After far too long, the follow-up to this video is finally out. In it, I fix many of the issues with the original tournament and run the whole thing again, except this time with all the Pokémon flattened to level 50. Check it out: ruclips.net/video/247qD1qulSQ/видео.html
Its been a while, and a lot of work for you, but are you interested in continuing your Research? I would die to know whos the strongest Hoenn Trainer is. Maybe even whit all the Frontier Brains at Lv 50 included.
I mean, if she is spamming rest then she is doing her job. Her entire position in that team is to lure out high-damage low-pp moves and make you exhaust them as early as possible. The only issue is that she uses pokemon with some really unfortunate matchups, both against starters (pikachu/venusaur defeats water, and charizard ice) and various other beasts/TMs likely found in your team. Should have had a team of snorlax.
I actually laughed out loud when I noticed Lorelei's draw ratio. That infamous soft lock... Pokemon Yellow actually added a check that goes "are you Lorelei? if so, don't spam rest just because it's psychic type"
@@zacharyadler4071 it’s the sole reason that the All Terrain Venomoth from Twitch Plays Pokémon was also known as the Dragon Slayer. Also, I am pretty sure it was Barrier, Lance’s Pokémon have illegal moves at their disposal, namely Dragonite with Barrier.
I love that the trainers with just magikarp actually won a few games, I can only assume they were against trainers who just self destructed immediately so they had magikarp to spare
Its fun of those funny quirks of red/blue. The ai is stupid enough to sometimes lose against lv5 magikarp. Its proven to be possible to train a lv5 magikarp all the way to evolution and beyond in victory road, to the dismay of many attemps to stuck a save into indigo plateu
I always just imagine... What is going on in the minds of those Electrode trainers? "Yes, all of my Pokemon battles are just my Pokemon exploding immediately and me desperately hoping the enemy's Pokemon is fainted so I can be declared the winner". That's such a sad, braindead life.
@@plasmakitten4261 According to the game, their happiness stat would be incredibly low. However, according to the Pokédex, Electrode actually *enjoy* exploding.
I mean... Doesn't he still have roar as a gym leader? Roar didn't have a function in gen 1 trainer battles... So it always failed. But it's still an option to use, so another wasted turn between super potion on full health.
@@willofthewind If I recall, Blaine's AI is bugged and doesn't check the amount of health before healing. From a lore standpoint, I like to think that he's gone senile
@@UntarLaManteca most tier lists aren't really objective as much as they try to be. SMOGON for example base tier lists off of usage. The theory is a really good pokemon is going to be used far more often then a bad one because people are more likely to win battles. Obviously this isn't always the case as some pokemon aren't able to be slapped on any team but have some kind of niche that no other pokemon can do issue is these team archetypes won't be as common so it's tier placement becomes lower despite being perfectly viable in the higher tier. We also have cases of pokemon being more viable in higher tiers while the meta of their own tier is completely inhospitable to them for example shedinja in gen 3 was actually a legitimate threat in Ubers but would be laughed at in OU.
It might be because the game prioritizes Giovanni's ground-type moves against Gastly, a Ghost/Poison type, but Gastly dodges all those attacks with its Levitate ability. Although I think the game does ignore secondary types (as demonstrated in the Ivysaur vs. Oddish battle, so maybe it's more complicated than that.
Was fissure a ground move in gen 1? I just realized I don’t think I’ve ever used it. Maybe Giovanni’s AI thought “fissure ground and OHKO” Another reason could be that Giovanni’s gen 1 move sets are awful and are filled with normal type attacks.
Lorelei _specifically_ got patched, so she would not have produced this amount of draws with her Pokémon Yellow AI, which keep sher, and _only_ her, from spamming rest.
As a data engineer in real life, I really appreciated the "look at all this fancy mathematics...realize we don't like it...and just fudge the numbers until it feels right." That's basically 70% of the data science projects I see nowadays 👍
@@RadkeMaiden At least engineers do something useful. Dedicated mathematicians rarely make useful contributions to science or society, just argue over "different kinds of infinities" and other pointless constructs that have no bearing on our actual, physical reality.
@@Litevaar... Until we discover something those "useless" rules actually apply to. That's a mathematician's job. Discover and create new, undiscovered (and thus unused) rules that COULD be useful for future generations. Without a mathematical rule that precisely accounts for what you are doing, every other branch of science is stuck and must either create a new rule from 0 or find an old rule that also applies for wathever they are doing. That's true for every branch science, form rocket launching to statistical analysis. If there's a situation never encountered before, you either search for an elusive theorem that somehow works with what are you doing and that no one has ever noticed before (and, when that happens, tha mathematician that discovered it suddenly becomes popular and important) or you ask to current mathematicians to create a new theorem from 0.
17:53 it's interesting to see that, even in the ai tournament, one of the only two loses prof oak received was against Lance, just like the lore suggests (Lance defeated oak to become champion)
The other loss from Oak is Green3 (Charizard). Presuming the player character would beat Oak in the end as well, he would have two A+ tier protégés. Not bad for a professor.
@@dimitripetrograd9808 and theoretically if you pick Bulbasaur you fight green Charizard and oak blastoise meaning the worst starter to pick is Bulbasaur
Hey, thanks for watching! I'm working on a follow-up to this video where I run the tournament again with all the Pokémon level balanced (and some other changes...). If you're interested, be sure to subscribe! (...it'll be a while, though)
I went back and battled her as an adult and she was stupid easy She's hard when youre a kid because you dont properly level your pokemon or catch and level new pokemon for your team early in the game
This was amazing. It actually shows that despite the flak gen1 code gets, it was actually fairly well designed to give the player enough of a challenge at every stage of the game to keep things interesting.
Gen 1 is incredibly well made considering it’s limitations. It’s just slightly unfortunate that all the optimizations and compression also made it very easily become a buggy mess.
The programming effort that went into this is absolutely incredible. Modding ASM based GB games is tough enough as it is, and to then script emulator behavior and save mods as well, AND THEN to run several hundred thousand simulations? THEN APPLY STATISTICS VISUALIZATIONS?! S S S T I E R
I'm impressed! As someone who does scientific programming and statistics professionally, I shudder at the thought of how much work it must have been to compose all that, run the simulations, evaluate the data and on top of it all animate it! I know, it's largely a matter of how familiar you are with a code (in this case, that of pokemon), but I can literally feel the frustration of all the failed attempts due to stupid mistakes it would have taken me. Great work!
