This is my third published article where I analyse the drafting patterns of professional teams using machine learning and data analysis techniques based on my skills I learnt in my degree. This time I’m tackling the drafts of Team Liquid’s very successful run at Starladder I-League StarSeries S3 this last weekend.
Straight into it, this was some seriously hard data to find patterns in. In 11 games liquid played 32 heroes out of a possible 55, with 19 of those only being played once (and 8 only being played twice). So in 11 games Liquid only played 5 heroes more than twice.
This makes my favourite algorithm for this type of analysis, apriori, pretty bad, as it relies on finding common pairs of attributes in a data set to find patterns. However, I still have a fair share of information about Liquid’s drafts to share.
Lets start with this graphic
The first thing I notice in this dataset is what I feel to be a weakness in Liquid’s drafts. Only three heroes had more than one player who played them. (And let’s face it – ‘Rubick can be played by both supports’ isn’t very interesting data) When I analysed Digital Chaos’ drafts at TI5 I was alarmed about how many heroes they could swap across the team which left huge ambiguity in the drafts, making it incredibly difficult for an opponent to predict lanes & game timings. (Remeber that support naga game? – link to my blog on DC if you’re interested in some old stats)
Now, does a team with this much versatility need to rely on cheeky tactics? Obviously not. It’s not just hero pick versatility though. Similar to EG Liquid tend to play a team fight oriented offlaner, however their supports are slightly more mobile than EG’s. Last time I looked at EG’s data, cr1t often played a lane support whereas Zai would be roaming the map and supporting Suma1l and Universe. However, with Liquid, both supports tend to play slightly more aggressive, but equally aren’t afraid to support Matumbaman when they need to.
Who is the carry now?
Also similar to EG there seems to be some question about who theposition one is. According to the average GPMs, Miracle- is actually the position one of team Liquid, but averages are messy, so lets do a count of games to see which player is given farm priority.
No surprises based on the hero picks & the averages, Miracle- does tend to have higher farm priority than Matumbaman, even though Matumbaman has almost half as many games with top farm priorty than Miracle- how come he’s only 20gpm behind? Take a look at the top 5 GPMs from Liquid’s run at Starladder. Although Miracle- comfortably holds the top spot, Matumbaman has massive gpms occupying the next top four spaces (there’s substantial drop off beyond these records too)
Anyway, no one really cares about GPMs that much, as there’s a lot going on in the games that can effect these, so looking at them as pure numbers could go on for hours – which no one wants to read.
Before we start getting into hero picks, there’s still some background on Liquid we need to look into, where’s their comfort zone? What kind of games timing to they aim for?
Now, yes this shouldn’t really be a line graph, but screw you High School statistics, and this is a relatively small data. But. Holy. Moly. Liquid lost every game that went over 45 minutes & won every game shorter and tended to have a strong peak around the 30-35 minute mark. Now, if I know this – Liquid knows this & this is what they will be aiming for in their drafts.
Liquid lost their first game of the tournament with Naga & Mirana as their position 1 & 2, these heroes were not picked again. The other game they lost was an OD/Timber/Brewmaster tri-core lineup that couldn’t take the game late against VGj’s Lifestealer/Sniper + Centaur core. As mentioned earlier Matumbaman & Miracle tend to pick two semi carries, and if the game calls for Matumbaman on Morphling, Miracle- will take a mid-game, mid-lane ember spirit. Equally, Miracle-’s Sniper requires Matumbaman (& MinD_ContRoL) to be an active frontliner with a Lifestealer and Centaur combination (Huh, i wonder where VGj got that idea from) 🤔
I could swear this was my economics teachers’ favourite word back in high school, and I’m going to use it a lot to discuss this combination. Lifestealear, Sniper & Centaur was picked up by the apriori algorithm as being fairly popular throughout starladder, and it makes a lot of sense.
Centaur & Lifestealer have a pretty strong mid game combination of infest bombs, high damage, tower pushing abilities & similar power curves so they often get picked together. Those heroes are so tanky and in your face it allows… well… *6.83 flashback shudders* sniper to peep peeps from the peepers.
The venn diagram to the left shows the relationship between these heroes at the LAN finals of Starladder.
In that one sniper game that didn’t contain a centaur, wings picked sniper then Liquid immediately took centaur from them to stop them getting that combination – which worked out very well for liquid as they already had the other sniper – Lone Druid (who is usually banned).
