Working out the bugs. launched an automatic match analysis launched an automatic match analysis

Admit it, you probably found yourself thinking after the end of a Dota 2 match at least once: “Damn, I had really bad luck with the team, I did everything right, but lost again because of stupid actions of my allies!”. I In DotA2, as well as in poker, it might be very difficult to recognize and realize one’s mistakes. The only difference is in poker the analysis of one’s action complicates the understanding of probabilities, while in DotA2 there is a huge selection of opportunities, tactical moves and available actions. The same action — whether it is purchase of artifacts, upping of a talent or navigating the map — might turn out to be very useful in one situation and a serious mistake in another.

A coach of one’s own. How our mistake analysis works.

Any metrics in the game correlate with winrate. Selection of a specific hero, buying a particular item, skill upgrading, moving to line, kills, assists, damage and damage dealt to towers — value of each of these metrics can raise or lower your chance of winning. In other words, our goal is to ‘train’ a model on a large number of matches (several million) to ‘understand’ how individual metrics affect probability of winning. After that we can import a specific match into the model, and it shall assess which way exactly specific metrics have influenced your probability of winning and what their absolute value is.

Every day we fill up our database of matches, engaging random matches from all regions and ranks — to make the model train on fresh data and adapt to patches and changes of meta. After each match you played, gets its results and analyzes the replay, comparing your performance to the average and top-end players’ values of your rank. Next, we can say what mistakes you made and what to do to fix them. The final result of the match analysis is a list of recommendations on what you should pay attention to in this game.

What we evaluate:

1. Draft

2. Post-match statistics in merits of GPM, XPM, Hero damage per minute, Last hit per minute, Kills per minute, Tower damage

3. Lasthits and denies during the laning stage

4. Deaths at the laning stage and mid game

5. Pings on the map (Alt + click)

6. Consumables bought

7. Some items (such as Quelling blade and Hand of Midas)

8. Participation in team fights

What we are going to evaluate soon:

1. Skills and talents powered

2. Purchase of all items

3. Navigating the map

4. Chat

Why is mistake analysis on the best? :)

Market evolves, various services emerge that provide similar functionality, but from a different perspective. For example: Gosu.AI for DotA2 and Mobalytics for League of Legends.

Unlike other services, our goal is to give the utmost clear, precise and concise feedback on each user’s action in a particular game, explaining why and what he/she did in the game wrong and how to fix it.

How we will develop this block further on:

We have a bunch of ideas and opportunities to improve feedback

1) We plan to lay out all mistakes made on a timeline in each game to make it clearer when there was a turning point in your actions;
 2) Every day we work on finding new insights and shifting them to user-understandable format of analysis;

3) We are going to add some more approbation to emphasize user’s strong sides, his or her strengths;
 4) Improvement of insight representation design, connection to the rest of the content of the site — video and quests — so that you could see examples of proper games immediately