The Elo system broke down slightly at the top- we do see that the Venusaur rival had the best ratio of the three- but at that point it's largely splitting hairs.
I'm not sure how I felt about the Elo score without knowing the 'order' which battles were fought. in some sense it doesn't matter what the Elo score is The trainer who fought their worst matches first would have the best score. But that could have been the difference between rival champ and oak fights
I feel like there is some match fixing going on here. Venusaur only lost twice, and at least one of those losses was vs. itself, which means it has at most one loss vs the top 3. The Charizard trainer fought two more matches than Venusaur, and Blastoise fought twenty more than it! Conspiracy theory: did they let Charizard continue until it was above Venusaur? And Blastoise until it was in between?
@@T_Lorentzen I mean, most of the trainers have different numbers of total battles, it's likely he used some program or something to randomly decide the number of fights and who they fought, so some trainers had more fights than others.
@@BillyBob-xx5jr Well he said in the video that he made every trainer fight against every other trainer, plus some more battles on top for good measure, so he definitely did not intend to skip any trainers at the very least. Yet Venusaur Green at 4th place had itself as its worst loss and 3rd place Green as its best victory. Because it only lost twice and one of those losses was itself, this means it somehow never fought either the Prof Oak at 1st or the Green at 2nd. Even besides the fact that a match against itself should definitely not count for scoring, something is definitely pretty screwy with how the matchmaking was done.
No, all Oak wanted to do was "fiddle with his Pokedex" and he did not believe they were for fighting. Agatha talks about him before you battle her in the Elite Four.
Wow, this video was absolutely phenomenal. Interesting methodology and explanations, and the visuals were incredible. High quality work man. Amazing job! Was super intriguing all the way through!
This is the most interesting video I've seen on RUclips so far, answering so many questions I was asking myself as a kid. Like : "How the fuck is Brock a gym leader, if every trainers we encounter on roads are better than him ?" Now I know he would lose to almost everyone in Kanto.
I am a fellow data scientist and programmer. I just wanted to say that while watching this video I couldnt help but feel respect for the effort, setup, and visual presentation. So cool
Dat coverage though. A grass/psychic, a water type with mad defence that has an ice move, and a fire type strong physical attacker that knows Take Down.
I've been wondering what his other three would be if he wanted them. assuming he would choose pokemon with two evolutions. I would assume Electrode for speed and electric, almost guaranteed turn 1 tunderwave. Rhydon for ground/rock tank and good moveset without using the same type twice I could really only think of Golbat only for chip damage and confuse ray.
@@TheNerdyGinger If I'd have to guess, it is either accidentally posted here and was meant for another comment OR it was a reply to post which has been deleted since.
Between the editing and visualization, and the actual coding and running the tournament, I can't imagine the amount of work! Really entertaining and cool! 🔥
Would love to see other generations done! Also including legendary battles might be a good idea. Like the Mewtwo battle is rock hard so I’d be curious where it ranks.
That was pretty cool. I almost want to see the lesser trainers specifically ran through the gym bosses / elite 4 in order. Just to see how many would actually gain badges.
I wonder if giving a flat level to all the trainers would yield some different results. it would seem more interesting to see the worst and best AI in the game if level is no object.
I was thinking the same thing. It'd also be more interesting if they all had 6 pokemon, using duplicates to fill in missing slots, starting from the front. Rival 1 would have 6 lv50 squirtles. Mistie would have 3 alternating lv50 staryus and 3 lv50 starmies. Agatha would have her lv50 team plus an extra gengar. The later teams would still have an advantage, but it'd be so much more mitigated.
Surely, attack selection is an indication of skill though, and who would decide which attacks they replace? And if they keep the same attacks, they'd stay disadvantaged.
@@DutchDread Well yeah, that would be the key difference. What we're basically arguing for though is an AI, PvP version of these battles, where the only advantages come from typing, movesets, and skill. A guy with 2 voltorbs doesn't lose to a guy with 3 magikarp just because the former randomly chose self-destruct twice.
@@JKBrown-vv1yw Why arbitrarily select "typing, moveset, and skill" as acceptable though? The question is do you want to see which is the best trainer, or which is the best AI? Because if it's the second, then any disparity in team composition makes your findings mostly worthless. If you really want to JUST test AI, you should combine every AI with every team in the game and see which does best with the most teams.
These digital graphics are amazing. I don't even expect visuals this awesome from double digits million channels. The Rolodex scroll of the character profiles was impressive. Well done AF.
17:10 So Venusaur Champ Green only lost twice, without draws and his worst loss being.. himself? Looks like the real #1 trainer, ELO is somewhat random.
Charizard champ had 2 more losses and had a worse “worst defeat” with an A- loss. If this were a more manual ranking, like how college football ranks teams; Venasuar champ would be higher than the other rival champs
@@holdenfegan1613 from what i can gather venusaur rival lost to charizard rival and oak with blastoise. Charizard rival lost to blastoise rival, oak with venusaur, oak with charizard and loralei. Rival with blastoise lost to all the oak battles, and the rival with venusaur. I dunno where that 5th battle came from, i can't find it because it's lowest loss is oak with venusaur, we can figure out it lost to oak with charizard. It didn't lose to the nonchampion rival fight since that trainers best win is against itself. And we know it won vs rival with charizard. Lost to rival with venusaur, and lost to oak vs blastoise. I'm guessing this other loss is a repeat battle against one of it's losses. The oak with blastoise lost to the rival with charizard and Lance. Oak with venusaur lost to rival with venusaur and oak with blastoise. So the top 5 should be 5th place rival with charizard since it lost to loralei which is way down on the ranking. Since i can only find 4 losses for the blastoise rival, it would take 4th. 3rd would be oak with blastoise since it lost to lance who's further down in the rankings. 2nd would be rival with venusaur since it's lowest loss would now be the 5th place rival with charizard. And first oak with venusaur since it's lowest loss is the now 3rd oak with blastoise. This ranking isn't including draws since we have no information on those and we don't know how many of those draws were rematched with a decisive winner. If we include draws. It goes 5th rival with charizard, 4th rival with blastoise, 3rd oak with venusaur, 2nd oak with blastoise, and 1st rival with venusaur. It would be nice if we could actually see the listings of all the battles and where the extra battles that some trainers have came from and who they drew against
Really fascinating! And I love the presentation. It’s only a matter of time before manual editing is made redundant by programmatically rendering scenes...
