This repository contains the code and model for the GuideEnricher: Protecting the Anonymity of Ethereum Mixing Service Users with Deep Reinforcement Learning. Read the full paper USENIX 24.
GuideEnricher requires Python 3.8 or later. You can install the required packages by running: conda create --name <env> --file conda_requirements.txt
Note: GuideEnricher is developed on ray 2.0.0 and tensorflow 2.10.0
To retrain the game, you can use the following command:
python evader.py --name "<name of your experiment>" \
--data_dir "<>" \
--log_dir "<>" \
--checkpoint_dir "<>" \
--no_addresses_agent_challenge_table 2 \
--agent_challenge_table "[2,3]" \
--agent_address_range_starts 0 \
--agent_address_range_end 250 \
--agent_mutable_address_range_start 10 \
--agent_mutable_address_range_end 20 \
--crowd_address_range_starts 250 \
--no_of_crowd 50 \
--no_of_wallets_for_each_crowd_agent 50 \
--amount_of_money_in_each_crowd 100 \
--fcnet_hiddens "[32, 32]" \
--no_iter 500
You can modify the parameters as your game. eg: agent_challenge_table, agent_address_range_starts, agent_address_range_end, agent_mutable_address_range_start, agent_mutable_address_range_end, crowd_address_range_starts, no_of_crowd, no_of_wallets_for_each_crowd_agent, amount_of_money_in_each_crowd, fcnet_hiddens, no_iter.
To run the pretrained game, load the agent's model and config (check model dir), and run :
python eval_evader.py
Read the extended appendix: here