• Follow our Instagram!
  • Curious about what we're currently doing about bots? Check out the latest updates from staff in this thread.

[WIP] Introducing OU Scanalizer - AI/ML teambuilding tool for SV OU

Status
Not open for further replies.
Saddam_Hisui

Saddam_Hisui

1790191139322.png



TL;DR:
  1. Paste your team
  2. Server calculates matchup against 24 teams (mostly sample teams + you can add your own)
  3. Inspect 1v1 matchups against selected team (it assumes tera is free and hp is 100%)
  4. Consider replacment of pokemon given by model

Basically, this website is powered by 2 models: One 6v6 model that takes two teams as an input and outputs probability of team 1 winning. Second one is 1v1 model. It predicst which of 2 mons wins given matchup. Models are based in part on scrapped replay data and in part on simulated data of fights between two foul play bots.

In section "Chances Against Teams" you see probabilities od your team winning against given team. Probs are calculated based on 2 foul play bots fighting. They are not perfect bots and in my simulations they were given very short time to think, so your milage might vary. The bot might not found best way to play given team. However if the matchup is good, even weak bots/players should have advantages.

If you click on expand 1v1 you will see matrix which pokemon fares good against which. It is a probablility of winning 1v1 fight from full HP and tera free to pop. Below you get info, that from all given teams which pokemon is consistenly worst matchup for your team.
1790193307155.png

Then there are top replacement for each slot. You can see what predicitive model thinks is best for your team based it's knowledge. +pp means that it improves average matchup against given teams, -pp means that it would be worse (still displayed, if you want to consider some changes). Also this parts want to find similar pokemon that;s being replaced. If it is too differen it will display "outside similarity threshold"

  • "Real teams" - sets found in teams pasted in smogon forums etc. Overall more robust.
  • "Inputed teams" - Sets from teams created by sampling from usage stats. You might find some bullshit there.
1790193496607.png
Finally "Find a replacement". You pick pokemon you are unsure, you get few suggestions. It is very much in love with Hydraple, Gholdengo, Gliscor and Gargancl. Probably the part I'm least confident about.

Whole project is very much work in progress, however I feel like it might be usefull to somebody.

Feel free to report any bugs or give suggestions! I've been working on this project for a long time. After quite a while I was more interested of how to make it work properly than how do create good teams, so I didnt get much opportunities to make use of it. However it seems to understand matchups and roles of given mons, so I feel good about it.

Pinkacross blunder Finchinator
 
Peak teambuilding aside, ""your"" (claude's) site doesn't take into account chip damage, hazards or team support which are 3 capital elements of mons strategy, thus it is unable to provide any meaningful advice. That Poliwrath + Politoed team unironically had more thought and care put into it than your sloptool; maybe take notes from it.
 
Delibird Heart said:
Peak teambuilding aside, ""your"" (claude's) site doesn't take into account chip damage, hazards or team support which are 3 capital elements of mons strategy, thus it is unable to provide any meaningful advice
Click to expand...
Click to collapse...
Only 1v1 matrix.

It writes it exactly in screenshot you have taken.

I also don't see how comming up with team that has 3 mons with sub 1% usage in OU supposed to represent anything.
 
@ grok do I win this battle
@ grok is my matchup good
@ grok should I forfeit

just click find battle and learn. We have ample resources on this site that help players improve their knowledge without the need for AI to give quite literally worse quality info
 
Finchinator said:
Please do not tag me with anything pertaining to AI Pokemon unless you wish for me to give my full, unfiltered opinion. And I don’t think you want that.
Click to expand...
Click to collapse...
I want.

I am a statistician by trade as well as education. I've been doing "AI" before "AI" was a thing.

I worked in a banking sector and created models. for example for credit risk scoring.

I've worked on this project for couple of weeks now. Through my education and experience I think I've made a good job.

If you think otherwise, great! Let me know what is wrong.

However go on beyond superficial "AI bad".
 
It takes someone especially careless about the integrity of the sport and the future of competitive Pokémon to try and simplify and offload the process of improving like this.

Machine Learning tools like this aren’t morally neutral. You’re either ignorant, apathetic or malicious to ignore that. Let’s hope you’re the former two.
 