Just wondering if bulbasaur family keeps pulling ahead of the other two starters in the lower tiers, because of enemy AI spamming poisonpowder/toxic futilely against bulba
0:10 What would happen if we had all the trainers battle each other? 3:00 2 Emulators are kept in sync. 3:27 Save-State Swapping between Emulators. 4:04 EXP gains --> Disabled by hard coding memory values. 5:04 True-Round Robin, (every trainer involved battles every other trainer) 6:00 DRAW Rules if the battle goes 75 turns w/o either side taking damage or 1,000 turns long. 6:32 ELO Scores change based on who you beat/lose to. 7:40 Mathematical Tiering with a Kernel Density estimation. *Ranking all The Trainers of Pokemon Gen-1 Kanto Region* 8:50 10:00 Rocker #2 beat Swimmer #13. Electric beats Water. 10:30 Brock, Pewter City Gym Leader. 11:00 C+ Juggler #5 The Mr. Mime trainer. 11:58 B- Misty, Cerulean City Gym Leader. 12:10 Beauty #10 got a 1,000 Turn Draw to happen using poisonpowder vs absorb. 13:22 Lt. Surge, Vermillion City Gym Leader 14:06 Erika, Celadon City Gym Leader. 14:15 Lorelei, Elite 4 Ice Trainer. _she prioritized rest_ 16:10 The A+ Tier, The Top 13 Trainers. Sabrina, Saffron City Gym Leader 17:07 #4 The Venasaur Champion, #3 The Blastoise Champion, #2 The Charizard Champion 18:00 #1 The Professor Oak Hidden Secret Post-Game, Final Boss. 19:19 The ELO Scores Leveling Off in The Middle.
the "bug" in Lorelei's code is quite interesting actually. I believe, if I remember it correctly, that Pikaspray Yellow made also a video talking about it. Lorelei used to work exactly that way you describe (and coded) prefering to use rest because it supposedly "super effective" against some pokemon. However, how everything works in gen 1 this can lead to a battle that can take forever. I believe that in later versions (the 3DS digital version), they fixed this by giving Lorelei her own personal AI, completely different from the most other AI trainers basic routines.
Wow. The quality information density in this video is top notch. And the editing is so good! How can you have so few views, dude? Subscribed, really hoping this channel blows up! PS: I would love to see something similar for the 2nd gen games!
Kid's nowadays don't want math, they want tier lists! *explains complex math used to generate tier list* XD Fantastic vid, my dude. Hats off to you for doing all the work, we needed to know!
have you considered submitting this as a study to a real journal? it may seem trivial, but there looks like enough data and structure here to be an actual study. either way, this is geniuniely one of the coolest things I've seen from a pokemon related video. thanks so much for this.
Actually, I know it's not intentional by design - but Lorelei is the first member of the Elite Four. It's up to those who wish to challenge the Elite four to defeat her. She is under no obligation to actually defeat her opponents. In Chess at least if you are the Titleholder and you draw against the challenger you retain the title. It is up to the Challengers to defeat her. So her rating is lower but the meta-strategy of drawing all her opponents works out nicely.
One thing you could have done to simulate the AI trainers wasting items is to give the player a stack of pokeballs. Then, when the AI trainer wastes an item, the player would throw a pokeball, wasting the turn as the other trainer blocks it. Great video though!
There's one thing that makes me curious about the Venusaur champion: he actually has less battles to work with than those around him, a total of 438, likely double counting the battle he had against himself. Other trainers also fought themselves, like with Green2 #11's greatest victory, but they have more battles, like 441, 444, or 440 and 458 for the other champions. So they literally had more time to gather elo rating; also, Venusaur champion only lost against himself and Charizard champion, with literally every other battle being a win in his book. I'm not sure if his performance compares to Charizard champion beating the #1, but it should be higher than Blastoise champion under normal circumstances.
This was super entertaining and well thought out! The graphics were especially cool in how they updated what was in the games while staying true to them
okay dude, "Misty Misty", reminds me of that super old pokemon parody vid pret, elo world and tom7, references to the actual ai code, maaaaaaan you're good!
This video is so well researched, explained and displayed - it rivals bachelor thesis I read... great work, that deserves so much more attention. Have a like and a sub! =)
This is brilliant work. I have recently been getting into creators like yourself that really get in deep with their research and work that is the subject of their videos. Your explanation of the rules of not only the actual game and of course how you devised the tournament was just absolutely delightful. It amazes me that there are actual good youtubers who put real love and effort into their videos. I wish you great luck on your future endeavors.
After far too long, the follow-up to this video is finally out. In it, I fix many of the issues with the original tournament and run the whole thing again, except this time with all the Pokémon flattened to level 50. Check it out: ruclips.net/video/247qD1qulSQ/видео.html
Off topic but the misty misty thing activated some dormant neurons about a certain 2006 shitty pokemon animation for me. Love it.
Its been a while, and a lot of work for you, but are you interested in continuing your Research?
I would die to know whos the strongest Hoenn Trainer is. Maybe even whit all the Frontier Brains at Lv 50 included.
Is that a Pokemon spoof reference at 3:30?
PLEASE do this for Gold/Silver/Crystal!
>Lorelei got a bug in the code that made her "better"
>Stills spams rest
Ah yes, the universal constant of Red and Blue
Gen1 gonna Gen1.
BuT PsYCHic>aLl
@@lisavauhti7675 what is my irl name
Now, say my name
I mean, if she is spamming rest then she is doing her job. Her entire position in that team is to lure out high-damage low-pp moves and make you exhaust them as early as possible.
The only issue is that she uses pokemon with some really unfortunate matchups, both against starters (pikachu/venusaur defeats water, and charizard ice) and various other beasts/TMs likely found in your team. Should have had a team of snorlax.
