A few days ago I posted some analysis of Wing Gamings’ draft from The International, where I used a rule mining algorithm to pull interesting stats as an exploratory project into applying data analysis techniques to Dota 2. You can read this blog here if you missed it. Today, as requested, I’m going to be doing a similar sort of thing, but with the second place team at this years TI, Digital Chaos.
Wings were an ‘unpredictable’ team, and what we’re looking for in Digital Choas is something a little different. I’m expecting to be able to make sense of the data a bit easier for DC. Their drafts weren’t as crazy as wings, but there is a few interesting stories. In the twilight games of TI4, most agree that VG perfected the meta, but NewBee knew how to counter it. In this article I’m going to convince you that TI6 did have a meta-game, and not only did DC know how to play it, but they knew how to play against it, and that’s why it took a meta-breaking team to bring them down.
A popular trend in Dota is picking heroes that can play multiple roles fairly early in the draft. DC did quite a bit of this in TI. The Graph below shows a count of the heroes that DC picked first and second in TI.
Simply, it shows that ~55% of DC’s first picks were ‘versatile’ heroes. There’s some inflation here from Mirana, as she was a pretty good hero this international, and I’ve counted Keeper of the Light as versatile, even though only DC’s support players played him – which I’d still argue is counts.
In the Wings’ draft analysis I covered the pick order of the different players of the team. I noticed a trend whereby their hard support player’s hero was always picked within the first 3 when the team had second pick in the draft. This is also Echoed in DC’s data, shown below.
Saksa’s hero was always picked within the first 3 heroes (remeber the first cluster accounts for two picks that are essentially the same). In the Wings’ data, Shadow’s hero tended to be picked second to last, or earlier, whereas MiSeRy often left Resolut1on & W33’s picks until a little later in the draft.
New trends emerge when we observe the games when DC had first pick.
Once again, Resolut1on’s hero is picked last, or second to last, with 0 picks in the first 3 stages. The standout carry of this years international was Drow Ranger. Resolut1on, or DC in general never actually played this hero, which is probably why they never needed to pick a carry early. (It was banned against them once however – by Wings [do ur research guys,🙄]) W33’s picks are shifted massively towards first pick in this graph – this is completely due to Mirana, all of those first picks are the Princess of the Moon.
Moo’s hero tended to be picked second [Beastmaster/Timbersaw/Batrider] (Team fight control heroes?), and MiSeRy tending to draft himself a hero around the first pick in second phase[Kunkka/Riki/Bounty Hunter/Sand King] (Roamers?).
Saksa’s picks are spread out a little more too, but on inspection you can see that Saksa’s hero tends to be picked earlier in the phases his hero is to be drafted. All of Saksa’s first picks are Shadow Demon or Oracle (3 times each), or Io (once – early in the tournament) the first two being heroes that are meta enough that they don’t show too much about the draft… or do they?
Okay, onto some real rule mining and hero match-ups. Shadow Demon was incredibly meta around Digital Chaos. Only untouched in three games of their international run. They played it 7 times, and played against it 8 – only banning it themselves twice (Both times against Fnatic) and it was banned against them a whopping 11 times. This begins the story of how DC played the meta. They were not afraid to play against shadow demon, they beat the hero 5/7 games they played against it, and they loved playing it, winning 6/7 shadow demon games.
Shadow Demon vs Batrider
I’ve always believed that Shadow Demon was a Batrider counter. Batrider comes in for a lasso, Demon guy puts his friend in a bubble, purges the bat and knocks him out of the sky. Amazingly, in the 6 games DC won with Shadow Demon, they actually banned Batrider in 5 of them. Three before the Shadow Demon Pick, and two after. In the game he wasn’t banned, Batrider was picked before.
There’s a few possible explanations for this – the three picks before could simply be that it’s TI and Batrider always seems to appear in the meta around this time of year, but banning Batrider after Shadow Demon raises question marks in that theory.
According to Dotabuff’s advantage algorithm, Shadow Demon has a 0.99% disadvantage against Batrider. It’s not by any means a huge number, but Batrider is apparently the 28th best hero to pick against Shadow Demon, which just sits him in the first quartile of counters. Discoveries like this reinforce that Dota 2 is an incredibly complex game. Just because Shadow Demon can disrupt Batriders initiation, it doesn’t mean the hero is a counter, there’s a whole game to play around that disruption that happens once or twice.
In the Shadow Demon vs Batrider matchups in DC’s TI run. DC’s Batrider beat eHome’s Shadow Demon in the first game, but DC’s Shadow Demon also beat LGD’s Batrider. Regardless of who counters who, all this really shows that DC could deal with both sides of the spectrum.
The table to the left shows the top 10 heroes sorted by combined pick & ban rates in all the games that DC played at TI. This sort of shows the meta game that DC experienced, and these 10 heroes make up for almost 40% of hero picks & bans in Digital Chaos games.
