Machine Learning Pokemon Battle Bot

1. Random Bot

Strategy: Just picks whatever move randomly

This is basically my control group. It literally just picks a random move from whatever's available each turn. No thinking, no strategy, just pure randomness. I needed this as a baseline to see if my other bots are actually doing anything smart or if they're just getting lucky. If my fancy bots can't beat a random number generator, then I know something's really wrong with my code.

2. Max Damage Bot

Strategy: Always go for the biggest number

This one's pretty straightforward. Every turn, it looks at all the moves it can use and just picks whichever one has the highest base power. So if it has Earthquake (base power 100) and Tackle (base power 40), it's always going with Earthquake. The interesting thing is that this bot completely ignores type effectiveness. It might use a Water move against a Water type even though that does basically nothing. But despite being super simple, it actually performs surprisingly well against the random bot, which shows that having any kind of consistent strategy is better than nothing.

3. Custom Strategy Bot

Strategy: Actually tries to play smart

This is where things get interesting. I built this bot to use a priority system where different strategic factors get weighted and combined to make decisions. Here's what it thinks about:

• Type effectiveness: This gets the highest priority. If it can hit something super effective (using Electric against Water, for example), it really wants to do that because super effective moves deal 2x damage.
• STAB bonus: STAB stands for Same Type Attack Bonus. When a Pokemon uses a move that matches its own type, it gets a 1.5x damage boost. So a Water type Pokemon using Water Gun does more damage than a Grass type using the same move.
• Dynamax timing: In Gen 8, you can Dynamax once per battle to make your Pokemon giant and boost its HP and move power. The bot tries to use this strategically when both Pokemon are relatively healthy so it can actually swing the momentum of the battle.
• Switching decisions: If the bot's current Pokemon is taking super effective damage or can't hit the opponent effectively, it'll switch to something with a better type matchup instead of just staying in and dying.
• Setup moves: Sometimes it's worth using a turn to boost your stats (using Swords Dance to double your attack, for example) instead of attacking immediately. The bot recognizes when it's safe to do this, such as when the opponent can't hit it very hard.

Performance: Gets about 59% win rate against the other bots, which is pretty solid. Against actual human players on the ladder though, it sits around 40% win rate, which honestly isn't terrible for a rule based bot.

Version 1: The First Attempt

The bot went 0 wins and 25 losses. It didn't win a single game. Looking back at the battle logs, I found some problems. The bot never used Dynamax at all, even though that's a huge mechanic in Gen 8. It would stay in bad matchups instead of switching out. And it never used setup moves, so it was missing opportunities to snowball advantages.

Improvements

I went back and made improvements based on what I learned from those 25 losses.

Then I ran another 25 battles and it went 10 wins and 15 losses. That's a 40% win rate. Going from 0% to 40% is a significant improvement. It's still losing more than it wins, but for a bot that's just following hardcoded rules without any learning, the results are acceptable.

What I Changed Specifically:

• Dynamax timing: Now it will Dynamax when the opponent has > 60% HP and the bot has > 40% HP. This ensures we're not wasting it on a Pokemon that's about to faint, and we're using it when it can actually turn the tide of battle.
• Setup move usage: The bot now recognizes when the opponent's attacks aren't doing much damage (resisted hits) and will take the opportunity to use setup moves to boost stats.
• Better switching: I made the penalty for staying in when your moves are not very effective way higher. So now if it's doing < 0.5x damage, it's much more likely to switch out to something better.
• Offensive pressure: When the opponent is at high HP, the bot prioritizes hitting hard rather than playing safe. This prevents it from getting out pressured early in the battle.