Content of the article: "Reverse Engineering of Team Secret’s Game. Part 2. (Translated from Russian)"
Hello! I present to you the second part of a series of articles analyzing Team Secret’s games. At the last ESL Germany tournament, Secret did not take first place, but my interest in their game has not faded, and as I hope, neither has yours. You liked the last part, so for the second part I did a lot more work and prepared a lot of insights from Secret’s games for you. I'll talk about smokes, warding, game economics, objectives. I'll show you some of the economic imbalance that is present in the game between Dire and Radiant. This economic imbalance is present both in the intra-team economy and in the in-game economy. And I will also show that some teams, for example Alliance, do not use smoke very effectively.
I would also like to remind you that everything that I write should not be taken as the truth; the game is quite complex and some of the actions of the players are quite difficult to quantify. I would also like to remind you that I have 3k MMR. So, my reasoning is most likely wrong.
All the conclusions that I will draw below are based on the games from the last three tournaments:
AMD SAPPHIRE OGA DOTA PIT EU / CIS
ESL One Germany 2020 Online powered by Intel
Team Secret played 44 matches in these three tournaments. I also want to note that general statistics were based on all matches of these tournaments, except for the matches of Team Secret. This is done so that the statistics are not biased, because Secret have an extremely high winrate.
Throughout this article, the central idea elucidated is that Secret’s game is not identical on Dire/Radiant, their Dire plan is different from their Radiant one. If you think that's the case for every team, then no. From the top teams (I checked only Liquid and Alliance), nobody else plays like that.
Smoke of Deceit
Watching the livestream while the team uses the smoke, have you wondered why they are doing this? How effective is it to use smoke now? Is it better to use a smoke as two or three? And in general, how should one assess the effectiveness of smoke usage? Usually, commentators try to answer this based solely on subjective feelings. In this part, I will answer these questions based on data. Let's start with who is on the team and at what minute they buy Smoke.
The graph shows that YapzOr usually buys the Smokes until 10 minutes. Next, let's look at what minute and how many Secret players use the smoke.
Secret often uses Smoke at 13 minutes and just for Dire and usually by two or three players. As Radiant, Secret doesn't have any obvious patterns, except for the 5-man smokes at the start. Consider these two cases. Let's see the position of the players at the time of using the smoke 13 minutes into the game.
We see that Secret use smoke mainly for their T1 and almost always for two players. Why for two? Why in the mid lane? And why more often for Dire? Let's figure it out further. The position of the Smoke application at the beginning of the game is also interesting.
We see an interesting moment Secret uses Smoke from the start of the game near T4 on the middle lane, and not from the base. If you think everyone is doing this, then no. Alliance also likes to use Smoke at the beginning of the game, but they break it right at the base. The purpose of this Smoke is usually to place deep defensive wards.
How should we evaluate how effective the Smoke was? For a start, of course, it would be good to understand the purpose of its use. However, it is certainly impossible to do this without direct contact with the players. But let's suppose that Smoke's task is to create an advantage for the team. The advantage in gold and in experience is quite a convenient tool for tracking changes in the situation on the map after applying the Smoke. But the question arises, what time window should be taken to assess the advantage? This is, in principle, not a simple question, and requires a separate study. For this article, I took a window of 120 seconds.
Thus, my interpretation of the effectiveness of the Smoke is the sum of the change from the total team cost in terms of gold or experience in the next 120 seconds after the Smoke is applied.
For example, at the 13th minute, Secret leads by 4,000 in gold. They use Smoke and at the 14th minute their advantage in gold has increased by a thousand, and is 5 thousand. On the 15th minute they lost a little, and their advantage was 4.5 thousand. Thus, the Smoke efficiency is (5000 – 4000) + (4000 – 4500) = 500.
Another option for assessing the effectiveness of Smoke is an assessment using the economic ROI ratio, in fact, this ratio shows the efficiency of investments. For example, an investment in Smoke is made up of the cost of Smoke + gold that you will lose when you move in Smoke (since you will stop farming) + experience that you will lose when you move in Smoke. Profit is made up of the gold and experience gained. By dividing Profit by Investment, you get the ROI ratio. However, the article turned out to be too long, so I won't calculate it, maybe some other time. For this article, I will choose a simpler interpretation.
