@Bookend I seek your advice on a project, if you'd got a minute or two
Working on ML for classifying 'other' flags.
First I run the text through SyntaxNet, then run the resulting 'paths' (what I call them, probably not the proper terminology) through TensorFlow.
The paths are unique, and there can be anywhere from 1 to (inf) in a single flag. Usually less than 10.
Anyway, I get a result from TensorFlow on each 'path'. Each result is an inference from that 'path' based on what I trained it earlier, either helpful or declined.
I'm running the raw result from TensorFlow through a softmax to get convenient percentages - for example, a single path might have [31%, 69%] as a result, being [Helpful, Declined]
I'm looking for a way to combine all of those results into one prediction for the whole flag.
Usually there's only one or two 'hot' paths - one path might have [1%, 99%] while the others are more like [45%, 55%].
I'm tempted to just take the hottest result and use it as a representation, but that feels like throwing away data.
I'll probably end up experimenting with a bunch of things, but I wanted to know if anything immediately jumps out at you as a 'standard' mathematical solution
My goal is to get a feel on each flag for how confident the system is either way.
@Quill TensorFlow's prediction and confidence for each path supplied.
just a sec.
can't give real data, because flags.
jumped (VERB,ROOT) -> fox (NOUN,nsubj) -> the (DET,det)
jumped (VERB,ROOT) -> fox (NOUN,nsubj) -> quick (ADJ,amod)
jumped (VERB,ROOT) -> fox (NOUN,nsubj) -> brown (ADJ,amod)
jumped (VERB,ROOT) -> over (ADP,prep) -> dog (NOUN,pobj)
Smoothed out a little, and with some of the crap removed, but that's what each path looks like. Think of it as a flattened grammatical tree (each branch combination has its own line).
So 'blue ice cream' might turn into 'cream -> blue' and 'cream -> ice'
@Undo one thing that comes to mind is looking at the lower edge of confidence interval, that is subtracting the second number from the first one. This is sometimes helpful for ranking things when the amount of information about each option is different.
[ SmokeDetector ] URL-only title, bad keyword in body, bad keyword in title, link at end of body, link following arrow in body, pattern-matching website in body, pattern-matching website in title: onlineskincarereports.co.uk/androx-extreme/ by Jacobdred on askubuntu.com…
@ArtOfCode When you're planning . . . to sleep . . . and then you start wasting three hours on an answer and sleep at four a.m., you can't just keep up with Be Nice.
> But the females love the proper amount of muscle Alpha Tren. Some of this are Rachelle lefervre, Cam Gigandet, Brad Pitt, Will Smith, Dwayne Johnson (now he's slimmed down), Hugh Jackman, such like.
I am not going to report this, as SD cannot possibly catch those, but if you have a strong itch to flag something: this post is obvious, trolling mockery of this one.
[ SmokeDetector ] Bad keyword in body, blacklisted website in body, blacklisted website in title, link at end of body, pattern-matching website in body, pattern-matching website in title: prohealthguides.com by tiger tom12 on drupal.stackexchange.com…
Well the links look related to the q, but I won't f it as we won't want a user who went on a rampage not to be blacklisted, unless he makes some more okay posts
[ SmokeDetector ] URL in title, bad keyword in body, bad keyword in title, link at end of body, pattern-matching website in body, pattern-matching website in title: lumalifteye.com/musc by user117692 on security.stackexchange.com…
AFAICT, we have a grand total of two women on the Smokey list. Perhaps that's down to this being a male-dominated field, but perhaps it's because the masculinity of the project is, for want of a better phrase, scaring people, women especially, off.
This is gonna be a touchy subject, but I think having an explicit policy that says we're inclusive and will try our damnedest to make everyone feel safe/comfortable goes a long way towards helping.
Edit Summary for the TLDR
People are asking for a summary due to the enormity of this thread of posts and comments
In summary this post is to discuss the link of diminishing numbers of women in programming and the research that reveals this is due to women feeling isolated within such a male do...
I don't know how you can encourage more women in particular, as the site isn't very heavy on women - and there are some that like to remain anonymous (gender anonymous)
I made my profile name and pic to me, so it's obvious I am a woman so other women can feel like there's someone else and less hesitant shrug Don't know if it helps @Åna??
Someone's gonna come back to that and say it's unnecessary because we haven't had any incidents. Which is true, but having such a policy in place marks us out as an inclusive space, instead of having to work that out for yourself.
@Yvette less active encouragement ("you should join us"), more passive encouragement ("this is a safe space if you wanna help out")
interestingly we lost one of our regular women from another chat room and then I left, so they lost both. I felt I was singled out and in part it was because I tried to catch on with the humour and couldn't make it work and would be very sensitive if I felt like I was singled out for censure
I think you may find you rub shoulder's with more women on SO than you may realise.
the thing with the humour on these sites, I've noticed is I don't understand a lot of it. I don't know if this is age, culture, gender or a mixture @BrockAdams, are you savvy with meta humour?
Anyway, ironically, one of the things that drew me here was the seriousness of the place, with a smattering of humour I really enjoy and actually understand :)
@TIPS it's still a lot of hard work to keep up with the boys, as much as I like them, I really love men, which is why I am comfortable in a predominately male field, but as a group and the humour, can be an effort (and there is the age thing also, it's not all gender)
@ArtOfCode "Charcoal HQ actively forsters an attitude of equal opportunity and tolerance. As such we welcome members from minority groups and are mindful to be respectful of our fellow users." ??