I actually laughed out loud when I noticed Lorelei's draw ratio.
That infamous soft lock... Pokemon Yellow actually added a check that goes "are you Lorelei? if so, don't spam rest just because it's psychic type"
yea lmao, that code is hilarious
lol? really? that#s awesome
I believe one of Lance's Dragon pokemon does it too with Amnesia.
@@Silith13 You might be talking about Lance's Dragonite using Agility
@@zacharyadler4071 it’s the sole reason that the All Terrain Venomoth from Twitch Plays Pokémon was also known as the Dragon Slayer.
Also, I am pretty sure it was Barrier, Lance’s Pokémon have illegal moves at their disposal, namely Dragonite with Barrier.
this is so cool!!!!!!!!!!
im here before the "hi" comments
oh wait thats the bold guy
Wow a comment by Seth that went largely unnoticed!
the bald king
@@a1b3a3c14nbcv Hi.
I love that the trainers with just magikarp actually won a few games, I can only assume they were against trainers who just self destructed immediately so they had magikarp to spare
Its fun of those funny quirks of red/blue. The ai is stupid enough to sometimes lose against lv5 magikarp. Its proven to be possible to train a lv5 magikarp all the way to evolution and beyond in victory road, to the dismay of many attemps to stuck a save into indigo plateu
I always just imagine... What is going on in the minds of those Electrode trainers? "Yes, all of my Pokemon battles are just my Pokemon exploding immediately and me desperately hoping the enemy's Pokemon is fainted so I can be declared the winner". That's such a sad, braindead life.
@@plasmakitten4261 Maybe they're just cruel. Their goal isn't to win, but to inflict as much pain on others as possible.
@@vanbaguette7368 I think that their Electrode's happiness stat would be so low we'd know about it if that were the case
@@plasmakitten4261 According to the game, their happiness stat would be incredibly low. However, according to the Pokédex, Electrode actually *enjoy* exploding.
I am 1000% sure Blaine is only on A tier since he can't use his signature move: super potion on his full health pokemon
I mean... Doesn't he still have roar as a gym leader? Roar didn't have a function in gen 1 trainer battles... So it always failed. But it's still an option to use, so another wasted turn between super potion on full health.
Why _does_ he do that so much, anyway?
@@willofthewind If I recall, Blaine's AI is bugged and doesn't check the amount of health before healing. From a lore standpoint, I like to think that he's gone senile
@@rjante2236 Fs in the chat for Blaine
I was honestly surprised he even made it that high 😂
"But kids don't like boring math! They like tier lists!"
*Proceeds to turn tier lists into math*
Objective tier lists are based on math
*to turn maths into tier lists.
I LOVE MATH!!!!!!!!!!!!!!
NOOOOOOOOOOOOOOOOO!
@@UntarLaManteca most tier lists aren't really objective as much as they try to be.
SMOGON for example base tier lists off of usage. The theory is a really good pokemon is going to be used far more often then a bad one because people are more likely to win battles. Obviously this isn't always the case as some pokemon aren't able to be slapped on any team but have some kind of niche that no other pokemon can do issue is these team archetypes won't be as common so it's tier placement becomes lower despite being perfectly viable in the higher tier. We also have cases of pokemon being more viable in higher tiers while the meta of their own tier is completely inhospitable to them for example shedinja in gen 3 was actually a legitimate threat in Ubers but would be laughed at in OU.
This video's production is insane and it shows. This is an S tier video. It's elo will surely beat out other videos on the yt algorithm
it didnt sadly
@@RohanDaDev I just saw this in the yt algorithm after so many months, so there is how, and this video definitely deserves it
@@VictorownzuA11 I literally got this recommend this too right now
This video is suddenly trending.
@@benign_oracle8875 That's awesome if this video is finally getting the algorithmic recognition it deserves.
This was extremely interesting, I appreciate the ton of research you put into this, thank you!
hey lightsen!
so when is part 4?
@@pokemonherogaming4669 2077
Waiting for Part 4 , lightsen
@@technicalchirag5757 Animation takes time, they're probably doing their best
Would like to point out that channeler #6 beat giovanni with just a lvl 24 ghastly
That is probably on the level of the Haunter who beat Lance.
It might be because the game prioritizes Giovanni's ground-type moves against Gastly, a Ghost/Poison type, but Gastly dodges all those attacks with its Levitate ability. Although I think the game does ignore secondary types (as demonstrated in the Ivysaur vs. Oddish battle, so maybe it's more complicated than that.
@@ryanskotzke3187 but gen 1 doesn't have abilities, so any ground moves would've been super effective.
Was fissure a ground move in gen 1? I just realized I don’t think I’ve ever used it. Maybe Giovanni’s AI thought “fissure ground and OHKO”
Another reason could be that Giovanni’s gen 1 move sets are awful and are filled with normal type attacks.
It might not have been the final Giovanni. You fight him three times and at least the first one basically only uses normal type moves
Lorelei _specifically_ got patched, so she would not have produced this amount of draws with her Pokémon Yellow AI, which keep sher, and _only_ her, from spamming rest.
It would be cool to see this done with Pokemon Yellow with the fixed mechanics
@@TheCOOKIECHEESEMAN I am really curious to see where Sabrina would fit with lvl 50 (!) Kadabra and Alakazam.
@@bartiz12 She'd be Top 10 I bet
She's just gonna be spamming Recover vs every other Psychic type.
"You'd think the Beauty would be sure to win, but... a-Lass".
That got me good for some reason.
Professor Oak taking revenge for the poor squirtle that stayed behind. Truly a touching moment
Fun fact: In Pokemon Yellow, Lorelei's data was altered to prevent her from softlocking you.
Yeah, there's basically code that checks
"Are you Lorelei? If so, don't spam Rest"
And I think it's the funniest code in any video game
17:13 All these years, I never realized they changed the rival's portrait for the final battle
His backpack apparently had a cool jacket inside.
