AdaptiveScalarizingEntryRanker (magpie API)

java.lang.Object
- magpie.optimization.rankers.BaseEntryRanker
- - magpie.optimization.rankers.MultiObjectiveEntryRanker
  - - magpie.optimization.rankers.AdaptiveScalarizingEntryRanker

All Implemented Interfaces:

java.lang.Cloneable, Options
```
public class AdaptiveScalarizingEntryRanker
extends MultiObjectiveEntryRanker
```
Transforms the values of multiple objectives to a single variable using automatically-determined weights. Employs a strategy attributed to Wierzbicki, where the goal is to minimize the function: max_i[(f_i(x)- z_i^ideal) / (z_i^worst - z_i^ideal)] + p / N * sum[(f_i(x)- z_i^ideal) / (z_i^worst - z_i^ideal)] where f is the vector objective function, z^ideal/z^worst is the best/worst solutions found so far for a certain objective, p is a tradeoff parameter, and N is the number of objectives.
Note: Rather than using the maximum and minimum values of each objective to mark the best/worst this code uses the 99% and 1% percentiles. This is used to prevent a single outlier from drastically effecting the normalized values of each objective function.
Usage: <p> -opt <maximize|minimize> <property> <ranker name> [<ranker options...>] [-opt <...>]
p: Trade-off parameter between favoring entries that are best in a single category, and those that are good in many (default=1.0)
property: Name of property to be optimized using this ranker
maximize|minimize: Whether the goal is to minimize this objective function
ranker method: Name of an BaseEntryRanker. Avoid using another multi-objective ranker
ranker options: Any options for that entry ranker
The "-opt" flag can be used multiple times, and the syntax for each additional flag is identical. Also, this function is designed to be minimized.

Field Summary

Fields
Modifier and Type	Field and Description
`protected java.util.SortedMap<java.lang.String,BaseEntryRanker>`	`ObjectiveFunction` Map of property name to objective function
`protected double[]`	`ObjectiveMaximum` Maximum value of each objective function in training data
`protected double[]`	`ObjectiveMinimum` Minimum value of each objective function in training data
`protected double`	`P` Trade-off between best in single objective, and generally-good
`protected int[]`	`PropertyIndex` Index of each property of interest

Constructor Summary

Constructors
Constructor and Description

AdaptiveScalarizingEntryRanker()

Constructors
Constructor and Description
`AdaptiveScalarizingEntryRanker()`

Method Summary

All Methods Instance Methods Concrete Methods
Modifier and Type	Method and Description
`void`	`addObjectiveFunction(java.lang.String property, BaseEntryRanker function)` Define a new objective function.
`void`	`clearObjectiveFunctions()` Clear out list of currently-defined objective functions
`AdaptiveScalarizingEntryRanker`	`clone()`
`BaseEntryRanker`	`getObjectiveFunction(java.lang.String property)` Get objective function for a certain property.
`double`	`getObjectivePercentile()`
`java.lang.String[]`	`getObjectives()` Get properties currently used in objective functions.
`double`	`getP()` Get tradeoff parameter.
`double`	`objectiveFunction(BaseEntry Entry)` Some kind of objective function that returns a double when given an entry.
`java.lang.String`	`printUsage()` Print out required format for options.
`void`	`setObjectivePercentile(double x)` Define percentile to use for approximating the minimum value of each objective function.
`void`	`setOptions(java.util.List<java.lang.Object> Options)` Set any options for this object.
`void`	`setP(double P)` Set tradeoff parameter between optimal in single category and decent in all
`void`	`setUseMeasured(boolean useMeasured)` Set whether to used measured (or predicted) class values
`void`	`train(MultiPropertyDataset data)` Train the entry ranker.

Methods inherited from class magpie.optimization.rankers.MultiObjectiveEntryRanker
train

Methods inherited from class magpie.optimization.rankers.BaseEntryRanker
isMaximizing, isUsingMeasured, rankEntries, rankEntries, rankEntries, runObjectiveFunction, setMaximizeFunction, sortByRanking

Methods inherited from class java.lang.Object
equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

- Field Detail
  - P
```
protected double P
```
    Trade-off between best in single objective, and generally-good
  - ObjectiveFunction
```
protected java.util.SortedMap<java.lang.String,BaseEntryRanker> ObjectiveFunction
```
    Map of property name to objective function
  - ObjectiveMaximum
```
protected double[] ObjectiveMaximum
```
    Maximum value of each objective function in training data
  - ObjectiveMinimum
```
protected double[] ObjectiveMinimum
```
    Minimum value of each objective function in training data
  - PropertyIndex
```
protected int[] PropertyIndex
```
    Index of each property of interest
- Constructor Detail
  - AdaptiveScalarizingEntryRanker
```
public AdaptiveScalarizingEntryRanker()
```
- Method Detail
  - clone
```
public AdaptiveScalarizingEntryRanker clone()
```
    Overrides:
    
    clone in class MultiObjectiveEntryRanker
  - setOptions
```
public void setOptions(java.util.List<java.lang.Object> Options)
                throws java.lang.Exception
```
    Description copied from interface: Options
    
    Set any options for this object. If parsing the options fails, return an Exception with the usage as its message.
    
