In this article, I formalize Pokémon teambuilding as an optimization problem and introduce two models:
• a no-interaction baseline
• an interaction model that incorporates synergy and counterplay
Both highlight strong, versatile attackers (Alakazam, Zapdos, Starmie). This makes sense given that the models are trained on data generated by a simple AI — but it deviates from human play. Tauros, considered the strongest Pokémon in Gen 1 OU, only ranks 7th.
The interaction model also captures intuitive dynamics. For example, it recognizes that Starmie (Water) is countered by Jolteon (Electric).
More broadly, the optimal counter team identified by the interaction model defeats the best baseline team over 65% of the time — close to the model’s predicted win rate.
From there, I derive a simplified Nash equilibrium for teambuilding.
The full article is here: https://sahovic.fr/algorithmic-pokemon-teambuilding
Let me know what you think!
• a no-interaction baseline
• an interaction model that incorporates synergy and counterplay
Both highlight strong, versatile attackers (Alakazam, Zapdos, Starmie). This makes sense given that the models are trained on data generated by a simple AI — but it deviates from human play. Tauros, considered the strongest Pokémon in Gen 1 OU, only ranks 7th.
The interaction model also captures intuitive dynamics. For example, it recognizes that Starmie (Water) is countered by Jolteon (Electric).
More broadly, the optimal counter team identified by the interaction model defeats the best baseline team over 65% of the time — close to the model’s predicted win rate.
From there, I derive a simplified Nash equilibrium for teambuilding.
The full article is here: https://sahovic.fr/algorithmic-pokemon-teambuilding
Let me know what you think!
