Rulearner Algorithm

A Rulearner Algorithm is a Rule Induction Algorithm that takes lattices as an input and outputs set of classification rules.

• AKA: Sahami's Rulearner Algorithm.
• Context:
  • It was developed by Sahami, (1995).
  • It can be implemented by a Rulearner System to solve a Rulearner Task.
  • It can also be implemented into a Supervised Classification System.
• Example(s):
  • Rulearner Algorithm Pseudo-code.
  • …
• Counter-Example(s):
  • CN2 Algorithm,,
  • FINDBESTRULE Algorithm,
  • FOIL Algorithm,
  • FURIA Algorithm,
  • LEM-based Algorithms,
  • Progol Algorithm,
  • RIPPER Algorithm,
  • RULEX Algorithm.
• See: Pattern Mining Algorithm, Decision Tree Induction Algorithm, Inductive Logic Programming, If-Then Rule, Firs-Order Logic Rule.

---

---

References

2004

• (Fu, et al., 2004) ⇒ Huaiyu Fu, Huaiguo Fu, Patrik Njiwoua, and Engelbert Mephu Nguifo (2004, February). "A Comparative Study Of FCA-Based Supervised Classification Algorithms". In: Proceedings of The International Conference on Formal Concept Analysis. DOI: 10.1007/978-3-540-24651-0_26.
  • QUOTE: GRAND builds a set of rules, CLNN & CLNB generates a set of meta-rules, LEGAL and GALOIS construct relevant concepts, RULEARNER induces ordered rules set or unordered rules set to represent learned knowledge (...)

    Here we show the experimental comparisons of four methods in the same Java platform: GRAND, LEGAL, GALOIS, RULEARNER (...)

    The classification performance of the four methods can be divided into two groups. GRAND, GALOIS can classify multi-data datasets and LEGAL, RULEARNER can only classify two-class datasets. The statistical analysis on the two-class datasets for the four methods shows that:

    • RULEARNER produces the highest average classification accuracies and LEGAL obtains the best running time among the four methods.
    • RULEARNER and LEGAL perform better than C4.5 on average classification accuracies, but C4.5 significantly outperforms the four methods in running times.
    • RULEARNER, LEGAL and C4.5 outperform GRAND and GALOIS in average classification accuracies.

1995

• (Sahami, 1995) ⇒ Mehran Sahami (1995). "Learning Classification Rules Using Lattices". In: Proceedings of the European Conference on Machine Learning (ECML-95). DOI:10.1007/3-540-59286-5_83
  • QUOTE: This paper presents a novel induction algorithm, Rulearner, which induces classification rules using a Galois lattice as an explicit map through the search space of rules (...) The Rulearner system is also capable of learning both decision lists and unordered rule sets allowing for comparisons of these different learning paradigms within the same algorithmic framework (...)

    The Rulearner algorithm takes as input a lattice, L, and a set of instance classification labelings, C, which correspond to the instance nodes in L. The algorithm produces a set of symbolic classification rules as output. The user is also able to specify a noise parameter to the system as a percentage figure by which each induced rule can misclassify some portion of the training instances that it applies to. Furthermore, the user can configure the system to induce either a decision-list or an unordered set of rules, and can also decide whether the classification rules induced should only identify one given labeling (i.e. derive only rules with a "positive" labeling) or if the rules should attempt to classify all instances in the training set. While the basic framework of the algorithm is general enough to properly deal with more than two classification labelings, we present the algorithm here as a binary classifier for easier understanding.

    

    Table 1: Pseudo-code for the Rulearner algorithm. Note that L is the input lattice and C is the classification labelings for the instances which were used in building L.

---