RiotAPI - Riot Games offers an API to gather various game information which is offered to the LCU client. With a registered API Key, any user can access this information. With the power of machine learning, we can gather deeper insights than typical win rates to help us improve at our own play.

Aid discovery and learning: While Teamfight Tactics is easy to pick up, making sense of TFT has its own set of challenges. 

Recommend Data-Backed Composition:  Help player craft team compositions based on their given set of units, and items. (Aspiration with item consideration)

Contextual statistics: Provide statistics specific to conditions where the subject is powerful. Not a general statistics like Overall Win Rate. 

Building team synergies is a key path to success in Teamfight Tactics, an exercise that player needs to do quickly in their head is find a set of  units  with that maximize traits given their current unit limits. 

This take a vase amount of dispersed knowledge, sprinkled all over the game. There has yet to be a popularized method to aggregate and make sense of all the information in one single visualization. Currently information about the units are displayed 1 dimensionally. 

• Via; Unit Cards, or Traits Display.

There has to be a better way to display this information! Origin and Classes generally unique per unit; by plotting units against these two features, we can build a table where scanning vertically or horizontally can help players quickly identify units of interest. 

By overlaying this table with additional information like key carries, and their items, players build effective compositions without decicated time or investment in testing these combinations.

Placement Regression:

Augment data are not included in the Riot API. 

We are interested in understanding how team composition affects placements --> How do we measure the strength of a composition.

Using an XGBoost Regressor, I was able to get a R² Score of ~0.59, which explains that 59% of the  in placement can be explained by Units (and traits) alone. But this the data has an underlying behavior where players selects best items for their units. So appropriate items are also implied. So appropriate items are also implied.

With the machine learning algorithms, we can understand it's decision process to understand what makes effective and powerful game winning positions. Sparse models (models that uses few features to make predictions) are the most interpretable. Most ML models are blackboxed and its difficult to extract knowledge from, but this is important for us if we want to extract learnings. 

So we will engineer features to help the algorithm process;

• Number of $x Cost Units

• Highest * of $x cost units.

• # of Trait Counts

• Highest Trait's Unit Requirement