    Parameters:
    
    Options - Array of options as Objects - can be null
    
    Throws:
    
    java.lang.Exception - if problem with inputs
  - printUsage
```
public java.lang.String printUsage()
```
    Description copied from interface: Options
    
    Print out required format for options. For consistency, should start with "Usage: ".
    
    Returns:
    
    Usage for this function
  - setUseMeasured
```
public void setUseMeasured(boolean useMeasured)
```
    Description copied from class: BaseEntryRanker
    
    Set whether to used measured (or predicted) class values
    
    Overrides:
    
    setUseMeasured in class BaseEntryRanker
    
    Parameters:
    
    useMeasured - Desired option
  - setObjectivePercentile
```
public void setObjectivePercentile(double x)
```
    Define percentile to use for approximating the minimum value of each objective function. This is to provide some measure of robustness against an outlier objective function value from seriously affecting how the function values are normalized.
    The maximum value is approximated as the (100 - x)% percentile
    
    Parameters:
    
    x - Desired percentile
  - getObjectivePercentile
```
public double getObjectivePercentile()
```
    Returns:
    
    Percentile value used to approximate minimum/maximum of an objective function
  - getP
```
public double getP()
```
    Get tradeoff parameter.
    
    Returns:
    
    Current tradeoff
  - setP
```
public void setP(double P)
```
    Set tradeoff parameter between optimal in single category and decent in all
    
    Parameters:
    
    P - Desired tradeoff (should be positive) [default = 1]
  - clearObjectiveFunctions
```
public void clearObjectiveFunctions()
```
    Clear out list of currently-defined objective functions
  - addObjectiveFunction
```
public void addObjectiveFunction(java.lang.String property,
                                 BaseEntryRanker function)
```
    Define a new objective function. Order in which you add these does not matter
    
    Parameters:
    
    property - Name of property to be optimized
    
    function - Objective function for that property
  - getObjectiveFunction
```
public BaseEntryRanker getObjectiveFunction(java.lang.String property)
```
    Get objective function for a certain property.
    
    Parameters:
    
    property - Name of property
    
    Returns:
    
    Objective function (null if not defined)
  - getObjectives
```
public java.lang.String[] getObjectives()
```
    Description copied from class: MultiObjectiveEntryRanker
    
    Get properties currently used in objective functions.
    
    Specified by:
    
    getObjectives in class MultiObjectiveEntryRanker
    
    Returns:
    
    Array of properties currently in use
  - train
```
public void train(MultiPropertyDataset data)
```
    Description copied from class: MultiObjectiveEntryRanker
    
    Train the entry ranker. This may be necessary for multi-objective methods that rely on the performance of other entries (e.g. Pareto fronts).
    Note: This might be an opportune operation to map the name of each property with their index to allow for faster operation of the EntryRanker#objectiveFunction(magpie.data.BaseEntry)
    
    Specified by:
    
    train in class MultiObjectiveEntryRanker
    
    Parameters:
    
    data - Dataset to use for training
  - objectiveFunction
```
public double objectiveFunction(BaseEntry Entry)
```
    Description copied from class: BaseEntryRanker
    
    Some kind of objective function that returns a double when given an entry.
    
    NOTE: This function needs to support using either the measured or predicted class depending on the settings
    
    Specified by:
    
    objectiveFunction in class BaseEntryRanker
    
    Parameters:
    
    Entry - Entry to be analyzed
    
    Returns:
    
    Evaluation of target function

Class AdaptiveScalarizingEntryRanker

Field Summary

Constructor Summary

Method Summary

Methods inherited from class magpie.optimization.rankers.MultiObjectiveEntryRanker

Methods inherited from class magpie.optimization.rankers.BaseEntryRanker

Methods inherited from class java.lang.Object

Field Detail

P

ObjectiveFunction

ObjectiveMaximum

ObjectiveMinimum

PropertyIndex

Constructor Detail

AdaptiveScalarizingEntryRanker

Method Detail

clone

setOptions

printUsage

setUseMeasured

setObjectivePercentile

getObjectivePercentile

getP

setP

clearObjectiveFunctions

addObjectiveFunction

getObjectiveFunction

getObjectives

train

objectiveFunction