News

---

Introduction:

We understand that the recent changes to our rating algorithm have caused some confusion. In response to your feedback, we have made further adjustments to the algorithm to enhance both its accuracy and rating volatility. This article aims to explain these changes and address any concerns regarding the fluctuation in ranks.

Enhanced Dynamics:

Previously, our ratings converged strongly, making it challenging for players to experience noticeable changes in their ranks unless they consistently demonstrated superior gameplay. The new changes we have implemented address this issue by increasing the overall rating volatility. As a result, players now have the potential to experience more sudden and significant rating changes.

However, it's important to note that our algorithm still maintains its fundamental principle of not assigning excessive MMR for matches that lack balance. This ensures that games with unbalanced teams, which would result in an easy victory, do not significantly impact the rating.

While higher volatility generally implies decreased rating accuracy, our previous implementation was not optimized to attain its maximum accuracy potential. Consequently, although the volatility changes result in a slight decrease of approximately 0.02% in accuracy, the implementation of our other changes, actually lead to an overall accuracy improvement.

Improved Predictive Accuracy:

Our primary goal was to increase the accuracy of match outcome predictions. Initially, we were able to predict 93% of match outcomes correctly, and we have now improved that number to approximately 98.5%.

Please note that these percentages represent the proportion of accurately predicted match outcomes, which means that perfectly balanced matches would ideally end in a draw. However, due to the absence of a draw outcome in Predecessor, we won't take those matches into consideration for our measurements. Measureing games that are expected to end in draw would result in a predictive accuracy of 63.6%.

New Thresholds

With the implementation of these refinements to our rating algorithm, we anticipate a shift in the rating range. As a result, we will need to adjust the thresholds for each rank accordingly. It's important to note that while the overall distribution of players in each rank will remain the same, the specific thresholds for each rank will be different. This adjustment is necessary to align the ranks with the updated rating system and ensure a fair representation of player skill levels.

Understanding the Data:

These are the test results that have guided our decision to implement the new changes: view data

Taking the specific example of the following line, we observe that out of the total 24,736 matches where our algorithm predicted a 60% chance of winning for a team, 14,572 matches resulted in a victory, the actually achieved result of 58.91% aligns with the initial expectation of 60%. Hence, our algorithm shows a deviation of 1.09% for this particular win probability.

60% -> 14572/24736 (58.91%) E: 1.09%

Thank you for your attention. We look forward to seeing you in game.