After further investigation, Bounty Hunter is also drafted in all of the games containing sniper & centaur, possibly the team fight presence of those two heroes (and the team around them) is enough to benefit from Bounty’s track gold allowing the team to snowball hard into pushes. Lifestealer is a good compromise if you don’t have access to a Bounty Hunter as the hero is naturally strong in that stage of the game without too many items anyway. Rage allows him some team fight presence along with Infest bombs, and coupled with Feast and open wounds the hero has the sustainability for pushes post team fight – made even easier with the high damage armlet, echo sabre & desolator build.
Now, Sniper is an excellent trade-off when Lone Druid gets banned (which is pretty much all the time) since 7.00 where talents were added & Lone Druid became a better version of the hero no one wanted to be in the game in the first place (Except maybe Pajkatt). And as we still have 10 centaur games to account for, it’s worth investigating this combination.
Surprsingly, this hero only accounts for 3 more of the Centaur games, and Lone Druid raises more questions with four more games not matching the pattern mentioned earlier, never even appearing in a game with lifestealer. Looking at Matumbaman’s picks above & dotabuff, this appears to be because both of these heroes prefer to have a support with them in lane — and druid can push towers with the bear anyway, right? The other four Druid games may not have a centaur, but they do include similar tanky team fight initiators, such as Tidehunter & Slardar.
I want to move away from this as it’s not really focused on Team Liquid, although they did play this strategy, but I’ll drop a quick fact that 4/7 of those centaur games, currently unaccounted for, contained a Weaver #DesolatorIsAGoodItem.
Although there isn’t that many patterns about actual hero picks, there’s some really consistent information in how teams approach the first phase of picks against Liquid, and one hero that Liquid really don’t want to play against. In 8/11 games, Liquid banned Nyx, always within the first phase of the draft.
Starting with the heroes win rate, Nyx Assassin played 5 games at starladder, and won 4. There’s a little correlation on the heroes’ timing, the only game Nyx lost was a 34 minute game by TNC against OG. Every game Nyx won was longer, spreading between 37 minutes and a 1 hour 18 minute game, which as we know is when Liquid tends to fall off.
It appears that this hero was banned due to his presence in the meta, and Liquid do not want to play against it, in fact, they don’t even play it themselves. It was never banned against them, only by them, and they only played against it once, their first game, which they lost — maybe that’s why they hate it.
Typically teams tend to be pretty consistent with what they have banned against them, more so then their picks themselves, for example, although Liquid’s picks were so wild, with only 5 heroes being played more than twice, in the first phase of the draft alone Liquid only had four heroes banned against them, but banned 10 different heroes – this is why Nyx stands out, as an incredibly consistent ban – they banned it against every team except for TNC (who had a 1-1 score on the hero [Lost to OG but Beat Secret])
On the flipside here’s all the heroes Liquid had against them in the first phase of the draft
One last data drop for the road, something that was picked up by apriori in Liquid’s drafting data
In every game that Liquid played against a Storm Spirit, they picked a Rubick. Storm Spirit is mainly magic damage, so the extra survivability for the team from Rubick’s Null Field is pretty handy against the hero. But the real reason is probably Rubick’s instant stun from telekinesis to stop Storm Ballin’ away. Rubick also has decent ganking potential on the mid lane, able to use fade bolt for that extra damage, and again, telekinesis is useful for those post-level-6 ganks.
In my first ever blog I spoke about the weaknesses of the algorithm I start with to analyse drafting data (apriori), and for smaller tournaments with teams that have insanely diverse hero picks (on the face anyway) the weakness is really prevalent. I believe I’ve shown that Liquid do have a style of dota they prefer, early pushing and aggression, team fight offlaners and having however many 9k players they have now.
Although their picks have different names they share a lot of attributes, mid-game carries, tanky team fighters in the offlane & decent damage roaming supports. It’s at this point I realise that looking at a hero only by it’s name is very naive, as there are a lot of trade offs for heroes that can offer the same thing, and the subtle differences in those picks could reveal a lot more information.
But hey, I’m not a team analyst, and scratching the surface is all I really have time to do 😀 but remember – there’s a lot more to dota than just drafts.
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. (Yes, I’ve had issues with this is the past)