Topic: Tagging Tool 2.8 + TagSet Extended (Read 726649 times)

Chris12 · « **Reply #300 on:** February 20, 2021, 09:18:48 am »

On to the tagging business. Goldo, I guess we have really different ideas about public sex. I have no idea how Onbed2 or Room2 could possibly be public. It's like Russell's teapot - sure, I can't prove that there is no other man in the room. But by the same logic, you also can't prove that there's a naked woman somewhere in the room, so it's also bisexual. Or that there's isn't a plug in her butt. You can "prove" pretty much anything with the Devil's proof.

Anyway, I don't wanna quarrel with you about this. I see your range for public is extremely wide, I can accept that.

In that case, I still think the location tags can be very useful to you, simply by putting them on some NOT list. (e.g. the "public acts" Fixation's not_list). If you find the time, implementing some logic adding a little flavor text for outside locations is probably the better solution, though. I think we all agree on this in principle.

Regarding 'onlookers', seems we agree about not adding the tag, though for different reasons. The tag would, in theory, allow you to find those images in town (or even the bar) where she's surrounded by people watching her in shock. Anyway, I also think it's ultimately not worth it, it's rare and would probably cause a lot of confusion. Rejected.

So conclusion. "public" stays as is, people can continue to tag inside images as public if they so desire. (Which I never disputed.) Just for the record. there's lots of public places not outside, e.g. in a school, an office building, or a bar or saloon. Anything but a private (bed)room is fine for me, too.
However, I agree we should leave the "public" definition vague, simply because it's an umbrella tag accounting for a really wide range of images. Like this, people in Goldo's camp can continue to tag public liberally, which is fine with me.

I hope everyone is happy with this and I didn't just argue this to death. I feel like ultimately, it was again a misunderstanding of tags in the context of the tagging tool and tags in the context of BK. I don't care at all how you use the tags inside BK, really. I'm completely fine if you exclude beach images from "public acts" if you want to reserve these for beach events, for example. Just put "beach" on the not list of the "public acts" Fixation. (BKinit_variables:1706)

@Jman: If Goldo really prefers the old, wider search, he can simply set the standard value for return_strict_immediately to False. Or fine-tune the value depending on fixation, or add a switch "more variety"/"better results" to the Settings screen, or whatever.

Anyway, I also lean towards better results here, but I can see how variety could also be a factor. In the case of get_fix_pic, you probably prefer exact results. It's weird to click "cum on her face" during sex and the pic shows cumming inside, although there would be "service on-face" images available.

Goldo · « **Reply #301 on:** February 22, 2021, 03:57:56 pm »

Sorry guys but I beg to differ on 'get_fix_pic()'.

Consider the following example:
- A girl pack has three doggy style pictures: "doggy sex.jpg", "doggy.jpg" and "doggy sex anal.jpg". The latter two being ambiguous pictures that can work for several sex acts. We know for a fact people will use all three tagging styles.
- Which one is best for 'doggy sex' advanced training? There's no way to tell. Maybe it's "doggy sex.jpg", but "doggy sex anal.jpg" seems equally legitimate for both acts, and "doggy.jpg" only has the one tag, so the pack maker probably thought it might be very fitting (perhaps the girl is bent against the wall, waiting for it, hence why no sex tag was added).
- Chris12's 'strict' algorithm will always return "doggy sex.jpg", which is wrong on two levels: not only will it reduce the pic variety that stand-alone "doggy.jpg" can provide, but it also excludes "doggy sex anal.jpg", even though it's perfectly valid for the 'sex' act: the tagger actively said so.

This kind of tag configuration is common in girl packs for all sorts of fixation tags, leading to people wondering why their pictures never actually showed up during training when using the algorithm you want to revert to. The way 'get_fix_pic' is set-up right now, all those pictures have a chance to show.

Jman · « **Reply #302 on:** February 22, 2021, 04:17:43 pm »

Okay, good point. I do think that giving some priority to the 'right sex act and no wrong sex act' pictures would be appropriate, so Chris's code can be useful there.

Goldo · « **Reply #303 on:** February 22, 2021, 04:54:30 pm »

Yes but 'wrong sex act' can be subjective. If a doggy picture is angled so that it could equally well be sex or anal, does that really count as the 'wrong sex act'?

Since I did not add AND, NAND and whatnot logic to the tagging system (on purpose, to preserve everyone's sanity), it's hard to differentiate between tags that are active together ('AND') and tags that can be either ('OR'). I try to infer that from context (group pictures are more likely 'AND' than 'OR', for instance).