Let’s start with the basics: society is at a crisis where critical thinking is being replaced by AI.

This may seem cliché coming from an older, more traditional member of the community who is boy-cries-wolf’ing the “omg robots are taking over”, but, unfortunately, it’s dangerously real and we are already well beyond points of no return. I try to be open minded to new ideas and advancements in society and Pokemon alike because creativity and innovation lead to better life experiences, but there are plainly obvious pitfalls and risks that have been downright ignored. To put it bluntly, this is depressing and recent developments have even lead to me being less motivated to play or contribute. And that says something because I’ve been as close to a constant in the equation of Smogon over the last 10 years as humanly possible.

Forgetting Pokemon for a moment, I’m a 27 year old dude who is graduated, moved out, working, etc. Yea, I’m a dinosaur on here, but let’s take a step back…Before AI, I had to tackle so many challenges in life without any assistance. Some came easy, but honestly a lot were challenging. Just to list a handful: learning to live on my own, learning to prepare very basic food options, learning to research important topics for school or work, learning to care for myself when I didn’t feel well or what help to get, learning to write based on my own opinions, and learning to actually communicate in a professional and social fashion that fit my needs and personality. All of these things took learning, classic trial-and-error that mandated critical thinking and build character. These things are the foundation of who I am. I imagine the vast majority of people can say the same for themselves (probably even in a more succinct fashion than I can).

Nowadays at the first sign of any conflict or even on instinct before trying to critically think or problem solve, everyone is just searching on google (or chatgpt or Claude or whatever the fuck) how to do everything and parroting exactly what AI says. And this is completely negligent, too, as AI isn’t always accurate, even erring to dangerous degrees frequently. It is not designed for this. Quite frankly, AI is designed for specific professions and fields that it makes sense for and public overuse of it is the worst thing to happen to the world in recent memory aside from literal catastrophic events like genocide and war. Humans and society aren’t even actually functional people anymore and it’s absurd how compromised and reliant we are. Like god damn. Learn shit. Grow as a human. Build character.

There is an obvious “connect the dots” when it comes to Pokemon. People automating how to build or how to play. This is a game meant to challenge our understanding, strategy, intellect, and even temperament (something I firsthand have struggled with a ton admittedly). What the fuck are we doing here? Who are we trying to serve or please? You could’ve hit 1500 and actually learned a thing or two in the time it took you to make this slop and post this.

I will do you a favor and assume you’re acting in good faith, but let me ask you: do you even derive satisfaction from this? Is it seriously more gratifying than playing, learning, improving, and winning? Have you even spent the time to assess this and critically think about your options? Do us all a favor and try that. Until then, get the hell out of here with this slop.
 
Last edited:
Mr Zangler said:
It takes someone especially careless about the integrity of the sport and the future of competitive Pokémon to try and simplify and offload the process of improving like this.

Machine Learning tools like this aren’t morally neutral. You’re either ignorant, apathetic or malicious to ignore that. Let’s hope you’re the former two.
Click to expand...
Click to collapse...
Excuse me. I can build a team for idk, 30 minuts, 1 hour. Take 5 test games and then "Oh shit, Im a bit too weak for DD kyurem".

Or, help myself with some sort of tool that will tell me. You will lose to X. You will lose to Y.

Teambuilding in OU is most dificult stage. Many, many players dont build their own teams. Inexperience builders have 1 or 2 good teams. Only few people really can pump out many good teams.

I've made a tool that doesn't build teams.

It is a tool that helps you find weakness in preexisting teams and suggest some changes. That given if suggestions are not garbage, beceause building good models is hard.

Finchinator said:
I will do you a favor and assume you’re acting in good faith, but let me ask you: do you even derive satisfaction from this? Is it seriously more gratifying than playing, learning, improving, and winning? Have you even spent the time to assess this and critically think about your options? Do us all a favor and try that. Until then, get the hell out of here with this slop.
Click to expand...
Click to collapse...
I'm mid level ou player. I'm mostly ~1500, but my best game is around 1851 elo. You can check it on my RMT threads.

And I think it was good thing to do becasue 1) this is a project relates to my work experience 2) would not it be good thing to make most difficult aspect of playing OU (ie teambuilding) easier for everybody?

