Notice: This website is continuously updated as new conference proceedings and journal volumes are published. Consequently, the figures presented here may differ from those reported in our publications.
A brief overview of our methodology is provided on the Methods page, along with references to publications documenting our approach and links to the code and data used in each study.
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.
Generalized Potential Heuristics for Classical Planning
Authors: Guillem Francès, Augusto B. Corrêa, Cedric Geissmann, Florian Pommerening
IJCAI 2019 | Venue PDF | LLM Run Details | Input Tokens: 17,731 Total number of tokens sent to the LLM as input for this paper's analysis. | Output Tokens: 5,947 Total number of tokens produced by the LLM (including reasoning/thinking tokens) for this paper's analysis.
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We illustrate the approach empirically on a number of standard domains, where we show that the generated heuristics will correctly generalize to all possible instances. and We have implemented a prototype of the above approach in Python... The experiments below run on Intel Xeon E3-1275 CPUs... |
| Researcher Affiliation | Academia | Guillem Franc es , Augusto B. Corrˆea , Cedric Geissmann and Florian Pommerening University of Basel, Basel, Switzerland EMAIL |
| Pseudocode | No | The text does not contain structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | Source code and benchmarks are available online.3 and the footnote 3 https://doi.org/10.5281/zenodo.3236083 |
| Open Datasets | Yes | Source code and benchmarks are available online.3 and the footnote 3 https://doi.org/10.5281/zenodo.3236083. Also, For each domain, we use training instances with a (reachable) state space of at most 15000 states. |
| Dataset Splits | No | The text mentions 'training instances' and an 'initial set of states S0' but does not provide specific details on how the dataset is split into training, validation, and test sets with percentages or counts. |
| Hardware Specification | Yes | The experiments below run on Intel Xeon E3-1275 CPUs using CPLEX v12.8 as a MIP solver. |
| Software Dependencies | Yes | The experiments below run on Intel Xeon E3-1275 CPUs using CPLEX v12.8 as a MIP solver. |
| Experiment Setup | Yes | For each domain, we use training instances with a (reachable) state space of at most 15000 states. To keep the set of candidate concepts small, we limit role composition to primitive roles... We set a maximum complexity of k = 8 when generating cardinality features and of k = 5 for distance features. The initial set of states S0 has 100 randomly sampled states along with their successors, plus all states in an arbitrary optimal plan. |