Based on the example above, it turns out that the effectiveness of the smoke is two numbers, the average difference in gold and experience in the next 120 seconds after use. Of course, you can combine these two numbers into one, taking the average for example, but let's leave them as two numbers, for clarity.
According to the graph of Smoke efficiency, you can see that using Smoke as two for Team Secret is much more profitable than as three up to the 20-minute mark. Further, everything strongly depends on the situation on the map and, on average, the use of Smoke on more than two players is justified. Similarly, it can be seen that Secret does not use Smoke for one single person.
Remember above I wrote that Alliance is better off not using Smoke. According to this metric, their effectiveness for Smoke up to the 20th minute is about zero in terms of gold and experience.
This also raises the question of how should one evaluate the effectiveness of wards? One the one hand, if an enemy appears on the ward, is the ward useful? On the other hand, if no one appears on the ward, then this is also information! Here I could not independently invent a quality metric for wards. Therefore, here I will just show you where Secret puts wards and with some subjectivity I will try to imagine why exactly there. Here I will slightly explain the yellow dots on the map, these are the places where the Observer Wards were planted based on 44 matches, broken down into 5-minute intervals. Note!!! This does not count where the wards were already placed (previous interval)! The lines are Kernel Density Estimation (KDE), showing the distribution density of the ward setup, ranging from dark red to yellow, where yellow is the center of the distribution.
It is worth paying attention to the following points.
On the interval (-2, 0) minutes, when playing as Radiant, Secret puts wards more often to protect the mid lane from ganks. This is due to the fact that in the game for the Radiant Nisha often plays more greedily and his networth is more important for the team than when playing as Dire. There are fewer defensive wards on the safelane, and more aggressive wards on the offlane. Also, for Radiant, there are places for setting deep wards, on the opponent's territory; I think that their goal is also for defense, but, interestingly, they were mainly placed through a Smoke of 5 people at the beginning of the game.
The next segments are (0, 5) and (5, 10) minutes. An extremely interesting picture, you can clearly see how attacking wards are practically not put behind Radiant by Secret, only defensive wards, and in defense of the mid lane. But, for Dire, the picture is mirrored, there are no defensive wards at all, only attacking ones, basically these wards are in the offlane area, and they are set by YapzOr for ganking mid.
In CIS teams and teams of lower rank it is accepted that only the position 5 buys and places wards. While in Secret, at the beginning of the game, it is YapzOr who actively wards.
Some information on the style of play can also be gleaned from the timing of destroying towers. Below are the timings for the destruction of towers by Secret.
The timings show that playing for Dire, Secret team demolishes towers faster for Dire than for Radiant by an average of 6 minutes. If you think it's okay and all teams play the same way, then no, not all. For example, on average, Alliance demolishes towers in about the same time on both sides.
Judging by the timing of the demolition of towers and wards, you can clearly see that Secret plays faster on Dire.
Depending on the side, Team Secret also changes how they farm the most valuable neutrals in the game. Below is a graph that shows how much, on average, a player gets by farming their triangle.
Pay attention to Nisha’s farm. Playing as Radiant, Nisha, on average, farms 400 gold from 10 to 15 minutes from his triangle. For Dire, in 10-15 minutes, Nisha farms an average of 320 gold. The difference is 80 gold, which is transferred to MATUMBAMAN’s farm. This implies that Nisha farms less as Dire, which means he moves more around the map. For comparison, here's the triangle farm from Alliance and Liquid.
Pay attention to the difference. Alliance plays exclusively for one person, Nikobaby, moreover, for some reason s4 as position 3 farms more neutrals than the midlaner.
You often hear from analysts that Secret is good at distributing farm. Have you ever wondered how efficient this farm distribution is? Do I need to do this at all? Below I will try to answer a number of these questions.
It's funny, but it turns out that the distribution of farming by team has a different winrate for Dire and for Radiant, and this dependence also changes over time.
First, how do you express the distribution of farm in a team? To do this, you can calculate the networth standard deviation within the team, at every minute of the match. The standard deviation is essentially the average deviation from the average, I will not give the formula, you can easily google what it is. Next, let's see what the average net worth deviation can generally be within the team.