It would be very gratyfying for me to see that fruit of my free time and experience makes people life easier. That some people would use it and found out "oh shit, I lose to cm val" quicker and with fewer lost points.
 
Saddam_Hisui said:
2) would not it be good thing to make most difficult aspect of playing OU (ie teambuilding) easier for everybody?
Click to expand...
Click to collapse...
It gets easier the more you do it. It gets better the more experience you have. You grow stronger the more you challenge yourself and critically think. This is my whole point. And you actively taking the easy way out is doing both yourself and the community a disservice.
 
Finchinator said:
It gets easier the more you do it. It gets better the more experience you have. You grow stronger the more you challenge yourself and critically think. This is my whole point. And you actively taking the easy way out is doing both yourself and the community a disservice.
Click to expand...
Click to collapse...
Some people are good players. Some peolpe are good builders.

Why peolpe who can play but cant build get a bit of help in team building? There are many peoples like that. There were many tournament games where guy lost only because one guy had a known team.

Do we want to preserve teambuilding as most dificult aspect of game?

I don't think painfuly tring each iteration of my team, loosing games and points on ladder is most fun part of the game.

Maybe you will tell me that calcdex and damgae calc technology was a mistake and people shold rememeber all possible damage rolls in hear?
 
Honestly, you are all overreacting about it. As the author said, it does not completely trivialize the teambuilding process as it mostly tries to point out MU weaknesses and bring up suggestions to improve the team. As for the fact it's AI or not, it's entirely irrelevant as what matters is if the tool is functional or not. Right now, we can all agree that this is not the case and, if I'm honest, I don't see any practical usages with that tool in its current state. If you want to troll over it, at least don't be randomly anti-AI and don't forget to explain how the real issue is also the poor quality of the product.
 
Saddam_Hisui said:
View attachment 865299


TL;DR:
  1. Paste your team
  2. Server calculates matchup against 24 teams (mostly sample teams + you can add your own)
  3. Inspect 1v1 matchups against selected team (it assumes tera is free and hp is 100%)
  4. Consider replacment of pokemon given by model

Basically, this website is powered by 2 models: One 6v6 model that takes two teams as an input and outputs probability of team 1 winning. Second one is 1v1 model. It predicst which of 2 mons wins given matchup. Models are based in part on scrapped replay data and in part on simulated data of fights between two foul play bots.

In section "Chances Against Teams" you see probabilities od your team winning against given team. Probs are calculated based on 2 foul play bots fighting. They are not perfect bots and in my simulations they were given very short time to think, so your milage might vary. The bot might not found best way to play given team. However if the matchup is good, even weak bots/players should have advantages.

If you click on expand 1v1 you will see matrix which pokemon fares good against which. It is a probablility of winning 1v1 fight from full HP and tera free to pop. Below you get info, that from all given teams which pokemon is consistenly worst matchup for your team.
View attachment 865304
Then there are top replacement for each slot. You can see what predicitive model thinks is best for your team based it's knowledge. +pp means that it improves average matchup against given teams, -pp means that it would be worse (still displayed, if you want to consider some changes). Also this parts want to find similar pokemon that;s being replaced. If it is too differen it will display "outside similarity threshold"

  • "Real teams" - sets found in teams pasted in smogon forums etc. Overall more robust.
  • "Inputed teams" - Sets from teams created by sampling from usage stats. You might find some bullshit there.
View attachment 865305Finally "Find a replacement". You pick pokemon you are unsure, you get few suggestions. It is very much in love with Hydraple, Gholdengo, Gliscor and Gargancl. Probably the part I'm least confident about.

Whole project is very much work in progress, however I feel like it might be usefull to somebody.

