Skip to content

Data Model

OptiScope uses a structured and extensible data model to represent optimization results. This model is designed to handle various types of optimization problems, from simple single-objective problems to complex multi-objective problems with many constraints and custom data types.

Core Concepts

The central class in OptiScope is OptimizationResult. It serves as a container for all data associated with an optimization run, including:

  • Design Variables: The input parameters that the optimizer adjusts.
  • Objectives: The values to be minimized or maximized.
  • Constraints: Conditions that the solution must satisfy (equality or inequality).
  • Observables: Additional quantities computed during the optimization that are not objectives or constraints but are useful for analysis.
  • Metadata: Information about the problem, solver, and variables.
  • Result Sets: Named subsets of solutions (e.g., "Pareto Front", "Feasible Solutions").

OptimizationResult

The OptimizationResult class organizes data into pandas DataFrames, ensuring efficient storage and manipulation. It guarantees that all data components (variables, objectives, etc.) are aligned, meaning the \(i\)-th row in the design variables DataFrame corresponds to the \(i\)-th row in the objectives DataFrame.

from optiscope.core import OptimizationResult

# Accessing data
design_vars = result.design_variables
objectives = result.objectives
constraints = result.inequality_constraints

Problem Metadata

The ProblemMetadata class stores high-level information about the optimization problem:

  • name: Name of the problem.
  • description: A brief description.
  • solver: Name of the solver used.
  • run_date: Date and time of the run.
  • n_evaluations: Number of function evaluations.

Variable Types

OptiScope defines specific metadata classes for different types of variables. These classes provide context such as units, bounds, and optimization direction.

Design Variables

Represent the decision variables of the problem.

  • Properties: lower_bound, upper_bound, is_discrete, possible_values, initial_value.

Objectives

Represent the goals of the optimization.

  • Properties: direction (MINIMIZE or MAXIMIZE), ideal_value, nadir_value.

Constraints

Represent the restrictions on the design space.

  • Properties: constraint_type (INEQUALITY, EQUALITY, BOUND), reference_value, tolerance.
  • Inequality: \(g(x) \le \text{reference\_value}\)
  • Equality: \(h(x) = \text{reference\_value}\)

Observables

Represent auxiliary data computed during the run.

Result Sets

ResultSets allow you to group specific solutions for focused analysis. Common examples include:

  • Pareto Front: The set of non-dominated solutions.
  • Feasible Set: Solutions that satisfy all constraints.
  • User Selections: Subsets manually selected by the user during analysis.

You can create, retrieve, and manipulate sets using the OptimizationResult methods:

# Create a set of feasible solutions
feasible_set = result.check_feasibility(set_name="feasible")

# Get data for a specific set
pareto_data = result.get_set_data("pareto_front", category=DataTypeCategory.OBJECTIVE)