Welcome to a deep dive into SeaPearl, a cutting-edge Constraint Programming (CP) solver that leverages Reinforcement Learning (RL) agents to enhance the efficiency of value-selection heuristics. Think of SeaPearl as a sophisticated toolkit for researchers, pushing the boundaries of what’s achievable in solving complex problems.

What is SeaPearl?

SeaPearl allows for solving various CP problems by employing RL techniques as a dynamic approach to derive solutions. Whether you are dealing with classic CP solving or seeking to employ advanced RL methods, SeaPearl provides a structured pathway to achieving your objectives.

The accompanying paper on arXiv offers further insights into the methodologies used. If you wish to cite SeaPearl in your research, please do!

Installation

To get started with SeaPearl, install it via the Julia package manager:

julia] add SeaPearl

Using SeaPearl

General Overview

SeaPearl can be utilized in two ways:

• As a classic CP solver with predefined heuristics
• As an RL-driven CP solver that learns through experience

1. Using SeaPearl as a Classic CP Solver

When you opt for the classic approach, you’ll be working with a sequence of steps to define your problem:

• Declare a variable selection heuristic:

  YourVariableSelectionHeuristic = SeaPearl.AbstractVariableSelectionTakeObjective

• Declare a value selection heuristic:

  BasicHeuristic = ValueSelection

• Create a Constraint Programming Model:

  trailer = SeaPearl.Trailer()
  model = SeaPearl.CPModel(trailer)

• Create variables, add constraints, and set objectives:

  SeaPearl.addVariable!(...)
  SeaPearl.addConstraint!(model, SeaPearl.AbstractConstraint(...))
  SeaPearl.addObjective!(model, ObjectiveVar)

2. Using SeaPearl as an RL-Driven CP Solver

The RL-driven approach unlocks more complex functionalities:

• Declare selection heuristics:

  CustomVariableSelectionHeuristic = SeaPearl.AbstractVariableSelectionTakeObjective

• Define an agent:

  agent = RL.Agent(policy=(...), trajectory=(...))

• Create various optional customizations: From rewards to state representations, you can tailor the user experience as needed.
• Create a custom problem generator:

  CustomProblemGenerator = AbstractModelGenerator

• Train the agent:

  metricsArray, eval_metricsArray = SeaPearl.train!(...) 

Understanding SeaPearl through an Analogy

Imagine SeaPearl as a chef in a kitchen. To create the perfect meal, the chef needs quality ingredients (variables and constraints), a recipe (model), and the right cooking techniques (heuristics). Whether the chef follows a conventional recipe (classic CP approach) or tries to come up with a brand-new dish through experience and feedback (RL-driven approach), the end goal remains the same: to create a dish that satisfies diners (solve the problem).

Troubleshooting Tips

• If faced with issues during installation, check your Julia environment settings and ensure that you’re connected to the internet.
• In case of errors while executing code, verify the syntax and confirm that all necessary packages are properly referenced.
• For a smoother experience, reference the SeaPearlZoo for working examples and the documentation.
• For more insights, updates, or to collaborate on AI development projects, stay connected with fxis.ai.

Conclusion

At fxis.ai, we believe that such advancements are crucial for the future of AI, as they enable more comprehensive and effective solutions. Our team is continually exploring new methodologies to push the envelope in artificial intelligence, ensuring that our clients benefit from the latest technological innovations.