NeSy4PPM documentation

NeSy4PPM is the first Python package designed for both single-attribute (e.g., activity) and multi-attribute (e.g., activity and resource) suffix prediction in predictive process monitoring. It implements a Neuro-Symbolic (NeSy) system that integrates neural models with various types of symbolic background knowledge (BK), enabling accurate and compliant predictions even under concept drift.

NeSy4PPM offers the following key features:

1. Symbolic knowledge decoding-time integration: contextualizes neural predictions with declarative or procedural knowledge, including DECLARE, MP-DECLARE (multi-perspective DECLARE), ProbDECLARE (probabilistic DECLARE), and Petri nets.

2. Flexible learning: provides multiple prefix encoding methods and supports LSTM (Long Short-Term Memory) and Transformer architectures.

3. SDFA-constrained decoding: combines neural next-event probabilities with a stochastic deterministic finite automaton (SDFA) during beam search, with optional pruning of infeasible transitions and suffixes that cannot terminate within the remaining prediction horizon.

Contents:

• NeSy4PPM
  • NeSy4PPM package
• Installation
  • Local Installation
  • PyPI Installation
  • Docker Installation
• Tutorials
  • NeSy4PPM: Suffix prediction Tutorial
  • 1 Data preprocessing
  • 2 Training
  • 3 Prediction
  • 4 Evaluation
• Repository Structure
• Publications
• Citing NeSy4PPM