Skip to content

Running the Planner

Guillem Francès edited this page Feb 8, 2017 · 1 revision

(I'll keep upgrading this as the planner evolves these days)

Common Configurations

Just as a quick reference, read below for further details:

Run a plain plain BFWS(w_#g, #g). Successors are generated naively:

python3 preprocessor/runner.py --tag feb8 --instance [...]/instance.pddl --driver lazybfws --options="bfws.rs=none,successor_generation=naive" --run

Same but using a match tree:

python3 preprocessor/runner.py --tag feb8 --instance [...]/instance.pddl --driver lazybfws --options="bfws.rs=aptk_hff,successor_generation=match_tree" --run

Same but using gringo to ground and perform reachability analysis:

python3 preprocessor/runner.py --tag feb8 --gringo --instance [...]/instance.pddl --driver lazybfws --options="bfws.rs=aptk_hff,successor_generation=match_tree" --run

Same but using an "adaptive" novelty evaluator that will try to use the best specialization (e.g. binary features only, etc.) before resorting to the generic and somewhat less performant evaluator:

python3 preprocessor/runner.py --tag feb8 --gringo --instance [...]/instance.pddl --driver lazybfws --options="bfws.rs=aptk_hff,successor_generation=match_tree,evaluator_t=adaptive" --run

Type of Simulation / Relevant Atomset

Controlled by the option bfws.rs. Currently can be:

  • none: No simulations whatsoever, plain BFWS(w_#g, #g)

  • aptk_hff: Run Nir's version of H_FF and compute a set of relevant atoms R_APTK_FF from it, so that #r can be used as usual in the search.

  • sim: No info about action preconditions and effects is used - an IW(k) simulation is run from the state and information about a set of relevant atoms R_SIM is extracted from there.

  • macro: No set of relevant atoms R is used. Instead, simulations are run from certain states, and the paths to subgoals are used as macros when expanding those states.

Running FS on FSTRIPS Problems.

FSTRIPS domains pose some restrictions on the planner options that can be used:

  • The match tree is probably not yet well suited for FSTRIPS:
  • Same for the --gringo parser option.