Jman · « **Reply #304 on:** February 22, 2021, 05:03:47 pm »

Since the code can't see the pictures, it doesn't matter. If one picture is tagged 'sex doggy' and the other 'sex doggy anal', then it's very likely the pack maker figures the first is a better representative of straight-up sex than the other. And even if he does not, the game should, so that pack makers can promote precisely tagged pictures and the player is not drowning in generic 'doggy' pictures regardless of the act he chose.

If you have 'sex doggy', 'sex anal doggy', 'sex anal doggy (2)', 'doggy' and 'doggy (1)', the two acts have disturbingly similar fixation picture pools, even though 'sex' has a picture that's presumably a good representative of the activity.

Goldo · « **Reply #305 on:** February 22, 2021, 07:20:37 pm »

This is where I'd like experienced pack makers like Leortha to pitch in; but I know that when I tag, at least, I do not make the hierarchy you assume. If I tag a picture 'doggy sex anal' it's because I think it's a good picture for both sex acts, in no way inferior to 'doggy sex'. It's just the angle that's different, but there's no law that doggy style should always be presented from a back angle only...

Besides, your rule would effectively make any picture tagged 'doggy sex anal' disappear from doggy style training completely, as long as there's at least one other doggy picture for each sex act.

Now that we worked frequency tags into the game, if a picture is very inferior to others, the tagger can always explicitely convey it by adding a low frequency tag (or, you know, by not including the 'sex' tag at all in the first place!).

Leortha · « **Reply #306 on:** February 22, 2021, 08:21:55 pm »

I tend to agree with Goldo here. Unless we have a logical reason why the presence or absence or a secondary tag makes the primary tag less viable, I would be inclined to want to treat all instances of the primary tag as equivalent. When I tag "doggy sex anal.jpg", generally in my mind the pic is perfectly suitable for either sex or anal (or doggy training). The guy is behind her, and you cannot tell which is actually happening. So for the illusion of the game, it fits fine used either way.

And I'm definitely not saying there are no cases where additional tags change things. The most obvious would the farm tags that block images from other uses. In general these are what I tend to refer to as "modifier" tags, where the whole purpose of the tag is to modify the use of another tag. But other than specific situations like this, I tend to consider all pics tagged "sex" to be equally representative of the act.

I will note that "doggy sex anal" is something of a unique case, since by the nature of the act exactly which act is happening can be obscured.

I'll note that if I tagged one pic "doggy sex" and 10 pics "doggy sex anal" I would absolutely not want the first to be displayed every single time the game is looking for a pic with doggy and sex.

Jman · « **Reply #307 on:** February 22, 2021, 10:13:00 pm »

Quote from: Goldo on February 22, 2021, 07:20:37 pm

your rule would effectively make any picture tagged 'doggy sex anal' disappear from doggy style training completely

Note that I am not proposing to use Chris's code as posted. I'm advocating for putting some priority on 'perfect' pictures. The easiest would be to just do the same as the current code, but with an extra loop for the 'right act, no opposite act', so these have a bigger chance to be picked.

Quote from: Leortha on February 22, 2021, 08:21:55 pm

I'll note that if I tagged one pic "doggy sex" and 10 pics "doggy sex anal" I would absolutely not want the first to be displayed every single time the game is looking for a pic with doggy and sex.

This is not the problem I'm trying to solve. The problem arises when you also tag another 10 pictures 'doggy'. Then 'sex' and 'anal' have almost indistinguishable picture pools, despite there being a clearly exclusive 'sex doggy' picture. You could have just tagged everything 'sex anal fetish service doggy cumshot', and the difference would be minor.

Or to put it bluntly: the number of pictures tagged in a certain manner is what matters, rather than covering all the tags with distinct pictures. I think the tagging tool ratings outright mislead pack makers to believe that is not the case.

The current procedure means lots of ambiguity in the pack will result in lots of ambiguity in what's happening on-screen. If that's the intention, okay. I don't find it all that fun to not be able to tell from pictures what the naughty bits are doing. YMMV.

_neronero · « **Reply #308 on:** February 22, 2021, 10:38:41 pm »

There's a niche feature I've been thinking about lately (but hesitant to share, there are plenty of other things to work on after all)
But in this discussion it might be interesting to consider:

I'm trying to use the Extras > CG - Girl Packs gallery to hone my tagging craft. This feature is already very useful as-is for playtesting purposes. You are able to identify which images are tagged into an impossible corner and you can correct for it if needed.

