I'd rather host everything there
1047 messages scanned by this new instance and still no hits
 
1:17 PM
Seems weird. I'd expect some
how long has it been running?
 
this instance for a little more than an hour, but I have been running this basically for 8 hours with just intermittent breaks and it has detected nothing
that's not completely out of line with expectations, there is not a large number of spam messages on SO and fewer still which are in some sense representative of what we see on other sites
chat.stackexchange.com/… shows about 20 from the time range and I could have had breaks when some of them were posted
but it's also not implausible that the model was retrained to avoid the false positives we reported before and is now just not very sensitive to contemporary spam
in any event, maybe I should simply run it on all messages, not just SO
gotta run now, will be back tomorrow unless there's something urgent ->
 
1:41 PM
Let's try it against a larger number of sites. Spam is spam. It should be able to find it no matter where it's posted
 

  last day (351 days later) »