Winter Major AI Predictions
Revenge of the Seeding
Edit: Added groups related visualization. Also updated groups scenarios/simulation is available here.
Edit 2: Day 2 update available here.
Edit 3: Day 3 update available here.
The Rocket League Winter Major in San Diego is about to begin, and I'm about to spoil the entire script, so read at your own risk...
The AI has gone into v1.2 (full changelog) which makes it even more powerful. In the process, I fixed a bug that caused regions not to be assigned at Gamers8. As a result, this change happens to boost the entirety of EU up by like 30 points (which would've made the Fall Major predictions more accurate oops).
I've also created a simulation like last major, and it shows just how much seeding in groups affects results. Some teams are expected to outperform teams that are higher up in the power rankings, so make sure you read to the end.
Table of contents:
- the AI power rankings for all 16 teams
- the top 25 players at the Major
- group simulations
- the odds of each team getting Top 8, 4, etc.
If you need more information on what the numbers in this chart mean, check out the FAQ here.
Let's go over some of the more controversial picks.
Oxygen > Team Liquid???
The AI doesn't care where you finish in a tournament. Liquid had a poor first regional, including a loss to Vertigo. If you don't know that team, then that kinda proves the point. Since then, they have been clawing closer and looking solid. However, Oxygen hasn't really had a single blunder this split, so they've kept ahead by a small margin.
G1 to the top?
I was considering making an entire article out of G1's tough road to the majors, but I figured there wasn't quite enough content unless you want to go over all the numbers.
In short: G1 has had the toughest brackets out of any major contending team in EU. Out of the 7 playoff games they've played, G1 had to play against a team that ended up making the major 7 times. That's right, every single time. Although, they have lost the majority of those, they've got another excuse. 3/4 bo7 losses were against the eventual tournament winners; the remaining one was still against Karmine Corp.
I wanted to see whether they've really had it the toughest, so I decided to calculate the average MMR of opponents at each stage of the tournament. You can find the full spreadsheet here if you're interested. One of the things that I immediately realized after collecting the data, is how G1 has been the 3rd best team in groups/swiss on series and game differential. Additionally, they've managed to do it while playing slightly above average rated opponents. However, it is indeed the playoffs, where they've been the most unfortunate, playing against opponents on average 70 points higher rated than what would be expected.
With some napkin math using the rating system, I was then able to calculate that G1 would've been expected to win about 2.5 series more if they had played opponents with the same average strength as other major qualifying teams. This is the sort of difference that could've seen them go to the first major or just comfortably avoid the tiebreaker this time around.
Overall, adjusting for the fact that later rounds are more difficult, the only team (with a decent sample size) that has had it worse than G1 is Solary. Here are the 5 teams with the highest/lowest round adjusted difficulty:
- SLY (+46)
- G1 (+29)
- FAX (+25)
- SUH (+20)
- BDS (+19)
- OXG (-44)
- KC (-28)
- SNQ (-20)
- MST (-17)
The only argument that speaks against G1 is that they've been unable to take down those eventual tournament winners. Therefore, I would not expect them to win this major either. And that means they are placed exactly where they should be, at the bottom of the pile of top teams.
The AI hates SAM and loves MENA
It doesn't, but it also doesn't blindly assume that a repeat of the last major is the best guess when we have a whole history of data in MENA's favor. First of all, let's dispell the myth that people are propagating about how poorly Falcons have been doing relative to SAM. They've been on a decline since Worlds, true. But were they actually outdone by SAM at Worlds? If you include Furia, absolutely. Unfortunately, that talent has just been removed from the region, so it doesn't matter for this comparison. When evaluating minor region performance, a good way is to only use the matches against major regions because it prevents any influence from how many easy regions they're playing. Taking all matches of The Club against EU/NA teams, they went 1-3 at Worlds; Falcons went 1-2; Secret went 0-2. You may say, that's comparing the #1 seed with the #2 and #3 seed, and that's true. I'm not trying to make the argument that MENA #3 (Vision) could beat SAM #3 (KRÜ). I'm only trying to make the argument that Falcons have outdone the current SAM talent at every event except the recent Fall Major.
At said major, Team Secret went 4-2 while The Club went 0-2 (vs. EU/NA). Before the event, not a single person would've picked Team Secret > The Club. So the question was, is Team Secret just way better than expected, or is The Club way worse? It's simply impossible to know. The best guess you can make for the region (based only on this major) is simply in between: 4-4 against EU/NA as measurement. Falcons have gone 0-3, which is definitely worse, but it's also not unexpected or abysmal. It's only one terrible event, and the current version of the AI actually predicted all recent international Falcon matches correctly (7/7). That means they were never losing as the favorites. It does include a few 49% predictions though, so it's still worrysome to lose that much. They are ranked behind all EU teams accordingly.