Since a "picture frequency" feature is already making its way into game, how complex would it be to give images that aren't "unlocked" yet a higher priority to be displayed (as long as it fits the right criteria ofcourse)?

Another nice addition to this gallery for tagging/playtesting purposes, that could prove equally useful in picture selection: A counter of how many times every image has been pulled.

But like I said... Plenty of other things to focus on; I don't know how much work all this would entail.

Chris12 · « **Reply #309 on:** February 23, 2021, 04:31:25 am »

Goldo's concerns are valid, they're exactly why I said that image variety might be a factor. I think Leortha summed things up nicely.

You could argue "sex doggy" (an image clearly showing sex) is better than "anal sex doggy" (a picture showing something that could be sex). Even then, seeing "sex doggy" 100 times is probably inferior to seeing "sex doggy" 50 times, and "anal sex doggy 1" and "anal sex doggy 2" 25 times each.

Long story short, you can control this with the return_strict_immediately parameter. You might even rename it to very_strict if that makes things a little clearer. But your concerns are probably valid enough to warrant a default value of return_strict_immediately = False. If you want the old algorithm back completely, you can also remove and_not_combined from do_if_act_list. (After all, it does make "sex doggy" more likely to appear than "anal sex doggy", even if you set return_strict_immediately = False)

Basically, setting this parameter to False is what Jman is looking for. (making the "perfect" images more likely to appear, while also giving the others a fair chance)

In general, I was just jumping ahead with my usual over-eagerness, and trying things out. Take the parts you can use, and leave out the things you have no use for.

____________________

Jman, I think what you are pointing out is actually yet another problem. What about images missing any appropriate sex tags? My thoughts:

You could try to solve this with auto-tagging. But it would have to be fine-tuned on a per-tag basis.
"doggy" could be auto-tagged "doggy sex" if there are no other sexual tags. Or is it "sex anal", after all?
You probably would have to add a default "case" for each lonesome subtag. Sounds like a lot of work.
You could also be lazy and add a fourth step to do_if_act_list . (Which does no additional filtering, allowing "doggy" into the main selection.)
But again, this would have to be controlled on a per-tag basis. E.g. "profile cosplay" should probably not just appear when you have "sex cosplay" available.
Keep in mind that "doggy" is always a possible fallback. If the strict search for "sex doggy" returns no images, it will fall back to "doggy"
I think adding special logic for this case is probably a lot of work. I think it's also likely to create more problems than it solves. I do not recommend it.
Ultimately, I think this is another artificial discussion. I just quick-checked my girls folder and out of 2260 images tagged doggy, 2230 have at least one of the 'sex' or 'anal' tags.

_________________--

_neronero - if the likelyhood for a "freq_high" picture is indeed higher, your CG gallery should quickly converge to the point you desire, anyway. I wouldn't worry too much about this.

The counter idea is kinda neat. On the other hand, you have to save all that data somewhere, maybe even inviting performance problems. I personally don't use the CG Gallery much at all.

Jman · « **Reply #310 on:** February 23, 2021, 09:26:01 am »

Quote from: Chris12 on February 23, 2021, 04:31:25 am

Ultimately, I think this is another artificial discussion.

From an ultra-practical, 'this is how the girl pack situation is' perspective, certainly. From a 'give pack makers more power' point of view, maybe not so much, as I think my "tag everything 'sex anal fetish service doggy cumshot' " example shows.

But this is again veering into whether pack makers should use secondary tags in a primary role or, indeed, concern themselves with how the game handles the tags at all.

Quote from: Chris12 on February 23, 2021, 04:31:25 am

I think what you are pointing out is actually yet another problem. What about images missing any appropriate sex tags?
Keep in mind that "doggy" is always a possible fallback. If the strict search for "sex doggy" returns no images, it will fall back to "doggy"

These are fine, they go under 'no right sex act, no wrong sex act', i.e. the last step. So the fallback pictures never really apply in this situation.

Goldo · « **Reply #311 on:** February 23, 2021, 06:16:43 pm »

If you don't mind, on the specific topic of picture-picking algorithms, I'd rather focus the conversation on practical issues rather than theorical ones. By that I mean, show me instances of immersion-breaking pictures (with screenshots), rather than discussing hypotheticals that may or may not cause problems.

Otherwise, I'd rather go with the maxim 'if it ain't broke, don't fix it' (I can certainly use bits of Chris12's code to add a loop and optimize code readability a bit, but I am reluctant to change the current behavior without proof it's broken).

