boone.util

Class Patterns

java.lang.Object

boone.util.Patterns

public class Patterns
extends java.lang.Object

A class supporting pattern operations.

Since:

August 2021

Version:

$Id: DataSet.java 2271 2015-03-22 jmory $

Author:

Helmut A. Mayer

Constructor Summary

Constructors

Constructor and Description
Patterns()

Method Summary

Modifier and Type	Method and Description
static java.util.List<java.lang.Double>	encode(int label, int n, java.util.List<java.lang.Double> target) Encode a class value into a 1-of-n encoding, i.e., the classical encoding of an output layer for classification.
static int	getMaxIndex(java.util.List<java.util.List<java.lang.Double>> patterns, int index) Returns the index of the maximal value in a pattern.
static PatternSet	getSubSet(PatternSet set, int first, int last) Returns a subset of the given pattern set.
static void	join(PatternSet a, PatternSet b) Joins a data set to another data set.
static void	mapInputsToInterval(PatternSet set, double low, double high, double min, double max) Maps all input values to the given interval.
static void	mapTargetsToInterval(PatternSet set, double low, double high, double min, double max) Maps all target values to the given interval.
static void	mapToInterval(PatternSet set, double low, double high, double min, double max) Maps all input and target values to the given interval.
static void	multiply(PatternSet set, int mult) Multiplies the input vectors in a pattern set such that the input vector has 'mult' times the size.
static void	normalizeInputs(PatternSet set) Normalizes the input patterns using the euclidean norm, so all input vectors are changed to unit vectors.
static void	shuffle(PatternSet set, int n) Reorders the first n patterns in this set randomly.
static void	swap(PatternSet set, int pos1, int pos2) Swaps the two patterns at position pos1 and pos2 of the given set including names, inputs, targets, and outputs.

Methods inherited from class java.lang.Object

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

Constructor Detail

Patterns

public Patterns()

Method Detail

join

public static void join(PatternSet a,
                        PatternSet b)

Joins a data set to another data set. The other data set is appended to the first data set. Also copies those properties from the other set, whose keys are strings being different from the string keys in the first data set.

Parameters:

a - a pattern set

b - another pattern set

getSubSet

public static PatternSet getSubSet(PatternSet set,
                                   int first,
                                   int last)

Returns a subset of the given pattern set. The subset is a deep copy including the properties. If the number of input patterns does not allow a proper subset, null is returned. If targets, outputs or names cannot be split into a subset, e.g., there are no names, the returned data set has an empty list for the corresponding category.

Parameters:

set - a pattern set

first - the first pattern (included in the subset)

last - the last pattern (excluded from the subset)

Returns:

the subset, null if too few input patterns

shuffle

public static void shuffle(PatternSet set,
                           int n)

Reorders the first n patterns in this set randomly. After shuffling the first n patterns may be any patterns from the set.

Parameters:

set - a pattern set

n - the number of patterns to be shuffled

swap

public static void swap(PatternSet set,
                        int pos1,
                        int pos2)

Swaps the two patterns at position pos1 and pos2 of the given set including names, inputs, targets, and outputs.

Parameters:

set - a pattern set

pos1 - the first position

pos2 - the second position

normalizeInputs

public static void normalizeInputs(PatternSet set)

Normalizes the input patterns using the euclidean norm, so all input vectors are changed to unit vectors.

Parameters:

set - a pattern set

mapToInterval

public static void mapToInterval(PatternSet set,
                                 double low,
                                 double high,
                                 double min,
                                 double max)

Maps all input and target values to the given interval. The minimal feature value will be mapped to 'low. The maximal feature value will be mapped to 'high'.

Parameters:

set - a pattern set

low - the lower interval boundary

high - the higher interval boundary

min - the minimal feature value

max - the maximal feature value

mapInputsToInterval

public static void mapInputsToInterval(PatternSet set,
                                       double low,
                                       double high,
                                       double min,
                                       double max)

Maps all input values to the given interval. The minimal feature value will be mapped to 'low. The maximal feature value will be mapped to 'high'.

Parameters:

set - a pattern set

low - the lower interval boundary

high - the higher interval boundary

min - the minimal feature value

max - the maximal feature value

mapTargetsToInterval

public static void mapTargetsToInterval(PatternSet set,
                                        double low,
                                        double high,
                                        double min,
                                        double max)

Maps all target values to the given interval. The minimal value of a target in the set will be mapped to 'low. The maximal value of a target will be mapped to 'high'.

Parameters:

set - a pattern set

low - the lower interval boundary

high - the higher interval boundary

min - the minimal feature value

max - the maximal feature value

encode

public static java.util.List<java.lang.Double> encode(int label,
                                                      int n,
                                                      java.util.List<java.lang.Double> target)

Encode a class value into a 1-of-n encoding, i.e., the classical encoding of an output layer for classification.

Parameters:

label - the class label

n - the number of classes

target - the target vector (if null, a new one is created)

Returns:

the modified target vector

multiply

public static void multiply(PatternSet set,
                            int mult)

Multiplies the input vectors in a pattern set such that the input vector has 'mult' times the size. The values of the original input vector are simply copied, e.g., a vector [1, 0, 1] becomes [1, 0, 1, 1, 0, 1] for 'mult' = 2. This is useful for channel inputs of a CNN.

Parameters:

set - a pattern set

mult - the factor, e.g., 2 doubles the input vector

getMaxIndex

public static int getMaxIndex(java.util.List<java.util.List<java.lang.Double>> patterns,
                              int index)

Returns the index of the maximal value in a pattern.

Parameters:

patterns - a set of patterns

index - index of pattern

Returns:

the index of the maximal value