This distribution is bimodal. On the graph we see that the average net worth deviation in the team from 0 to 7000 has approximately the same rate of occurrence for both Dire and Radiant, and not otherwise. This graph shows that, on average, there are more games for Radiant, where, for example, carry has a lot, relative to supports. Now it is possible to estimate the net worth deviation in winrate on the intervals including two distribution modes (1000, 6000) and (6000, 10000).
The graph below shows the relationship between the distribution of net worth in a team over time and winrate. These statistics were built from all matches of the three tournaments, the list of which I gave above. Secret’s games did not fall into these statistics, as I wrote above.
The point on the graph is the mean of the standard deviation within the team for a given minute. Note that there is a non-linear dependence here that is mirrored by the Dire / Radiant side. You can interpret it like this:
Playing Dire, before the 25th minute, it is better that there is a player who has a lot of farm relative to the rest of the team, after the 25th minute the relationship is reversed. That is, it is ideal that after the 25th minute that the players who did not receive farm get it. The situation is reversed for the Radiant. In the late game, the situation is mirrored.
Example: When playing for Dire, supports and mid lane are better off moving more around the map, creating space and farm for the carry player; this farm is realized by the player in about 25 minutes, helping other players to get their farm, which they did not receive until the 25th minute. And so on until 40 minutes, in the deep lategame, the carry should have relatively farmed teammates. The opposite is true for the Radiant.
As you can see, this style, all other things being equal, adds ~1.5% to the probability of winning in the mid game and ~10% in the late game. It seems that this is not a lot, but this is only one aspect of the game, and if the amount of prize money is several million, it can and should be investigated. Now take another look at the gold from farming their triangle by Team Secret, how they use Smoke and how they ward. Playing as Dire, Nisha in the midgame farms less than on Radiant, and, as you saw earlier, in the chapter about Smoke, he often moves under Smoke, creating space for MATUMBAMAN.
Inflation and deflation advantages
Now it's time to answer the question as to why Secret have different styles of play for Dire and Radiant. And why when playing against them even during the draft, you need to understand that for different sides they have different timings, different warding, different Smokes, different farming and, accordingly, picks and builds that will be different. This is partly due to the imbalance in the intra-team economy. But that's not all.
And as I wrote above, there is an economic imbalance in the game, which also exists in the distribution of farm within the team. Start with a thought experiment. Think, for example, do you think a 3k gold advantage is a lot? Of course, each of you will answer, depending on what minute, and yes, this is true. And so, you already understand that the value of a unit of gold advantage changes over time. But then try to answer the following question, "Is three thousand gold advantage in the 5th minute better than in the 10th or not?" This is what you really need to think about! And I will tell you in advance that the advantage of 3k at 10 minutes has a winrate higher than at the 5th. Because in game it is important that you have not only the gold but also the experience in order to realize it. The same idea is expressed by NS. Okay, now try to answer, do Dire and Radiant have the same winrate with a 3k advantage at 10 minutes? I will tell you, no! This is another economic imbalance in the game.
In this part I will try to answer the question of how much a unit of gold is worth in winrate for Dire and for Radiant, and how it changes over time. To do this, take the matches of tier 1 teams <'Natus Vincere', 'Alliance', 'Ninjas in Pajamas',' ViKin.gg ',' OG ',' Nigma ',' Team Liquid ',' VP.Prodigy ',' mudgolems '' Evil Geniuses'> and their games in the last three tournaments. Let's calculate which winrate has the advantage in gold and experience at every minute of the game. Please note there is no Team Secret in the list of teams! This is because they have an extremely high winrate and are an outlier for this statistic. I already wrote about this above.
First, you need to understand what kind of advantage the parties have on average. To do this, you can build the distribution of the advantage in the game.
Already at this point, it seems strange why Radiant players on average have an advantage of up to 7500 gold, and up to 5000 experience more often than Dire players. The interquartile range (IQR is the difference between 75 and 25 quantile) for the distribution of gold will be equal to 3738, for experiment 3942. For example, take the advantage range from 2000 to 4000 (2000, 4000). The IQR value falls within this range.
The graph below shows the value of the advantage (2000, 4000) of gold and experience in winrate over time.
The line shows the graph of the mean, and the bars show the confidence interval of 85%. A confidence interval of 85% is an interval in which a random variable falls with a probability of 85%.
