Constraints

Constraints are a fundamental component in non-trivial fitting operations. They can also be used to affirm the minimum/maximum of a parameter or tie parameters together in a model.

Anatomy of a constraint

A constraint is a rule which is applied to a dependent variable. This rule can consist of a logical operation, relation to one or more independent variables or an arbitrary function.

Constraints on Parameters

easyscience.Objects.Base.Parameter has the properties builtin_constraints and user_constraints. These are dictionaries which correspond to constraints which are intrinsic and extrinsic to the Parameter. This means that on the value change of the Parameter firstly the builtin_constraints are evaluated, followed by the user_constraints.

Constraints on Fitting

easyscience.fitting.Fitter has the ability to evaluate user supplied constraints which effect the value of both fixed and non-fixed parameters. A good example of one such use case would be the ratio between two parameters, where you would create a easyscience.fitting.Constraints.ObjConstraint.

Using constraints

A constraint can be used in one of three ways; Assignment to a parameter, assignment to fitting or on demand. The first two are covered and on demand is shown below.

from easyscience.fitting.Constraints import NumericConstraint
from easyscience.Objects.Base import Parameter
# Create an `a < 1` constraint
a = Parameter('a', 0.5)
constraint = NumericConstraint(a, '<=', 1)
# Evaluate the constraint on demand
a.value = 5.0
constraint()
# A will now equal 1

Constraint Reference

Built-in constraints

These are the built in constraints which you can use

User created constraints

You can also make your own constraints by subclassing the easyscience.fitting.Constraints.ConstraintBase class. For this at a minimum the abstract methods _parse_operator and __repr__ need to be written.