eloialonso / diamond

DIAMOND (DIffusion As a Model Of eNvironment Dreams) is a reinforcement learning agent trained in a diffusion world model. NeurIPS 2024 Spotlight.
https://diamond-wm.github.io
MIT License
1.53k stars 101 forks source link

can you provide the code? #2

Closed ZhikanggFu closed 4 months ago

ZhikanggFu commented 4 months ago

can you provide the code of calculate results/data/DIAMOND.json

AdamJelley commented 4 months ago

Each value in the results json corresponds to the final evaluation of a single training run (see Launching a training run in the README). The evaluation code can be found in collect_test (specifically, the value used is the final_return_mean on line 298), which is called at the end of run, both in src/trainer.py. Hope that helps!