def _get_features(self, state: tuple[Pokemon, Pokemon], move: Move):
# TODO: Optimize feature extraction
pokemon_a, pokemon_b = state
type_mod = Type.dmg_modifier(move.info.type, pokemon_b.species.types[0]) * \
(Type.dmg_modifier(move.info.type, pokemon_b.species.types[1]) if len(pokemon_b.species.types) > 1 else 1)
type_mod /= 4 # Normalize the value between 0 and 1
dmg_class_mod_stat_atk, dmg_class_mod_stat_def = (pokemon_a.stats.attack, pokemon_b.stats.defense) \
if move.info.damage_class is DamageClass.PHYSICAL else (pokemon_a.stats.special, pokemon_b.stats.special)
dmg_class_mod = (dmg_class_mod_stat_atk /
(dmg_class_mod_stat_atk + dmg_class_mod_stat_def))
return [
pokemon_a.hp / pokemon_a.stats.total_hp,
pokemon_b.hp / pokemon_b.stats.total_hp,
dmg_class_mod,
type_mod,
(move.info.power / 250),
(move.info.accuracy if move.info.accuracy is not None else 1),
# move.info.hit_info.min_hits,
# move.info.hit_info.max_hits,
int(move.info.high_crit_ratio),
# move.info.priority,
move.info.drain,
# move.info.healing,
# int(move.pp == 0)
# TODO: Add more features
]
def _calc_modifier(self, attacking_pokemon: PokemonSpecies,
defending_pokemon: PokemonSpecies,
move_used: MoveInfo) -> float:
rand_modifier = random.uniform(0.85, 1.0)
if move_used.name == ATTACK_SELF:
return rand_modifier
stab = 1.5 if move_used.type in attacking_pokemon.types else 1
type_modifier = Type.dmg_modifier(move_used.type,
defending_pokemon.types[0])
if len(defending_pokemon.types) > 1:
type_modifier *= Type.dmg_modifier(move_used.type,
defending_pokemon.types[1])
return rand_modifier * stab * type_modifier
def _types_advantage(self, pokemon_a: PokemonSpecies,
pokemon_b: PokemonSpecies) -> Matchup:
weight = math.prod(
Type.dmg_modifier(at, dt)
for at, dt in itertools.product(pokemon_a.types, pokemon_b.types))
if weight > 1:
return Matchup.ADVANTAGEOUS
elif weight < 1:
return Matchup.DISADVANTAGEOUS
else:
return Matchup.NEUTRAL
