Bounded-cost Search Using Estimates of Uncertainty
Authors: Maximilian Fickert, Tianyi Gu, Wheeler Ruml
IJCAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We empirically evaluate XES and the new BEES variants on both planning and heuristic search benchmarks. ... The experiments were run using the lab framework [Seipp et al., 2017] on a cluster of 2.2 GHz Intel Xeon E5-2660 CPUs. |
| Researcher Affiliation | Academia | Maximilian Fickert 1 , Tianyi Gu 2 , Wheeler Ruml2 1Saarland University, Saarland Informatics Campus, Germany 2University of New Hampshire, USA |
| Pseudocode | No | The paper does not contain structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | We implemented the bounded-cost search algorithms in Fast Downward [Helmert, 2006].1 The experiments were run using the lab framework [Seipp et al., 2017]... 1See https://github.com/fickert/fast-downward-xes. |
| Open Datasets | Yes | We use the benchmark set from the bounded-cost track of the 2018 International Planning Competition (IPC), which contains 180 instances from 9 domains, and the instances of the satisficing tracks of all previous IPCs, where we use the upper bounds from Planning.Domains [Muise, 2016] as the cost bounds... |
| Dataset Splits | No | The paper uses benchmark sets from the IPC and Planning.Domains but does not explicitly provide details about training, validation, or test dataset splits in terms of percentages or sample counts for these experiments. |
| Hardware Specification | Yes | The experiments were run using the lab framework [Seipp et al., 2017] on a cluster of 2.2 GHz Intel Xeon E5-2660 CPUs. ... Domain-specific solvers were implemented in C++2 and run on 64-bit Linux systems with 3.16 GHz Intel E8500 Core2 duo processors and 8 GB of RAM. |
| Software Dependencies | No | The paper mentions 'Fast Downward [Helmert, 2006]' and the 'lab framework [Seipp et al., 2017]', and C++ for domain-specific solvers, but does not provide specific version numbers for these software dependencies. |
| Experiment Setup | Yes | To compute the debiased heuristic ˆh(n) for d PTS, BEES, and XES, we use ˆh(n) = h(n)+ ϵh ˆd(n) where ϵh is the mean one-step error in h [Thayer et al., 2011]. We initialize ϵh with 100 virtual samples to avoid a large fluctuation of ˆh values at the beginning of the search. The initial value is set to make ˆh optimistic, using an initial one-step error of zero on the search domains (where the heuristics are admissible), and 0.5 on the planning domains (where the heuristic is inadmissible). All algorithms use a dual queue for preferred operators. |