These are chat archives for beniz/deepdetect

4th
Jun 2018
cchadowitz-pf
@cchadowitz-pf
Jun 04 2018 14:47
hi @beniz - taking a look at #430 in relation to #325. It does seem like I'm getting different values, but I don't believe they're lining up with the values returned by the associated python script (still). I'll investigate a bit further though. If I understand #430 correctly, was the mean not being subtracted when provided via the input connecter as an array of values?
Emmanuel Benazera
@beniz
Jun 04 2018 14:49
It wasn't subtracted unless something broke it in the meantime.
cchadowitz-pf
@cchadowitz-pf
Jun 04 2018 14:50
you mean unless something broke it between last time we were looking into it and now?
Emmanuel Benazera
@beniz
Jun 04 2018 14:51
Yes
cchadowitz-pf
@cchadowitz-pf
Jun 04 2018 14:55
i see. i'm wondering if now that it's using them, i have the order wrong... i'm getting essentially the opposite class in DD that the python script is returning
0.990728974342 NSFW from the python script and 0.9972344040870667 safe for work from dede (true classification should be NSFW)
Emmanuel Benazera
@beniz
Jun 04 2018 15:15
If you've pulled the PR it is effective, I've tried with the yahoo model
Send me the pic url as a PM and I'll try when I can
cchadowitz-pf
@cchadowitz-pf
Jun 04 2018 15:17
oh i'm definitely seeing the output change when i tweak the mean values i'm providing to dede, but it hasn't aligned the outputs with the original yahoo model's python script output (unless you're saying you managed to do that?)
Emmanuel Benazera
@beniz
Jun 04 2018 15:22
No I haven't tested against the original python code, just highlighted it into the ticket
cchadowitz-pf
@cchadowitz-pf
Jun 04 2018 15:22
:+1: