ShIOEnv_40cmd_7x10K

Datasets of Linux command inputs and their observed execution behaviors collected from the ShIOEnv environment for 40 utilities. Each dataset is curated using different methods of argument construction from a defined context-free grammar (CFG): unconstrained random truncated (UCRT) : randomly select productions from the full set of productions, truncated randomly to reduce argument redundancy. unconstrained policy network (UCPN-m0) : policy network updated using proximal policy optimization over 20,000 episodes with a redundancy score margin of 0, selecting from the full set of productions. unconstrained policy network (UCPN-m50) : policy network updated using proximal policy optimization over 20,000 episodes with a redundancy score margin of 0.50, selecting from the full set of productions. grammar-constrained random truncated (GCRT) : randomly select productions from valid expansions, truncated randomly to reduce argument redundancy. grammar-constrained policy network (GCPN-m0) : policy network updated using proximal policy optimization over 20,000 episodes with a redundancy score margin of 0, selecting from valid expansions. grammar-constrained policy network (GCPN-m50) : policy network updated using proximal policy optimization over 20,000 episodes with a redundancy score margin of 0.50, selecting from valid expansions. NL2Bash : Bootstrapped NL2Bash dataset adapted to be executable in the default container provided in ShIOEnv. Refer to dataset_dist_.png for distributions of each field in each dataset. The generating environment and agent are available on GitHub: https://github.com/synlab-jragsdale/ShIOEnv/tree/main