Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].
Strong Stubborn Set Pruning for Star-Topology Decoupled State Space Search
Authors: Daniel Gnad, Jörg Hoffmann, Martin Wehrle
JAIR 2019 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We evaluate our DSSS techniques from both a theoretical and a practical perspective. For the theoretical part, we analyze exponential separations: example task families scaling in a size parameter n, exponentially separating some method A from another one B in that A yields search space size polynomial in n while B yields search space size exponential in n. ... For the practical evaluation, we run our techniques on standard benchmarks from the International Planning Competition (IPC). |
| Researcher Affiliation | Academia | Daniel Gnad EMAIL Saarland University Saarland Informatics Campus Saarbr ucken, Germany J org Hoffmann EMAIL Saarland University Saarland Informatics Campus Saarbr ucken, Germany Martin Wehrle EMAIL University of Basel Basel, Switzerland |
| Pseudocode | No | The paper describes algorithms and methods verbally and through logical steps in proofs (e.g., Definitions 4, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15 and Theorems 1, 2, 3, 6, 7), but it does not contain explicitly labeled 'Pseudocode' or 'Algorithm' blocks with structured steps formatted like code. |
| Open Source Code | No | Our implementation extends the decoupled search planner of Gnad and Hoffmann (2018), which is based on the Fast Downward planning system (FD) (Helmert, 2006). We use the Lab software package to execute the experiments (Seipp, Pommerening, Sievers, & Helmert, 2017). |
| Open Datasets | Yes | We focus our experimental evaluation on optimal planning and proving unsolvability of planning instances... We evaluate our algorithms in optimal planning on all STRIPS benchmarks from the optimal tracks of the International Planning Competition (IPC) ( 9818). ... We use two different configurations. Table 5 shows results for blind search... In Table 6, we additionally use the hmax heuristic as a deadend detector (Bonet & Geffner, 2001). For all planners based on decoupled search we switch to a pricing function that only distinguishes between reachable (price 0) and unreachable leaf states (price ). We show results on all domains from the Unsolvability IPC 16 and those from Hoffmann et al. (2014) where not all instances have been reused for the competition. |
| Dataset Splits | No | The paper refers to using "standard benchmarks from the International Planning Competition (IPC)" and "domains from the Unsolvability IPC 16", implying predefined datasets. However, it does not explicitly provide information on training/test/validation splits, sample counts, or specific splitting methodologies within the paper's text for reproducing the data partitioning. |
| Hardware Specification | Yes | All experiments are conducted on a cluster of Intel E5-2660 machines running at 2.20 GHz, with a time cut-off of 30 minutes and a memory limit of 4 GB. |
| Software Dependencies | No | Our implementation extends the decoupled search planner of Gnad and Hoffmann (2018), which is based on the Fast Downward planning system (FD) (Helmert, 2006). We use the Lab software package to execute the experiments (Seipp, Pommerening, Sievers, & Helmert, 2017). ... When using h LM-cut, we include Complementary 2 (C2) to the comparison, the best non-portfolio planner in the optimal track of IPC 18 (Franco, Lelis, & Barley, 2018). |
| Experiment Setup | Yes | All experiments are conducted on a cluster of Intel E5-2660 machines running at 2.20 GHz, with a time cut-off of 30 minutes and a memory limit of 4 GB. ... We compare the performance of decoupled strong stubborn sets pruning (DS3) to its base components, i.e., standard search with strong stubborn sets pruning (S3), decoupled search (DS), as well as standard search without pruning (B). We use the strategy fullsyntactic-SSS-EC , which is the strongest variant of SSS pruning that is implemented in a recent Fast Downward. ... We implement a simple safety belt mechanism to disable the pruning in planning instances where stubborn sets are not effective. To do so, we switch the stubborn sets off if after the first 1000 expanded states less than 1% of the transitions have been pruned. |