Versatility

Valorant has 22 different playable agents which can each be classified into one of four roles. Prior to a game, a team goes through "Agent Select," where each player picks the agent (and thus the role) that they'll play during the game. Two players on the same team cannot play the same agent, and there are certain patterns that are generally maintained: at most two duelists, one sentinel, one controller, and at least one initiator.

Not every role is equally desirable. Many players prefer to play duelist over controller, for example; some players end up with a role that they are essentially forced to play due to the rest of the team's selections. Sometimes a player will "fill," or wait for their teammates to pick their agents and then make their decision based on the resulting vacancy that needs to be filled.

The goal of this page is to identify the players who exhibit the most versatility by selecting different agents and roles. On the other hand, there are some players who predominantly play on the same agent and role. It is generally a commendable trait to be able to fit the needs of the team in different ways.

I calculated a naive metric called Versatility Index for this purpose. The versatility metric is a measure of diversity in a player's choice of agent or role, naively based on counts (not taking into account their performance). The measure we use is entropy, a concept from information theory. In the context of player versatility, entropy can be interpreted as the amount of "uncertainty" or "diversity" in the player's agent or role selection. The entropy \(V\) of a player's agent or role distribution is calculated as \(V=-\sum_{i}p_i\log_2(p_i)\), where \(p_i\) represents the proportion of times the \(i\)-th agent or role was chosen. The entropy value will be larger when the distribution of selected agents or roles is more uniform, indicating higher versatility. Conversely, a player who always chooses the same agent or role will have an entropy of zero, indicating low versatility.

I'll repeat the fact that this metric does not take performance into account. It only considers the amount of times an agent or role was played. The obvious problem with this approach is that a player who is very good at one role and very bad at the other three roles may be rated as equally versatile as a player who is very good at every role (if their role counts are identical). In reality, the latter player is obviously more versatile. It's important to understand what exactly a metric is capturing: in this case, it is capturing the amount of diversity in a player's agent or role selection.