His appearance changes a couple of times throughout the game
@@ssjbread2803 He gets a backpack and does a different pose, but then he's wearing an outfit they brought back in HGSS
@@michaelcarlton1484 Can see why they gave him a Blastoise in the anime: he's got DA DRIP
As a data engineer in real life, I really appreciated the "look at all this fancy mathematics...realize we don't like it...and just fudge the numbers until it feels right." That's basically 70% of the data science projects I see nowadays 👍
That's mad!
@@michaelbread5906 Don't worry, we just say the words "linear regression" and "deep learning" a bunch until management gets scared and approves it.
Science is more of an art than a science...
@@RadkeMaiden At least engineers do something useful. Dedicated mathematicians rarely make useful contributions to science or society, just argue over "different kinds of infinities" and other pointless constructs that have no bearing on our actual, physical reality.
@@Litevaar... Until we discover something those "useless" rules actually apply to.
That's a mathematician's job. Discover and create new, undiscovered (and thus unused) rules that COULD be useful for future generations. Without a mathematical rule that precisely accounts for what you are doing, every other branch of science is stuck and must either create a new rule from 0 or find an old rule that also applies for wathever they are doing.
That's true for every branch science, form rocket launching to statistical analysis. If there's a situation never encountered before, you either search for an elusive theorem that somehow works with what are you doing and that no one has ever noticed before (and, when that happens, tha mathematician that discovered it suddenly becomes popular and important) or you ask to current mathematicians to create a new theorem from 0.
17:53 it's interesting to see that, even in the ai tournament, one of the only two loses prof oak received was against Lance, just like the lore suggests (Lance defeated oak to become champion)
The other loss from Oak is Green3 (Charizard). Presuming the player character would beat Oak in the end as well, he would have two A+ tier protégés. Not bad for a professor.
@@dimitripetrograd9808 Well he does hook you up with an awesome starter. Just imagine starting with like a caterpie or something.
I'm pretty sure Oak renounced his title of Champion and that's why the seat was vacant. I mean, they did send him an email asking him to return.
@@TheMissingLink2 Dang it, you're right!!! Just a few more minutes and could have fought him... if only games worked that way xD
@@dimitripetrograd9808 and theoretically if you pick Bulbasaur you fight green Charizard and oak blastoise meaning the worst starter to pick is Bulbasaur
“I ran 87000 battles for over 4000 hours of battling-“
Dude that’s worth a subscribe right there. Holy crap.
Yeah why is noone talking about the effort went into this video like seriously
i dont think he let 4000 hours of battleing happen in real time. you can speed this up by magnitude of orders
@@martinsz441 oh!! Since it's an AI he can speed up🤔🤔🤔
I think so...
i mean assuming non-stop, 4000 hours is a bit under half a year which fits right into his upload schedule
Really cool video :)
O hi Kosmic
E
Ε
E
hi
So if Blue was the worst trainer in the game at the first battle, that means everytime I lose against him was *I* the worst trainer on the game lol
Yes. Lol
Depends on what Pokemon he had.
@@wewhoareabouttodiesaluteyo9303 Only if you chose Bulbasaur.
@@user-mh9dx7nz2r Only if you managed to lose with Bulbasaur*
@@user-mh9dx7nz2r only if you choose *Charmander*
This video is really well made, lot of effort put in. I really hope it will be picked up by the algorithm soon!
We must all make a sacrifice to the Almighty Gorithm and pray for a bountiful engagement harvest.
It already has!
Welp
Just showed up for me. So maybe now?
It did, this got recommended to me and I'm very happy it was!
The idea of two oddish just endless using absorb on each other forever just makes me laugh.
"How do lesbians have- you know what nevermind"
Boys: Wow, a girls only school! I wonder what goes on in there
Girls: ODDISH VS IVYSAUR FOR ETERNITY
these memes are better than the boys locker room ones
im so upset he didnt let the lass win
Korega Requiem Da
Boys: Probably nothing interesting, wanna play smash?
Hey, thanks for watching! I'm working on a follow-up to this video where I run the tournament again with all the Pokémon level balanced (and some other changes...). If you're interested, be sure to subscribe! (...it'll be a while, though)
@CJ Who are you on about?
That sounds really interesting!
i know this would probably take forever but try G&S when you get time because well the AI should be smarter and i think Lorelai might be a- tier
Pog
@CJ lmao
Gold and silver please. I want to see if Whitney’s miltank really is that difficult.
Yes this would be an interesting thing to see, how much the difficulty fits the curve once the quirks of gen 1 have been ironed out.
I went back and battled her as an adult and she was stupid easy
She's hard when youre a kid because you dont properly level your pokemon or catch and level new pokemon for your team early in the game
Female geodude is great for making her cry in like 5 minutes
I was also thinking about miltank during this. Excited to see what comes of it.
That would definitely take soooo much longer to make haha two regions and so many more trainers
"I've included timestamps in the description, so if a part doesn't interest you, you can skip around."
Oh, I won't be doing that, thank you.
This was amazing. It actually shows that despite the flak gen1 code gets, it was actually fairly well designed to give the player enough of a challenge at every stage of the game to keep things interesting.
I thought something similar, this demonstration really highlights how well balanced and thought out this game is.
Gen 1 is incredibly well made considering it’s limitations. It’s just slightly unfortunate that all the optimizations and compression also made it very easily become a buggy mess.
12:11 If you ever feel useless remember someone with four level 21 Pokémon lost to a level 5 Bulbasaur
Wow, this is amazing
Pero que hace aqui señor gary, ah claro, video de pokemon.
ojo que se nos hace gringo
That “Misty Misty” sound bite unlocked an old memory that I totally forgot existed.
right?? I miss the days of pokemon spoof,,
whats it from?
@@jesusramirezromo2037 it's from pokemon spoof!! ruclips.net/video/GaVklXhNBP4/видео.html
Yeah, I had to pause the video and sit for a minute trying to figure out why it sounded so familiar
literally one of the first videos i ever saw on youtube
The "Misty Misty" at the beginning was a blast from the past
FRUIT SALAD YUMMY YUMMY
The programming effort that went into this is absolutely incredible. Modding ASM based GB games is tough enough as it is, and to then script emulator behavior and save mods as well, AND THEN to run several hundred thousand simulations? THEN APPLY STATISTICS VISUALIZATIONS?!
