How an A.I. is Becoming the World's Best Pokemon Player

• showdown

  5 237 6.3 Python

  A Pokemon Showdown Battle Bot written in Python (by pmariglia)

He doesn't reveal any big details about the algorithm, but from what I can tell his final iteration is some kind of expectiminimax + pruning, where essentially the bot is looking forward a few turns and based on all possible (or usually all "reasonable") outcomes, picks the moves that minimize the worst outcome, aka your traditional "safe" player. There are definitely limitations to this approach, first of all is the depth of search. To give context, Stockfish regularly searches to 30 moves ahead. I am pretty sure that a lot of implementations are limited to a mere 3 moves ahead, as illustrated here (https://github.com/pmariglia/showdown/issues/53). So, I suspect that ~1600 is going to be a pretty hard limit because you can't use this approach by itself to really develop long-term strategies. Another thing I wanted to point out is that this algorithm is also really prone to exploitability- knowing that your opponent is always a safe player is going to lead to easy manipulation by the best players. He even mentions it here in his Smogon post:

• InfluxDB

  www.influxdata.com featured

  Power Real-Time Data Analytics at Scale. Get real-time insights from all types of time series data with InfluxDB. Ingest, query, and analyze billions of data points in real-time with unbounded cardinality.

  

Suggest a related project