Kunkka was played 6 times by DC. They only lost with the hero once, but it was the only game they played Kunkka without Shadow Demon. Conversely, DC beat Shadow Demon + Kunkka 4 times, but only beat a solo Kunkka 1/3 times. [beat Fnatic – Lost to VG.R & Wings]
DC played against Elder Titan 6 times. They beat him 5 times, and in four of those games they had a Shadow Demon. In the game they lost Shadow Demon was untouched, (This was the Jakiro game vs EG in the Semi-finals)
So what’s the secret to beating Kunkka + Shadow Demon that DC figured out? There is actually a stat here. 4/5 of the times the enemy picked the duo, DC banned Luna – the one time the trio got through, DC did actually win the game anyway. DC beat Luna two more times, once when she was played with just Shadow Demon, no Kunkka, and once when she was played without Kunkka or Shadow Demon.
DC played Luna 3 times. All three times with Shadow Demon, and twice with Kunkka, winning every game, but it was only banned against them twice. (Fnatic & LGD)
The data suggests that Winter Wyvern was sometimes seen as a replacement Shadow Demon for Digital Chaos. She was picked 4 times, and in all of those games Shadow Demon was banned. They won 3/4 of these games.
There were three points that stood out somewhat to apriori in the data set. The link between DC picking Oracle, Morphling, and banning Io. The Venn Diagram to the left shows the relationship. Io was banned a disproportionate amount of times compared to the other two heroes being picked, but what’s most interesting is the middle section, and the one above it. Oracle and Morphling were picked together three times, but only when Digital Choas had banned Io, however, when picked alone, there was picked the same frequencey irregardless of wether or not Io was banned. This data technically suggests that DC believed Io was a counter to an Oracle & Morphling pair, but not the heroes on their own – which is super crazy, but to be honest – it is probably just a quirk in the data. As last time, the floor is open to theories regarding this statistic. (There was only one game when DC’s opponent banned Io – Oracle & Morphling were also banned DC vs LGD)
Every time the enemy picked Juggernaut, DC picked Mirana. The base statistics seems to hint that there was some sort of matchup on the works here. In the table to the right I’ve digested the heroes’ pick orders, and whether or not the Mirana (DC in this case) won the game. There could be a few things happening here, but I won’t be investigating them fully. Mirana is a versatile pick, if juggernaut is the counter to her as a core, is it the counter to her as a support, or vice versa? Was she seen as countered in the first game, then attempted in the next two? Or finally, is this just a product of DC picking Mirana in so many games (13), that coincidentally, she happens to appear in all of the games against Juggernaut?
So, what have we learnt about DC in this brief overview of their TI6 drafts? Referring back to combinations like Shadow Demon & Kunkka, it appears that not only did they know how to play them, but they also knew how to play against them, the understanding of the Luna combination also appeared extremely important in their games at the International. Their knowledge of versatile heroes definitely helped them throughout the tornument. I still remeber literally jumping out of my seat in that game where Naga Siren was actually a support, and the Venge was the carry, and the combination with Razor’s early push was incredibly clever. DC appeared to use Shadow Demon as an Elder Titan counter 4/6 games, and only picked Morphling when Titan was banned, or the supports had already been picked.
This last section of the blog is unrelated to Digital Chaos, and is just a response to some general questions & feedback I was given from my last blog.
I posted my first Blog on Saturday night, and the response was really positive, so thanks for that. I’ve tried a few different methods of showing my data and statistics this time, in an attempt to spruce up the blog based on some feedback I was given, please let me know if it’s working. I also deleted everything I wrote about the pick order mining I gathered from DC’s data. I didn’t like how it was presented in the last blog post, referring to different picks in the drafting phase is such a mouthful that I’m not sure it’s actually worth the bother, especially as the stats were basically just dumb stories. Maybe in the future I’ll come up with a better way to talk about this data – but I actually don’t think it’s suited to a long form blog, and I quite like the simple graphs showing it to be honest.
I got a few messages requesting source code or just general advice for data science students. First, good on you for choosing the best field in computer science, rite? Second, the code for this project isn’t up to my standard. I currently work between 4 & 6 days a week, and most of the programming is spread out haphazardly over my free time. I’m currently looking to change my job (this blog started to expand my portfolio), so hopefully I can allocate more time to this if it continues to grow. As for advice, I’m more than happy to help you out, and I’m currently working on a little page of ‘things I wish I knew’ when I started my final year project, I’ll try to get that posted to my respondents some time this week.
Finally, I was told by my ‘manager’ that I need to build an online profile from this. I don’t currently use twitter, but I’ve had the account since I was like, 12 or something, and the majority of people who follow me I’m sure are bots. So feel free to drop me a follow on there, I’ll start posting some micro-analytics [or half-arsed statistics, to give them their real name] to see if I can scare the bots that have been following me for years. @omgtrafficlight
Short disclaimer. I am more than happy for my articles to be posted elsewhere if they are translated to share with readers I can’t access, but please – ask for my permission first, or at least notify me. My last article was translated and uploaded here and translating the comments gave me feedback from another side of internet culture, which was great, but I like to know about these kinda things when I’ve poured literally hours into these.