S S S
T I E R
This video is fantastic! Super interesting stuff
it's spelled "impostor," not "imposter"
Yooo
@@Sir_Isaac_Newton_ both are correct
I would 100% watch this for every single generation
Oh heck yes sign me up!
Maybe with something more linear than ELO the other times around, but either way sign me up!
14:06 Everyone just going to ignore how this one random rocket grunt made it to A-?
It's because he has a Hypno I guess.
Having a physic in gen 1 makes him good.
I do wish he noted what’s the farthest a trainer class maid it and what team they had
@@MikeJohnson-nr4yo there's another at 14:04 without any psychic types.
@@AlphaSquadZero Yeah I counted about 4 rockets in A-
piman: "the AI isn't limited by PP constraints"
Me: same
I'm impressed! As someone who does scientific programming and statistics professionally, I shudder at the thought of how much work it must have been to compose all that, run the simulations, evaluate the data and on top of it all animate it! I know, it's largely a matter of how familiar you are with a code (in this case, that of pokemon), but I can literally feel the frustration of all the failed attempts due to stupid mistakes it would have taken me. Great work!
The Elo system broke down slightly at the top- we do see that the Venusaur rival had the best ratio of the three- but at that point it's largely splitting hairs.
I'm not sure how I felt about the Elo score without knowing the 'order' which battles were fought. in some sense it doesn't matter what the Elo score is
The trainer who fought their worst matches first would have the best score. But that could have been the difference between rival champ and oak fights
I feel like there is some match fixing going on here. Venusaur only lost twice, and at least one of those losses was vs. itself, which means it has at most one loss vs the top 3. The Charizard trainer fought two more matches than Venusaur, and Blastoise fought twenty more than it! Conspiracy theory: did they let Charizard continue until it was above Venusaur? And Blastoise until it was in between?
@@T_Lorentzen I mean, most of the trainers have different numbers of total battles, it's likely he used some program or something to randomly decide the number of fights and who they fought, so some trainers had more fights than others.
@@BillyBob-xx5jr Well he said in the video that he made every trainer fight against every other trainer, plus some more battles on top for good measure, so he definitely did not intend to skip any trainers at the very least. Yet Venusaur Green at 4th place had itself as its worst loss and 3rd place Green as its best victory. Because it only lost twice and one of those losses was itself, this means it somehow never fought either the Prof Oak at 1st or the Green at 2nd. Even besides the fact that a match against itself should definitely not count for scoring, something is definitely pretty screwy with how the matchmaking was done.
The win loss ratios didn’t seem consistent between the trainers that confused me? Like at the top some of them clearly fought more games than others.
Dang. Maybe Oak retired from battling because everyone else was too easy?
No, all Oak wanted to do was "fiddle with his Pokedex" and he did not believe they were for fighting. Agatha talks about him before you battle her in the Elite Four.
Agatha even said back in his young days, he was a great trainer
Go play a Rom Hack called Pokemon Rocket Edition, it is the best fanmade rom hack, and it goes into the story of Prof. Oak
Wow the presentation of the cards was so nice to look at. I can't imagine the work involved in this. Kudos.
The quality is way to high for this video to be only watched by around 1000 people. Hope it takes off.
Holy shit, I was surprised that at the time of me watching there's only 17k views on this vid.
It is now!!
@@rawbinYT I'm surprised there's only 45k views
@@waterdrinker_ I'm surprised there is only 50.5k views
@@TwinGrimm 60k now
You already know you have to do this for Gen 2 now. The people need to know where Youngster and his Top % Rattata would end up
s tier, obviously
How would it treat rematches, though?
@@Mernom As separeted trainers, like he did for this one.
The music in the "how the ai works section" is a chiptune version of the mario 64 file select theme.
Well played.
Wow, this video was absolutely phenomenal. Interesting methodology and explanations, and the visuals were incredible.
High quality work man. Amazing job! Was super intriguing all the way through!
I predict comments about Undertale songs
wow it's Elu :3
oh wow it's the morenatsu ost guy!!
oh wow it's the undertale ost guy!!
gUYs lOoK iTS THe uNdERtaLe oSt gUY
"Professor oak is gonna win" was the first thing I said when I clicked on this video
Proof?
This is the most interesting video I've seen on RUclips so far, answering so many questions I was asking myself as a kid. Like : "How the fuck is Brock a gym leader, if every trainers we encounter on roads are better than him ?"
Now I know he would lose to almost everyone in Kanto.
In the Lore ,Gym Leader have multiple team that they use depending on how many badge the Challenger have .
This man joined youtube on 2006 and is still making relevant videos, what a legend
I am a fellow data scientist and programmer. I just wanted to say that while watching this video I couldnt help but feel respect for the effort, setup, and visual presentation. So cool
The Strongest non impoartant character is Cool Trainer at number 14 using only 3 Pokemon. Executor, Cloyster, and Arcanine.
Huh, that's pretty interesting. That puts him well above many important trainers.
Dat coverage though. A grass/psychic, a water type with mad defence that has an ice move, and a fire type strong physical attacker that knows Take Down.
I've been wondering what his other three would be if he wanted them. assuming he would choose pokemon with two evolutions.
I would assume Electrode for speed and electric, almost guaranteed turn 1 tunderwave.
Rhydon for ground/rock tank and good moveset
without using the same type twice I could really only think of Golbat only for chip damage and confuse ray.
This was a really cool vid
it's spelled "impostor," not "imposter"
@@Sir_Isaac_Newton_ what are you talking about?
@@TheNerdyGinger If I'd have to guess, it is either accidentally posted here and was meant for another comment OR it was a reply to post which has been deleted since.
"so son are you going to build rockets or work on AI for cyborgs?"
no just....just pokemon stuff
This is how TheZZAZZGlitch's videos make me feel.
This is legitimately one of the best produced videos I've ever seen. This should easily have 10 times the amount of views it has right now.
See you guys in a year when this vid has 5 million views
Meet back up, December 16, 2021?
Aye, will jot it down on my calendar
😭
This is seriously something superior to a college dissertation, holy hell
And it only took him a single day to make! He is a genius of unparalleled abilities!
This was so well explained and so well represented with visual data! You have a real talent dude.
