Attentive Relation Network for Object based Video Games
This is an implementation of the paper Attentive Relation Network for Object based Video Games
- Python = 3.6
- tensorflow = 1.12
- gym = 0.12.1
- stable-baselines = 2.8.0
- opencv-python = 4.1.0.25
- matplotlib = 3.1.0
Train an agent: nohup python a2c_main.py --env=BreakoutNoFrameskip-v4 --net=ARNSRF --n_cpu=16 --gpu=4 --total_steps=5000000 --date=20210210 --seed=0 &
Evaluate during training: nohup python a2c_evaluate.py &
Evaluate an agent in the modified Breakout: nohup python a2c_evaluate_zero_shot_po.py --model_dir=the_model_path/ &
Train a visualizable agent: nohup python a2c_main_vis.py &
See the visualization: run visualize.ipynb in jupyter notebook