com.pawjaw.classification.crf.lmcbt.configurations
Class Configuration

java.lang.Object
  extended by com.pawjaw.classification.crf.lmcbt.configurations.Configuration
Direct Known Subclasses:
DefaultConfiguration

public abstract class Configuration
extends java.lang.Object

Single point for configuring all options for training and inference.


Field Summary
 int boosted_tree_threads
          Number of threads used in multithreaded tree boosting.
 int boosting_iterations_between_tests
          Boosting itereations to run between evaluating stopping criteria when validation data is provided for training.
 boolean cache_expanded_true_features
           
 int expanded_features
          Number of features in each individual element (aka point) including features incorporated through sliding window and previous point label.
 int label_count_excluding_start_label
          Number of labels for elements in the data not including the start pseudo-label.
 int label_count_including_start_label
          Number of labels for elements in the data including the start pseudo-label.
 int max_boosting_iterations
          Maximum number of boosted tree iterations.
 int max_leaves_per_tree
          Maximum leaves per boosted regression tree.
 int min_boosting_iterations
          Minimum number of boosting iterations regardless of validation data and min_relative_accuracy_improvement.
 double min_relative_accuracy_improvement
          Minimum relative accuracy improvement required to continue boosting when training includes validation data.
 int point_features
          Number of features in each individual element (aka point).
 double regression_tree_shrinkage
          Shrinkage for regression tree smoothing.
 boolean report_training_progress
           
 int splitter_threads
          Number of threads used in multithreaded sufficient statistics collection portion of node splitting during regression tree construction.
 int valid_window_offset
          Offset into feature index where valid window features start.
 int window_radius
          One side of the sliding window (e.g.
 
Constructor Summary
Configuration(int point_features, int window_radius, int label_count_excluding_start_label)
           
 
Method Summary
protected abstract  int getBoostedTreeThreads()
           
protected abstract  int getBoostingIterationsBetweenTests()
           
protected abstract  boolean getCacheExpandedTrueFeatures()
           
protected abstract  int getMaxBoostingIterations()
           
protected abstract  int getMaxLeavesPerTree()
           
protected abstract  int getMinBoostingIterations()
           
protected abstract  double getMinRelativeAccuracyImprovement()
           
protected abstract  double getRegressionTreeShrinkage()
           
protected abstract  boolean getReportTrainingProgress()
           
protected abstract  int getSplitterThreads()
           
 
Methods inherited from class java.lang.Object
clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait
 

Field Detail

window_radius

public final int window_radius
One side of the sliding window (e.g. window_radius = 3 means use features from previous 3 and next 3 elements in the sequence).


point_features

public final int point_features
Number of features in each individual element (aka point).


splitter_threads

public final int splitter_threads
Number of threads used in multithreaded sufficient statistics collection portion of node splitting during regression tree construction.


expanded_features

public final int expanded_features
Number of features in each individual element (aka point) including features incorporated through sliding window and previous point label.


valid_window_offset

public final int valid_window_offset
Offset into feature index where valid window features start.


max_leaves_per_tree

public final int max_leaves_per_tree
Maximum leaves per boosted regression tree.


boosted_tree_threads

public final int boosted_tree_threads
Number of threads used in multithreaded tree boosting. Can utilize as many simultaneous threads as the value in label_count_including_start_label.


max_boosting_iterations

public final int max_boosting_iterations
Maximum number of boosted tree iterations. May be less if training includes validation data and min_relative_accuracy_improvement is greater than zero.


min_boosting_iterations

public final int min_boosting_iterations
Minimum number of boosting iterations regardless of validation data and min_relative_accuracy_improvement.


boosting_iterations_between_tests

public final int boosting_iterations_between_tests
Boosting itereations to run between evaluating stopping criteria when validation data is provided for training.


label_count_excluding_start_label

public final int label_count_excluding_start_label
Number of labels for elements in the data not including the start pseudo-label.


label_count_including_start_label

public final int label_count_including_start_label
Number of labels for elements in the data including the start pseudo-label.


regression_tree_shrinkage

public final double regression_tree_shrinkage
Shrinkage for regression tree smoothing. Typical range is 0 to 100.


min_relative_accuracy_improvement

public final double min_relative_accuracy_improvement
Minimum relative accuracy improvement required to continue boosting when training includes validation data. Typical range is 1e-3 to 1e-4.


report_training_progress

public final boolean report_training_progress

cache_expanded_true_features

public final boolean cache_expanded_true_features
Constructor Detail

Configuration

public Configuration(int point_features,
                     int window_radius,
                     int label_count_excluding_start_label)
Method Detail

getSplitterThreads

protected abstract int getSplitterThreads()

getMaxLeavesPerTree

protected abstract int getMaxLeavesPerTree()

getBoostedTreeThreads

protected abstract int getBoostedTreeThreads()

getMaxBoostingIterations

protected abstract int getMaxBoostingIterations()

getMinBoostingIterations

protected abstract int getMinBoostingIterations()

getBoostingIterationsBetweenTests

protected abstract int getBoostingIterationsBetweenTests()

getRegressionTreeShrinkage

protected abstract double getRegressionTreeShrinkage()

getMinRelativeAccuracyImprovement

protected abstract double getMinRelativeAccuracyImprovement()

getReportTrainingProgress

protected abstract boolean getReportTrainingProgress()

getCacheExpandedTrueFeatures

protected abstract boolean getCacheExpandedTrueFeatures()