Between the editing and visualization, and the actual coding and running the tournament, I can't imagine the amount of work! Really entertaining and cool! 🔥
Would love to see other generations done! Also including legendary battles might be a good idea. Like the Mewtwo battle is rock hard so I’d be curious where it ranks.
That was pretty cool. I almost want to see the lesser trainers specifically ran through the gym bosses / elite 4 in order. Just to see how many would actually gain badges.
holy shit the editing on this video is insane
Oh yeah Misty classic catchphrase "Misty, Misty!"
DUMBLEDORE! Snape, snape...SEVERUS snape.
@@wewhoareabouttodiesaluteyo9303 Ron, Ron, Ron Weasley!
@@stephenrolle2671 Harry potter..harry potter...ooo....harry potter...harry potter...ahhh
Okay but what’s the video that was actually from??? It’s from like 10 years ago but I NEED to know!
@@carsonlove531 look up Pokemon Trashanimation episode 1
Dude that “Misty! Misty!” Brought back memories of that video from like 15 years ago lol
I wonder if giving a flat level to all the trainers would yield some different results.
it would seem more interesting to see the worst and best AI in the game if level is no object.
I was thinking the same thing. It'd also be more interesting if they all had 6 pokemon, using duplicates to fill in missing slots, starting from the front. Rival 1 would have 6 lv50 squirtles. Mistie would have 3 alternating lv50 staryus and 3 lv50 starmies. Agatha would have her lv50 team plus an extra gengar. The later teams would still have an advantage, but it'd be so much more mitigated.
Surely, attack selection is an indication of skill though, and who would decide which attacks they replace? And if they keep the same attacks, they'd stay disadvantaged.
@@DutchDread Well yeah, that would be the key difference. What we're basically arguing for though is an AI, PvP version of these battles, where the only advantages come from typing, movesets, and skill. A guy with 2 voltorbs doesn't lose to a guy with 3 magikarp just because the former randomly chose self-destruct twice.
@@JKBrown-vv1yw Why arbitrarily select "typing, moveset, and skill" as acceptable though?
The question is do you want to see which is the best trainer, or which is the best AI? Because if it's the second, then any disparity in team composition makes your findings mostly worthless.
If you really want to JUST test AI, you should combine every AI with every team in the game and see which does best with the most teams.
But don't basically all of them have the exact same AI, just with differences in type matchups? It would just be a test of type coverage and RNG.
These digital graphics are amazing. I don't even expect visuals this awesome from double digits million channels. The Rolodex scroll of the character profiles was impressive. Well done AF.
came over from mandjtv. thought i'd give the original a watch as well for the hard work put in.
very nice work
I'ma save everyone 20 minutes:
S-Tier: Youngster Joey
Irrelevant-Tier: Everyone else.
Joey was probably near the bottom. I know this is a joke comment, but still.
Youngster Joey's in Gold/Silver/Crystal, not Red/Blue.
@@BioniclesaurKing4t2 So the OC screwed up anyway. Ok.
That ratatta is clearly op....
That moment when you're playing Soul Silver randomizer and Joey comes at you with a Rayquaza.
17:10 So Venusaur Champ Green only lost twice, without draws and his worst loss being.. himself? Looks like the real #1 trainer, ELO is somewhat random.
and with less total battles judging by the w/d/l
Venusaur champion was robbed
You're right. I'd like to see the list updated with W/L ratio rather than ELO
Charizard champ had 2 more losses and had a worse “worst defeat” with an A- loss. If this were a more manual ranking, like how college football ranks teams; Venasuar champ would be higher than the other rival champs
@@holdenfegan1613 from what i can gather venusaur rival lost to charizard rival and oak with blastoise. Charizard rival lost to blastoise rival, oak with venusaur, oak with charizard and loralei. Rival with blastoise lost to all the oak battles, and the rival with venusaur. I dunno where that 5th battle came from, i can't find it because it's lowest loss is oak with venusaur, we can figure out it lost to oak with charizard. It didn't lose to the nonchampion rival fight since that trainers best win is against itself. And we know it won vs rival with charizard. Lost to rival with venusaur, and lost to oak vs blastoise. I'm guessing this other loss is a repeat battle against one of it's losses. The oak with blastoise lost to the rival with charizard and Lance. Oak with venusaur lost to rival with venusaur and oak with blastoise. So the top 5 should be 5th place rival with charizard since it lost to loralei which is way down on the ranking. Since i can only find 4 losses for the blastoise rival, it would take 4th. 3rd would be oak with blastoise since it lost to lance who's further down in the rankings. 2nd would be rival with venusaur since it's lowest loss would now be the 5th place rival with charizard. And first oak with venusaur since it's lowest loss is the now 3rd oak with blastoise. This ranking isn't including draws since we have no information on those and we don't know how many of those draws were rematched with a decisive winner. If we include draws. It goes 5th rival with charizard, 4th rival with blastoise, 3rd oak with venusaur, 2nd oak with blastoise, and 1st rival with venusaur. It would be nice if we could actually see the listings of all the battles and where the extra battles that some trainers have came from and who they drew against
Don't you realise that he played way too less amount of games, I think this was kinda sad for him
0:46 I can never escape pannenkoek2012 music whenever a youtuber explains something
Really fascinating! And I love the presentation. It’s only a matter of time before manual editing is made redundant by programmatically rendering scenes...
Just wondering if bulbasaur family keeps pulling ahead of the other two starters in the lower tiers, because of enemy AI spamming poisonpowder/toxic futilely against bulba
Those would be advantageous matchups for the Charmender as well, but yeah. Not attacking at all is worse than landing non-effective hits.
@@leirbag1595 Attacking charmander with super effective attacks against fire type:
Ground.
0:10 What would happen if we had all the trainers battle each other?
3:00 2 Emulators are kept in sync. 3:27 Save-State Swapping between Emulators.
4:04 EXP gains --> Disabled by hard coding memory values.
5:04 True-Round Robin, (every trainer involved battles every other trainer)
6:00 DRAW Rules if the battle goes 75 turns w/o either side taking damage or 1,000 turns long.
6:32 ELO Scores change based on who you beat/lose to.
7:40 Mathematical Tiering with a Kernel Density estimation.
