Trained by a genetic algorithm, uses LSTMs and models various opponents. Plays 6max sit and go.
It's aimed at exploiting weak opponents rather than beating state of the art.
Some work I contributed to a while back. Realized it could interest some people some I'm sharing the code + study here.
It's all at https://github.com/tamlhp/deepbot-poker/tree/master/ the report is at https://github.com/tamlhp/deepbot-poker ... report.pdf
I'm putting two key figures of the report here to spark interest.
Feel free to use, happy to get feedback
| Attachments: |
 action-freq.png [ 42.1 KiB | Viewed 2396 times ] |
 rois.png [ 35.96 KiB | Viewed 2396 times ] |