Feel free to report any bugs or give suggestions! I've been working on this project for a long time. After quite a while I was more interested of how to make it work properly than how do create good teams, so I didnt get much opportunities to make use of it. However it seems to understand matchups and roles of given mons, so I feel good about it.
Click to expand...
Click to collapse...
"Chatgpt how do I play because Im to lazy to learn a pokemon battle sim type post" in all seriousness though AI has its place and uses and it's definitely not here, one of the fun parts of playing is Learning people are already at a point where they just ask AI to make them a team, like what level of lazy does someone gotta be to be like "Instead of actually learning through trial and error while along the way picking up useful game information, I'll just ask artificial intelligence!" Which is a stupid idea anyways AI HHallucinates information essentially speaking AI has a growing habit of making up information that doesn't actually exist and tries to make you think it's right. And now you want to use said AI for team building? Im sorry are you living in the same reality as us? Or in your own little world? That's just not gonna work, like actually good players will Not need this like what is this for? the lower ladder players that won't understand half of this anyways, it's an ambitious project bound to fail. Like seriously this feel like a slap in the face that people actually think that people should need to use AI to play a Video game like what level of lazy ignorance is this? I might be coming off as rude or blunt but this is my honest opinion.
 
"Ah yes, let me use the thing that constantly hallucinates things that are wrong, and has been shown to struggle at teambuilding time and time again, as a teambuilding tool and analysis of matchup issues."
View attachment 865325
1790203649174.png

How the actual fuck does this unviable dogshit have over a 1/3 chance of winning this matchup.

You have tried to use AI to work as a "team analysis and matchup spread calculator" when:
1) AI is known to just make shit up, as evidenced by the fact we just barely dodged a nuclear war caused by people blindly trusting AI.
2) AI is currently at the point where it cannot go band for band vs a top-level human player because AI is only good at playing certain ways currently.
3) AI is known to be garbage at teambuilding.
4) This does not analyze anything important and dumbs things down to 1v1s, which is not how pokemon is played.
 

Attachments

  • 1790203557647.png
    1790203557647.png
    39.8 KB · Views: 50
Saddam_Hisui said:
Excuse me. I can build a team for idk, 30 minuts, 1 hour. Take 5 test games and then "Oh shit, Im a bit too weak for DD kyurem".

Or, help myself with some sort of tool that will tell me. You will lose to X. You will lose to Y.
Click to expand...
Click to collapse...
The tool is called experience. It's called meta knowledge. It's called asking good players for advice. If you want to learn why they say something, ask them why they say it. Don't just pray to the "all-knowing" advanced autofill what you're gonna get beat by.
Saddam_Hisui said:
I want.

I am a statistician by trade as well as education. I've been doing "AI" before "AI" was a thing.

I worked in a banking sector and created models. for example for credit risk scoring.

I've worked on this project for couple of weeks now. Through my education and experience I think I've made a good job.

If you think otherwise, great! Let me know what is wrong.

However go on beyond superficial "AI bad".
Click to expand...
Click to collapse...
Let me expand beyond AI bad. AI is bad at doing every part of pokemon you are trying to use it for. You are the fucking bread companies putting AI into the bread, even though AI isn't edible. The tool utterly fails at what it is supposed to do because it lacks any ability to account for everything important.
Saddam_Hisui said:
Some people are good players. Some peolpe are good builders.

Why peolpe who can play but cant build get a bit of help in team building? There are many peoples like that. There were many tournament games where guy lost only because one guy had a known team.

Do we want to preserve teambuilding as most dificult aspect of game?

I don't think painfuly tring each iteration of my team, loosing games and points on ladder is most fun part of the game.

Maybe you will tell me that calcdex and damgae calc technology was a mistake and people shold rememeber all possible damage rolls in hear?
Click to expand...
Click to collapse...
You can get better at building by actually fucking building. That's called being human. Make, Test, Fail. Make, Test, Fail. Make, Test, Fail. Make, Test, Succeed. Growth at anything comes from the struggle, it comes from pushing yourself. I used to not be a good builder, but I actually practiced and got advice and now I am a competent builder.

