Learning CNF Theories Using MDL and Predicate Invention

Authors: Arcchit Jain, Clément Gautrais, Angelika Kimmig, Luc De Raedt

IJCAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our experiments demonstrate that Mistle can learn CNF theories accurately and works well in tasks involving compression and classification. We consider the tasks of compression and classification, and the following questions:
Researcher Affiliation Academia Arcchit Jain1 , Cl ement Gautrais1 , Angelika Kimmig1 and Luc De Raedt1,2 1KU Leuven, Dept. of Computer Science; Leuven.AI, B-3000 Leuven, Belgium 2AASS, Orebro University, Sweden
Pseudocode Yes Algorithm 1 Mistle and Algorithm 2 compress(T)
Open Source Code No The paper mentions the software Mistle is written in and the third-party libraries it uses, but does not provide a link or explicit statement that the source code for Mistle itself is openly available.
Open Datasets Yes For this experiment, we use the UCI datasets [Coenen, 2003; Dua and Graff, 2017].
Dataset Splits Yes To test whether Mistle s theories are not only highly compressed but also accurate, we compare Mistle with KRIMP on a classification task, using 10 fold cross-validation on the UCI datasets.
Hardware Specification Yes Experiment 1 is run on Intel i5 (2.3Ghz; 16Gb RAM) while experiments 2 and 3 are run on Intel i7 (3.4 GHz; 16Gb RAM).
Software Dependencies No Mistle is written in Python 3.7 and uses SPMF library [Fournier-Viger et al., 2016] for closed frequent itemset mining and Pico SAT library [Biere, 2008] for SAT solving. While Python 3.7 has a specific version, the libraries SPMF and Pico SAT do not have their version numbers explicitly stated in this sentence, only cited papers.
Experiment Setup No The paper does not provide specific details on hyperparameters, model initialization, or other system-level training settings. It only mentions the use of 10-fold cross-validation and the choice of minimum support threshold for KRIMP and Mistle.