Regardless of everything I've said in defense of the current rating. I do think that SAM should probably have 20-30 points more right now. I'm not saying that because I lack confidence in my AI, I do it because I know how it works. Team Exeed have been tearing up the SAM region, taking down every top team at least once. But they've come literally out of nowhere, moreso than maybe any team before them. Rating systems are relative measures of skill; when one team goes up, another team goes down. In the case of Exeed that change is very big because they started at a very low rating. That means that the AI thinks all the other SAM teams are now worse compared to regions who haven't played against Exeed. But which one is more likely? One team getting better, or all other teams getting worse? Until international play recalibrates the region rating, it will negatively affect the teams at LAN. The points margin likely doesn't actually change any positions in the rating, but it makes it closer.
I think this list is actually very accurate with a few odd picks. It's certainly better than 4/5 unspecified analysts' lists. Let's see how they actually show up at the major though. LAN buffs are not included.
Atomik > Atomic
If you want to learn more about how the player rating works, I've got a full article on the topic. It's stats based and Atomik is essentially a slightly worse Firstkiller in that regard. He's always at least solid, which is very needed to stay consistently at the top of the rankings. He's also helped by the fact that the AI rates G1 generally higher than most analysts seem to.
ExoTiiK > itachi
This is just the perfect example for illustrating why consistency is important. In the first regional you could split hairs between who should be MVP between vatira and itachi. In the second regional itachi looked like the best player in Europe. He was top 10 and ahead of ExoTiiK. A lot of people took notice and started ranking him very highly. But what happened in the third regional? He completely vanished on the stat sheet. Stats aren't everything, but then why was he able to do it the previous regionals? Whatever the reason for that may be, the AI doesn't care for it and punishes it massively. Hence, consistency is key.
Sad > kv1
Did I say LAN buffs are not included? Sad had an absolutely amazing Fall Major. Since he can only lose rating when losing series, he's been able to keep a lot of that rating. kv1 has been better in the online split, but there simply haven't been enough series against high quality teams to let him overtake Sad.
AI Simulated Groups
AI Simulated Major
|Name||Top 1||Top 2||Top 4||Top 8||Top 12||E(Points)||E($)|
|FaZe Clan||22.2%||36.5%||57.9%||90.5%||95.8%||24.6||$ 41.1k|
|Karmine Corp||18.3%||32.2%||52.4%||84.1%||96.7%||23.1||$ 37.1k|
|Team Secret||0.4%||1.9%||9.4%||41.7%||92.8%||13.1||$ 12.0k|
|Gen.G Esports||13.6%||26.0%||48.9%||79.4%||95.4%||21.7||$ 32.5k|
|Team Liquid||10.6%||22.7%||46.8%||88.7%||99.2%||21.7||$ 30.4k|
|Oxygen Esports||12.8%||24.9%||47.4%||77.9%||95.1%||21.3||$ 31.5k|
|G2 Esports||8.0%||18.4%||38.7%||77.9%||90.2%||19.5||$ 25.8k|
|Team Vitality||1.2%||4.3%||13.7%||39.4%||80.7%||13.1||$ 13.0k|
|Complexity Gaming||7.3%||16.6%||36.9%||69.4%||92.6%||18.7||$ 24.6k|
|Ground Zero Gaming||0.0%||0.1%||1.3%||13.3%||71.6%||9.8||$ 7.9k|
|KRÜ Esports||0.0%||0.3%||1.9%||13.4%||27.7%||8.1||$ 5.5k|
|Team Falcons||0.7%||2.8%||10.4%||32.6%||74.8%||12.1||$ 11.2k|
|Gaimin Gladiators||0.0%||0.0%||0.2%||3.8%||36.4%||7.7||$ 5.3k|
This table is the result of 1,000,000 tournaments simulated, each with 40 series. For a total of 35 million series sampled from the AI probabilities, including game differential.
As you can see, Team Secret is actually expected to outdo Falcons in terms of average points. Generally a lot of results are because of the "Group of Life" with Team Liquid, Team Secret, Ground Zero Gaming, and Gaimin Gladiators. Aside from the group participants, 2nd and 3rd from group A get to play against them, which boosts FaZe/G2/G1/KRÜ's chances of making it to top 8.
The odds for winning the tournament by region are:
NA - 51.2%, EU - 47.8%, MENA - 0.7%, SAM - 0.4%, OCE - 0.01%, APAC - 4 in a million.
The expected points per team are:
NA - 18.7, EU - 19.4, MENA - 12.1, SAM - 10.6, OCE - 8.4, APAC - 7.7.
The expected prize money per team is:
NA - $26k, EU - $27k, MENA - $11k, SAM - $9k, OCE - $6k, APAC - $5k.