Not to mention, I find the iteration aspect of teambuilding fun. I enjoy seeing it as a puzzle that I can solve. Shifting parts around to allocate different matchups to different pokemon to find ways to make teams more consistent, more resilient against cheese, more capable of surviving a bad matchup.
DIYUSI said:
Honestly, you are all overreacting about it. As the author said, it does not completely trivialize the teambuilding process as it mostly tries to point out MU weaknesses and bring up suggestions to improve the team. As for the fact it's AI or not, it's entirely irrelevant as what matters is if the tool is functional or not. Right now, we can all agree that this is not the case and, if I'm honest, I don't see any practical usages with that tool in its current state. If you want to troll over it, at least don't be randomly anti-AI and don't forget to explain how the real issue is also the poor quality of the product.
Click to expand...
Click to collapse...
I don't think that it's particularly random to be against the thing that is bankrupting the world, poisoning our water and air, and corroding our minds, sense of self, creativity, and agency.
Saddam_Hisui said:
Whelp, I guess I should catch on it. Right after I finish "Marxist critique of playing stall on ladder" and "Moral dilemas in webs HO".
Click to expand...
Click to collapse...
I love how you focused on the "moralities" part and missed the actual point of what was said. The intent behind the comparison to ethics and humanities in STEM is because it is analysis of how the thing you are making will impact the world. That is being compared to your inability to analyze how this tool will impact your own capabilities.
 
Saddam_Hisui said:
View attachment 865299


TL;DR:
  1. Paste your team
  2. Server calculates matchup against 24 teams (mostly sample teams + you can add your own)
  3. Inspect 1v1 matchups against selected team (it assumes tera is free and hp is 100%)
  4. Consider replacment of pokemon given by model

Basically, this website is powered by 2 models: One 6v6 model that takes two teams as an input and outputs probability of team 1 winning. Second one is 1v1 model. It predicst which of 2 mons wins given matchup. Models are based in part on scrapped replay data and in part on simulated data of fights between two foul play bots.

In section "Chances Against Teams" you see probabilities od your team winning against given team. Probs are calculated based on 2 foul play bots fighting. They are not perfect bots and in my simulations they were given very short time to think, so your milage might vary. The bot might not found best way to play given team. However if the matchup is good, even weak bots/players should have advantages.

If you click on expand 1v1 you will see matrix which pokemon fares good against which. It is a probablility of winning 1v1 fight from full HP and tera free to pop. Below you get info, that from all given teams which pokemon is consistenly worst matchup for your team.
View attachment 865304
Then there are top replacement for each slot. You can see what predicitive model thinks is best for your team based it's knowledge. +pp means that it improves average matchup against given teams, -pp means that it would be worse (still displayed, if you want to consider some changes). Also this parts want to find similar pokemon that;s being replaced. If it is too differen it will display "outside similarity threshold"

  • "Real teams" - sets found in teams pasted in smogon forums etc. Overall more robust.
  • "Inputed teams" - Sets from teams created by sampling from usage stats. You might find some bullshit there.
View attachment 865305Finally "Find a replacement". You pick pokemon you are unsure, you get few suggestions. It is very much in love with Hydraple, Gholdengo, Gliscor and Gargancl. Probably the part I'm least confident about.

Whole project is very much work in progress, however I feel like it might be usefull to somebody.

Feel free to report any bugs or give suggestions! I've been working on this project for a long time. After quite a while I was more interested of how to make it work properly than how do create good teams, so I didnt get much opportunities to make use of it. However it seems to understand matchups and roles of given mons, so I feel good about it.

Pinkacross blunder Finchinator
Click to expand...
Click to collapse...
i'm so fucking tempted. i could put you on blast so hard right now. give me a fucking excuse
 
DaddyBuzzwole said:
i'm so fucking tempted. i could put you on blast so hard right now. give me a fucking excuse
Click to expand...
Click to collapse...
ANinjaDude said:
"Ah yes, let me use the thing that constantly hallucinates things that are wrong, and has been shown to struggle at teambuilding time and time again, as a teambuilding tool and analysis of matchup issues."
View attachment 865325
View attachment 865326
How the actual fuck does this unviable dogshit have over a 1/3 chance of winning this matchup.

You have tried to use AI to work as a "team analysis and matchup spread calculator" when:
1) AI is known to just make shit up, as evidenced by the fact we just barely dodged a nuclear war caused by people blindly trusting AI.
2) AI is currently at the point where it cannot go band for band vs a top-level human player because AI is only good at playing certain ways currently.
3) AI is known to be garbage at teambuilding.
4) This does not analyze anything important and dumbs things down to 1v1s, which is not how pokemon is played.
Click to expand...
Click to collapse...

Jesus Christ, you guys are acting like this tool killed your mother.

Fine, have fun!
 
Status
Not open for further replies.
Back
Top