GENSYNTH: Synthesizing Datalog Programs without Language Bias

Abstract

Techniques for learning logic programs from data typically rely on language bias mechanisms to restrict the hypothesis space. These methods are therefore limited by the user's ability to tune them such that the hypothesis space is simultaneously large enough to include the target program but small enough to admit a tractable search. We propose a technique to learn Datalog programs from input-output examples without requiring the user to specify any language bias. It employs an evolutionary search strategy that mutates candidate programs and evaluates their fitness on the examples using an off-the-shelf Datalog interpreter. We have implemented our approach in a tool called GenSynth and evaluate it on diverse tasks from knowledge discovery, program analysis, and relational queries. Our experiments show that GenSynth can learn correct programs from few examples, including for tasks that require recursion and invented predicates, and is robust to noise.

S_YNTHBX: An Example-guided Synthesizer for Bidirectional Programs on Relations

“…There has been significant work on automatically synthesizing Datalog programs from examples [22], [26], [27], [28]. Alps [27] adopts a bidirectional search strategy with top-down and bottomup refinement operators over the syntax of Datalog programs to traverse the space of possible programs efficiently.…”

“…While DiffLog minimizes the difference between the weighted set of candidate rules and the reference output using numerical optimization, ProSynth uses query provenance to scale the CEGIS procedure and employs a SAT solver for constraint solving. GenSynth [22] learns Datalog programs from examples without requiring any templates, by introducing an evolutionary search strategy that mutates candidate programs and evaluates their fitness on examples using a Datalog solver. While we aim at the synthesis of well-behaved bidirectional programs (get, put) on relations, those works only target the synthesis of unidirectional programs.…”

“…Third, many other general synthesis methods (e.g., Refs. [22], [26]) possibly take examples on relations as input and independently synthesize a get and a put, but they cannot guarantee the well-behavedness of the synthesized pair (e.g., see Example 2.2).…”