org.omegahat.Numerics.Optimizers
Class ModelObjectiveOmegahat

java.lang.Object
  |
  +--org.omegahat.Numerics.Optimizers.ModelObjectiveOmegahat

All Implemented Interfaces:

ModelObjective

public class ModelObjectiveOmegahat

extends java.lang.Object

implements ModelObjective

Create an object to evaluate a model function and possible gradients, starting from one or more Omegahat expressions.

Field Summary

protected Evaluator	evaluator The evaluator object to be used to evaluate the objective.
protected ExpressionInt	initialize An expression to be evaluated when the object is first intialized by the optimizer.
protected ExpressionInt	model The expression defining the model objective, at the current parameter values.
protected java.lang.String[]	parameterNames

Constructor Summary

ModelObjectiveOmegahat()
Create a model objective for which the model (as an Omegahat expression) and the evaluator will be specified later.

ModelObjectiveOmegahat(ExpressionInt model)
Create the complete model objective with the given model formula and an empty evaluator of class BasicEvaluator.

Method Summary

void	assignParameters(double[] pars) Assign the parameters to the evaluator's database.
void	assignParameters(ModelPointNumericInt arg) Assign the parameters from the model point to the evaluator's database.
NumericArray	collectGradient(ModelPointNumericInt x) Retrieve the gradient from the database, if it is there and non-null.
NumericMatrix	collectHessian(ModelPointNumericInt x) Retrieve the hessian from the database, if it is there and non-null.
double	collectValue(ModelPointNumericInt x) Retrieve the value from the database and assign it into the value property of the model point argument.
ModelPoint	eval(ModelPointNumericInt arg) Evaluate the model objective at this point.
Evaluator	getEvaluator() Accessor for evaluator field
NumericArray	getGradient(ModelPointNumericInt x) Retreive the return the computed gradient of the objective, and store it in the ModelPoint argument, if that is non-null.
NumericMatrix	getHessian(ModelPointNumericInt x) Retreive the return the computed hessian of the objective, and store it in the ModelPoint argument, if that is non-null.
ExpressionInt	getInitialize() Accessor for initialize field
ExpressionInt	getModel() Accessor for model field
java.lang.String[]	getParameterNames() Accessor for parameterNames field
double	getValue(ModelPointNumericInt x) Retreive the return the computed value of the objective, and store it in the ModelPoint argument, if that is non-null.
boolean	handleError(ModelPointNumericInt current, java.lang.Throwable e) Handle an exception or error that occurred during the fitting.
Evaluator	setEvaluator(Evaluator value) Accessor for setting evaluator field
ExpressionInt	setInitialize(ExpressionInt value) Accessor for setting initialize field
ExpressionInt	setModel(ExpressionInt value) Accessor for setting model field
java.lang.String[]	setParameterNames(java.lang.String[] value) Accessor for setting parameterNames field

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Field Detail

evaluator

protected Evaluator evaluator

The evaluator object to be used to evaluate the objective.

If not specified, the initializing code will create one.

initialize

protected ExpressionInt initialize

An expression to be evaluated when the object is first intialized by the optimizer.

This may be used to attach database(s) or do other one-time computations.

model

protected ExpressionInt model

The expression defining the model objective, at the current parameter values.

The method for communicating the parameter values to the expression and for communicating the can be controlled in several ways. By default, the parameters are assigned under the name "Parameters" in the evaluator's default database. If the property parameterNames is set to a String[] array, the individual parameters will be assigned with the corresponding names.

When the expression completes, it should assign the value of the model objective with the name "Value"; if the gradient or hessian is computed by the expression, these should be stored as "Gradient" or Hessian respectively.

We expect to define alternative communication mechanisms in the future; in any case, other communication mechanisms can be defined by sub-classing this class and over-riding the assignParameters, getValue, getGradient, or getHessian methods as appropriate.

parameterNames

protected java.lang.String[] parameterNames

Constructor Detail

ModelObjectiveOmegahat

public ModelObjectiveOmegahat()

Create a model objective for which the model (as an Omegahat expression) and the evaluator will be specified later.

ModelObjectiveOmegahat

public ModelObjectiveOmegahat(ExpressionInt model)

Create the complete model objective with the given model formula and an empty evaluator of class BasicEvaluator.

Method Detail

getEvaluator

public Evaluator getEvaluator()

Accessor for evaluator field

setEvaluator

public Evaluator setEvaluator(Evaluator value)

Accessor for setting evaluator field

getInitialize

public ExpressionInt getInitialize()

Accessor for initialize field

setInitialize

public ExpressionInt setInitialize(ExpressionInt value)

Accessor for setting initialize field

getModel

public ExpressionInt getModel()

Accessor for model field

setModel

public ExpressionInt setModel(ExpressionInt value)

Accessor for setting model field

getParameterNames

public java.lang.String[] getParameterNames()

Accessor for parameterNames field

setParameterNames

public java.lang.String[] setParameterNames(java.lang.String[] value)

Accessor for setting parameterNames field

eval

public ModelPoint eval(ModelPointNumericInt arg)

Evaluate the model objective at this point. Update the point to contain the computed value and, if appropriate, derivatives.

Specified by:

eval in interface ModelObjective

assignParameters

public void assignParameters(double[] pars)

Assign the parameters to the evaluator's database.

assignParameters

public void assignParameters(ModelPointNumericInt arg)

Assign the parameters from the model point to the evaluator's database.

Specified by:

assignParameters in interface ModelObjective

getValue

public double getValue(ModelPointNumericInt x)

Description copied from interface: ModelObjective

Retreive the return the computed value of the objective, and store it in the ModelPoint argument, if that is non-null.

Specified by:

getValue in interface ModelObjective

getGradient

public NumericArray getGradient(ModelPointNumericInt x)

Description copied from interface: ModelObjective

Retreive the return the computed gradient of the objective, and store it in the ModelPoint argument, if that is non-null.

By convention, returns null if the gradient is not being computed.

Specified by:

getGradient in interface ModelObjective

getHessian

public NumericMatrix getHessian(ModelPointNumericInt x)

Description copied from interface: ModelObjective

Retreive the return the computed hessian of the objective, and store it in the ModelPoint argument, if that is non-null.

By convention, returns null if the hessian is not being computed.

Specified by:

getHessian in interface ModelObjective

collectValue

public double collectValue(ModelPointNumericInt x)

Retrieve the value from the database and assign it into the value property of the model point argument. Returns the value. The evaluation in the Omegaghat evaluator must have been done for this object, so this is essentially a utility method for subclasses that do not want to use the eval method.

collectGradient

public NumericArray collectGradient(ModelPointNumericInt x)

Retrieve the gradient from the database, if it is there and non-null. Assign it into the gradient property of the model point argument. Returns the gradient.

collectHessian

public NumericMatrix collectHessian(ModelPointNumericInt x)

Retrieve the hessian from the database, if it is there and non-null. Assign it into the hessian property of the model point argument. Returns the hessian.

If the model point has not been evaluated, first evaluate it.

handleError

public boolean handleError(ModelPointNumericInt current,
                           java.lang.Throwable e)

Handle an exception or error that occurred during the fitting.

The default handler just prints some error information and sets the value to NaN.