It can be seen that the advantage in gold and in experience up to about 15 minutes results in a greater winrate for Dire than for Radiant. You can also see how the winrate time advantage decreases from about 12 to 20 minutes, and how it grows from zero to 12 minutes. After 25 minutes, the confidence interval becomes too large, and the statistics for the most part in (2000, 4000) are not so obvious, this is actually quite understandable in the late stage of the game, as such an advantage does not mean anything.
You can build a linear model along the segments (0, 10>, <10, 20) to understand how quickly the cost of the advantage grows and is lost.
Notice how the value of the advantage in (2000, 4000) varies differently over time, depending on the side. The point on the graph is the average advantage in the game for a given minute. For Dire it is red, for Radiant it is green. If, for example, playing as Dire, the advantage of ~3000 gold at the 5th minute has a winrate of 65%, then the same advantage for the same side at the 10th minute has a winrate of 72%. The difference is 7%, which is hard to ignore. You can see how quickly the Dire advantage over Radiant is depreciating. If you build a polynomial regression, you can see the very peak when the cost of an advantage stops growing.
The dotted lines mark the peaks of the graph. You can see that the gold advantage value peaks at 14 minutes with Dire and 19 minutes with Radiant. For experience, these peaks are at 12 and 16 minutes. If we combine the timing by gold and by experience, it turns out that the statistically best time for an advantage of 2000-4000 gold/xp playing for Dire is about 13 minutes, and for Radiant 18 minutes. The game Dota 2 contains economic strategy, and it is necessary to have an advantage in this game when it is statistically most profitable (of course, provided that the team is able to realize this advantage). Now take another look at Secret's Smoke timings for Dire. That is why Secret's play for Dire is slightly faster in midgame than for Radiant! I am not suggesting, for example, that a 5th minute advantage is bad, I am just arguing that statistically this is not the optimal time.
Does Team Secret play that style because they may also see some kind of imbalance in the game between Dire and Radiant? Hard to tell. Maybe yes, maybe they feel it without all these statistics. But the fact that they play different Dota for Radiant and Dire is a fact! It seems to me that part of this style was borrowed from OG’s game after TI9. At The International 2019, playing for Dire, OG danced at the enemy fountain already at the 25th minute, but they played more greedily as the Radiant. Check out the games against LGD, Newbee and even the grand finals, there are even more greedy picks just for the Radiant! For some reason, no one noted this point; apparently, no one except Puppey.
Does such preparation allow you to win? An interesting point. It seems to me that in the CIS, especially, there is a belief that the winner is the one who presses their buttons better, respectively, and the best player is the one with the higher MMR. But I am convinced that, starting from a certain level, provided that the teams are in shape, mechanical skill is approximately the same for everyone, and it’s precisely strategy that decides the game. Moreover, the strategy must have some abstraction, so that it cannot be destroyed by two bans.
PS: I spent a really long time writing all this, please like and leave a comment, even if you think it all is complete nonsense. As with the previous part, if this one receives positive feedback, then the next part will be about the Holy of Holies of Team Secret team – their drafts, and I will also try to capture the same builds. Maybe I'll write a neural network that will learn from Secret drafts. Thanks for reading!
- Team Ares maphacking on Moon Studio Asian League?
- Relic platinum score based on warframe.market prices
- Three advice to help you get better at Dota 2
More about Dota 2Post: "Reverse Engineering of Team Secret’s Game. Part 2. (Translated from Russian)" specifically for the game Dota 2. Other useful information about this game:
- Carry players, y’all need to get your shit together.
- I AM SO SICK OF SMURFS (australia)
- Goodbye Dota
- Advanced Supporting : Pull Camp Power Ratings
- Mid matchup being a luck game? Can’t secure last pick? Just pick Lina mid!
- The truth (?) of pseudo RNG
- MKB proc chance is lower than 75% – Part 2 – In-game testing
- Turbo Mode Meta Heroes?
- Struggling to keep up with live coaching? Why not try written analysis coaching to raise your MMR?
- Upcoming mouse/squirrel hero
Top 10 NEW Games of November 2020
November 2020 is filled with tons of games to pay attention to thanks to the upcoming launch of PS5 /Xbox Series X and beyond. Here's a roundup of the big ones.
Top 10 Best Video Games of 2020 (So Far)
In times of uncertainty, video games allow us to escape from the stress of the real world. For this list, we’ll be looking at some of the best games released in the first half of 2020.