*Ranking all The Trainers of Pokemon Gen-1 Kanto Region*
8:50
10:00 Rocker #2 beat Swimmer #13. Electric beats Water.
10:30 Brock, Pewter City Gym Leader.
11:00 C+ Juggler #5 The Mr. Mime trainer.
11:58 B- Misty, Cerulean City Gym Leader.
12:10 Beauty #10 got a 1,000 Turn Draw to happen using poisonpowder vs absorb.
13:22 Lt. Surge, Vermillion City Gym Leader
14:06 Erika, Celadon City Gym Leader.
14:15 Lorelei, Elite 4 Ice Trainer. _she prioritized rest_
16:10 The A+ Tier, The Top 13 Trainers.
Sabrina, Saffron City Gym Leader
17:07
#4 The Venasaur Champion,
#3 The Blastoise Champion,
#2 The Charizard Champion
18:00 #1 The Professor Oak Hidden Secret Post-Game, Final Boss.
19:19 The ELO Scores Leveling Off in The Middle.
"When you're hot, you're hot!"
Sweats in pokemon puzzle
Yeah I immediately recognized that. Pokemon Puzzle League had some amazing voice clips "CL-L-L-L-LOYSTER"
Mewtwo PTSD...."I'm going to win"
I thought I was the only one who recognized the voice from the game, nice to see other people who played this game with its awesome voice clips!
Why is RUclips just NOW recommending this to me?
This is amazing work. Definitely worth more than the 1.2K likes. props to you
the "bug" in Lorelei's code is quite interesting actually. I believe, if I remember it correctly, that Pikaspray Yellow made also a video talking about it.
Lorelei used to work exactly that way you describe (and coded) prefering to use rest because it supposedly "super effective" against some pokemon. However, how everything works in gen 1 this can lead to a battle that can take forever. I believe that in later versions (the 3DS digital version), they fixed this by giving Lorelei her own personal AI, completely different from the most other AI trainers basic routines.
incredible how prof. oak #2's worst defeat was against himself
Alakazam is such a good Pokemon that Sabrina outranks half of the Elite Four
In Gen 1, Mewtwo and Alakazam are Gods
Rewatching before I check out your new vid! Love this and can’t wait to see what’s next
Wow. The quality information density in this video is top notch. And the editing is so good! How can you have so few views, dude? Subscribed, really hoping this channel blows up!
PS: I would love to see something similar for the 2nd gen games!
Was killed on the spot by that Pokémon Parody audio of Misty saying her own name. Over an entire decade came crashing down on me in an instant.
"Misty, Misty!" I love the reference to Pokémon Spoof
Ayyeee my boy Rival Fival getting up in that A- tier
So we’re just going to ignore how amazing this editing is?
MandJTV sent me! This video is great and I love the editing in it. So slick and smooth.
The production value for this is absurd, I’m very impressed
Now I'd really like to see how each trainer would fare in a turnament environment, having each Pokemon set to level 50
Kid's nowadays don't want math, they want tier lists! *explains complex math used to generate tier list* XD
Fantastic vid, my dude. Hats off to you for doing all the work, we needed to know!
Surprised there was no mention of Lance's Barrier spamming Dragonite
Considering it's his last pokemon, I assume most trainers never even saw it, and those who could had the potential to take it out.
"Video is split in parts, feel free to skip around."
Oh boy, that must be a long video then!
*Video is 20 minutes*
Oh.
have you considered submitting this as a study to a real journal? it may seem trivial, but there looks like enough data and structure here to be an actual study.
either way, this is geniuniely one of the coolest things I've seen from a pokemon related video. thanks so much for this.
Alternative title: Youngster Joey exposed.
To be fair, he's a gen 2 trainer. So maybe next time.
FBI open up!
Actually, I know it's not intentional by design - but Lorelei is the first member of the Elite Four. It's up to those who wish to challenge the Elite four to defeat her. She is under no obligation to actually defeat her opponents. In Chess at least if you are the Titleholder and you draw against the challenger you retain the title. It is up to the Challengers to defeat her. So her rating is lower but the meta-strategy of drawing all her opponents works out nicely.
It’s a lot of work to make the statement “the enemies get tougher farther into a video game” a 10 minute video
One thing you could have done to simulate the AI trainers wasting items is to give the player a stack of pokeballs. Then, when the AI trainer wastes an item, the player would throw a pokeball, wasting the turn as the other trainer blocks it. Great video though!
When i was a kid, my brother told me if you throw enough pokeballs you can eventually catch them. I wasted a lot of pokedollars.
@@androsp9105 LOL You would have been a good Rocket member.
@@MundaneMuser I always wanted the option to join when the guy on nugget bridge offers.
Fantastic video. If there ends up being a follow up, I'd love to see how each trainer fares in a level-adjusted tournament.
You got your wish!
Could this be done with the later generations? This is really interesting.
There's one thing that makes me curious about the Venusaur champion: he actually has less battles to work with than those around him, a total of 438, likely double counting the battle he had against himself. Other trainers also fought themselves, like with Green2 #11's greatest victory, but they have more battles, like 441, 444, or 440 and 458 for the other champions. So they literally had more time to gather elo rating; also, Venusaur champion only lost against himself and Charizard champion, with literally every other battle being a win in his book. I'm not sure if his performance compares to Charizard champion beating the #1, but it should be higher than Blastoise champion under normal circumstances.
I think that this is one of the most interesting things I've seen someone do with a pokemon game in a long time.
This was super entertaining and well thought out! The graphics were especially cool in how they updated what was in the games while staying true to them
okay dude,
"Misty Misty", reminds me of that super old pokemon parody vid
pret, elo world and tom7, references to the actual ai code, maaaaaaan you're good!
This video is so well researched, explained and displayed - it rivals bachelor thesis I read... great work, that deserves so much more attention. Have a like and a sub! =)
This is brilliant work. I have recently been getting into creators like yourself that really get in deep with their research and work that is the subject of their videos. Your explanation of the rules of not only the actual game and of course how you devised the tournament was just absolutely delightful. It amazes me that there are actual good youtubers who put real love and effort into their videos. I wish you great luck on your future endeavors.