Combining the Delete Relaxation with Critical-Path Heuristics: A Direct Characterization
Authors: Maximilian Fickert, Joerg Hoffmann, Marcel Steinmetz
JAIR 2016 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments on IPC benchmarks show that these theoretical advantages can translate into empirical ones. [...] We next introduce our basic notation as well as the ΠC and ΠC ce compilations in Section 2. We spell out our direct characterization of h+(ΠC) in Section 3, and we spell out our generalized relaxed plan extraction methods in Section 4. We summarize our implementation and experiments in Section 5, before concluding in Section 6. |
| Researcher Affiliation | Academia | Maximilian Fickert EMAIL Jörg Hoffmann EMAIL Marcel Steinmetz EMAIL Saarland University, Saarbrücken, Germany |
| Pseudocode | Yes | Algorithm 1: Relaxed plan extraction from h1. [...] Algorithm 2: Relaxed plan extraction from h C. [...] Algorithm 3: Greedy selection of a subset-maximally C-feasible set of supported subgoals G in C-relaxed plan extraction. |
| Open Source Code | No | The paper states: "We implemented h C, h CFF, and h CFF nc in FD (Helmert, 2006)." While FD is an open-source planner, the paper does not explicitly state that the authors' specific implementation of these heuristics or their own code is publicly available or provide a link. |
| Open Datasets | Yes | We used the benchmarks from the satisficing tracks of the two most recent International Planning Competitions, IPC 11 and IPC 14. |
| Dataset Splits | Yes | For each domain, each test suite has 20 instances (some domains have been used in both IPC 11 and IPC 14 so have two test suites). |
| Hardware Specification | Yes | The experiments were run on a cluster of machines with Intel Xeon E5-2660 processors running at 2.2 GHz. The memory limit was set to 4 GB. |
| Software Dependencies | Yes | We implemented h C, h CFF, and h CFF nc in FD (Helmert, 2006). |
| Experiment Setup | Yes | Throughout, we use FD s lazy-greedy best-first search with a dual open queue for preferred operators (Helmert, 2006), which profits from the search space pruning afforded by preferred operators, yet preserves completeness by keeping the pruned nodes in the second open queue. [...] We apply separate runtime limits for C-generation and search respectively, and we will report only about the performance of search not about that of C-learning. Given this, T merely serves as a means to keep the experiments feasible even for large size limits x. We fix T to 30 minutes. [...] Our strategies are: Arbitrary: Choose an arbitrary best supporter, i. e., the first one we find. [...] Random: Choose a random best supporter. [...] Difficulty (h C only): This is the tie breaking mechanism used in FF (Hoffmann & Nebel, 2001). It selects a best supporter that minimizes the summed-up hmax values of the supporter s preconditions. |