As for _neronero's suggestion, I share Chris12's concern about runaway performance problems. However, I could add it as a debug option, off by default, that could be activated through editing BKsettings.rpy or using the console.

Jman · « **Reply #312 on:** February 23, 2021, 09:21:47 pm »

Quote from: Goldo on February 23, 2021, 06:16:43 pm

By that I mean, show me instances of immersion-breaking pictures (with screenshots)

How would I do that for picture variety? The problem was never that you're getting immersion-breaking pictures, the problem was and is that one can inadverdently tag in a manner that doesn't make much of a difference in-game.

If you want an example, I found Lum from Urusei Yatsura to have tags like this: 2 exclusive 'sex cob', 3 shared 'sex group cob' and 8 'cob with no sex acts tagged'. The first two are by far the best fit when looking for 'sex cob', and show 30% of the time with the current algorithm. Going with 'return_strict_immediately = False' would push it slightly over 50%, which I find very reasonable from a variety POV.

Chris12 · « **Reply #313 on:** February 24, 2021, 04:23:11 am »

I think this is probably a matter of personal preference. As such, you could consider leaving this choice to the players through some option in the settings screen. E.g.

Image Variety
Low : try to find the most fitting images (return_strict_immediately = True)
Medium : give most fitting images a bonus, but also include others (return_strict_immediately = False)
High : Pick from all basically fitting images (and_not_combined removed from do_if_act_list)

If you go with something like that, it's probably best to remove the return_strict_immediately parameter from the function parameters. Instead, it could be set at the start of get_fix_pic()

return_strict_immediately = (image_variety <= 1) # Only return immediately with Image_Variety 'Low'
(...) # Initialize do_if_act_list with the two lambda's only
if (image_variety <= 2) : do_if_act_list.insert(0, and_not_combined) # Add an additional step with stricter criteria for Image_Variety 'Low' and 'Medium'

This assumes values of 1-3 for image variety.

Goldo · « **Reply #314 on:** February 25, 2021, 10:46:22 am »

Quote from: Jman on February 23, 2021, 09:21:47 pm

Quote from: Goldo on February 23, 2021, 06:16:43 pm
By that I mean, show me instances of immersion-breaking pictures (with screenshots)
How would I do that for picture variety? The problem was never that you're getting immersion-breaking pictures, the problem was and is that one can inadverdently tag in a manner that doesn't make much of a difference in-game.

If you want an example, I found Lum from Urusei Yatsura to have tags like this: 2 exclusive 'sex cob', 3 shared 'sex group cob' and 8 'cob with no sex acts tagged'. The first two are by far the best fit when looking for 'sex cob', and show 30% of the time with the current algorithm. Going with 'return_strict_immediately = False' would push it slightly over 50%, which I find very reasonable from a variety POV.

Sorry but I don't understand your numbers here. First, 'sex group cob' pics are completely outside of the discussion since group pictures are not used during sex training.

This leaves us with 2 exclusive 'sex cob' and 8 'cob with no sex acts tagged'. The chance to show one of the exclusive 'sex cob' pics is 50%, so 25% each. The chance to show one of the 8 generic cob pictures is also 50%, so in turn that's a 6.25% chance to show each of those pictures.

Isn't that the behavior you're asking for?

HentHighSchool Development Forum

News:

Author Topic: Tagging Tool 2.8 + TagSet Extended (Read 726649 times)

Chris12

Re: Tagging Tool 2.8 + TagSet Extended

Goldo

Re: Tagging Tool 2.8 + TagSet Extended

Jman

Re: Tagging Tool 2.8 + TagSet Extended

Goldo

Re: Tagging Tool 2.8 + TagSet Extended

Jman

Re: Tagging Tool 2.8 + TagSet Extended

Goldo

Re: Tagging Tool 2.8 + TagSet Extended

Leortha

Re: Tagging Tool 2.8 + TagSet Extended

Jman

Re: Tagging Tool 2.8 + TagSet Extended

_neronero

Re: Tagging Tool 2.8 + TagSet Extended

Chris12

Re: Tagging Tool 2.8 + TagSet Extended

Jman

Re: Tagging Tool 2.8 + TagSet Extended

Goldo

Re: Tagging Tool 2.8 + TagSet Extended

Jman

Re: Tagging Tool 2.8 + TagSet Extended

Chris12

Re: Tagging Tool 2.8 + TagSet Extended

Goldo

Re: Tagging Tool 2.8 + TagSet Extended