Notebooks and Experiments

This code base is tested under Python 3.9.18

1. Ready-to-run notebooks that are mentioned in Theorem 1&2.

• kMemBR.ipynb demonstrates how to compute BRs in IPD
• kMemNE_full.ipynb demonstrates how to compute NEs in IPD

2. Compute BR to strategies in Axelrod's IPD library

python axl2tab.py

3. Compute BR in ITD

# usage: python OPPO_MEM MY_MEM NUM_TRAIN_EPISODE MAX_ITD_ROUND TRAIN_or_EVAL, e.g.,
python itd_learn.py 2 1 1e6 20 train

4. RL in the Pursuit Domain

1. NE seeking

# Usage: python marl.py --alg "$ALG" --cfg "$CFG" --stack "$STACK" --it "$IT" --device "cuda:$DEVICE", e.g.,
python marl.py --alg DQN --cfg MidCatchMap16 --stack 4 --it 30e6 --device cuda:0

2. 1-to-7 best response

# Usage: python marl_sbr.py --alg "$ALG" --cfg "$CFG" --stack "$STACK" --it "$IT" --device "cuda:$DEVICE" --load-op "$OP" --op-stack "$OPSTACK", e.g.,
python marl_sbr.py --alg DQN --cfg MidCatchMap16 --stack 4 --it 10e6 --device cuda:0 --load-op DQN_MidCatchMap16_8_60e6_2025-09-09-190423 --op-stack 8

3. 4-to-4 best response

# Usage: python marl_team_sbr.py --alg "$ALG" --cfg "$CFG" --stack "$STACK" --it "$IT" --device "cuda:$DEVICE" --load-op "$OP" --op-stack "$OPSTACK", e.g.,
python marl_team_sbr.py --alg DQN --cfg MidCatchMap16 --stack 4 --it 10e6 --device cuda:0 --load-op DQN_MidCatchMap16_8_60e6_2025-09-09-190423 --op-stack 8

Notes

• The detailed usage can be further found using the --help flag.
• Only some prototype pertained model is included.