I have a simple Bottle wrapper up and running now, scanned about a dozen posts so far, no FPs yet

60, still no FP

github.com/tripleee/deepsmoke-bottle <- I'll add a Readme with a pointer to a Docker image as soon as I have that uploaded

@Undo the demo is now having its log file displayed to this websocket

the output is JSON with the deepsmoke result included but no details

I'd like to expose the score from Keras as well, should be easy but I really need to focus on real work now

at least now with a stateful wsgi server (on localhost to boot) the individual deepsmoke checks are way below one second each

2550 messages scanned, no spam identified ...

@tripleee What are you running it on?

@thesecretmaster Undo donated an Ubuntu instance in Azure for me to run this on

but with Docker you should be able to run it pretty much anywhere

But what data?

the same posts that Smoke Detector scans

Oh, so it's watching the live websocket?

it's connecting to the Stack API using a stub version of the Smoke Detector code

@thesecretmaster yeah

And ... no spam so far?

@tripleee "anywhere" meaning anywhere you have Linux (which with Docker also means MacOS)

@thesecretmaster nnnnope

4881 messages scanned, 10 bad NS (I used the same code to test the bad NS when it wasn't in prod yet) and zero deepsmoke

@tripleee Weird. Is the code someplace?

I'm preparing to wrap up for the weekend here but if you have any quick questions, I'll be working on a report for another 20min or so

@thesecretmaster indeed it is, look in the deepsmoke-demo-double-down branch in my fork of SmokeDetector

@thesecretmaster that's just the Bottle wrapper

so I have one Docker with Deepsmoke and Bottle inside and another running (a fork of) SmokeDetector, connecting to the first one to get the Deepsmoke verdict from it

@tripleee Where's the important bit?

@thesecretmaster the one which connects to the Stack API and scans posts is in deepsmoke-demo.py

OK

... or something like that, deepsmoke-ws.py maybe? it's not hard to find if you locate the correct branch in the repo

I found it

@tripleee Could it just be the API being weird?

Or maybe @TajyMany modified the code and hasn't pushed to GH yet?

@thesecretmaster you mean the Stack Exchange API not returning full posts? or the Bottle API losing the message?

I think I have looked at debug logging at every stage but so far I was unable to produce any spam verdicts with the model I got from you ... if you have some messages which produced FPs for you, maybe I should try with them

@tripleee It seems like an issue between you and Tanmays API

Sure

1 sec

docker run --rm -it -p 8080:8080 tripleee/deepsmoke-bottle:68c68cce2483

allows you to run it locally on port 8080

That is spam for me.

works for me too

$ curl 'http://localhost:8080/spam?body=%3Cp%3EUse%20defines%20to%20stringify:%3C/p%3E%0A%0A%3Cpre%3E%3Ccode%3E#define%20LENSTR_(x)%20#x%0A#define%20LENSTR(x)%20LENSTR_(x)%0A%3C/code%3E%3C/pre%3E%0A%0A%3Cp%3Ethen%20you%20can%20use:%3C/p%3E%0A%0A%3Cpre%3E%3Ccode%3E#define%20MAXLEN%203%0A%0Achar%20a%5BMAXLEN%20+%201%5D;%0Ascanf(%22%22%25%22%22%20LENSTR(MAXLEN)%20%22%22s%22%22,%20a);%0A%3C/code%3E%3C/pre%3E'
{"score": 0.5577008128166199, "text": "<p>Use defines to stringify:</p>\n\n<pre><code>", "spam": true}

so I guess the API actually works

anyway, gtg, have a good weekend!

and thanks for the sample