jidt/java/source/infodynamics/measures/discrete/BlockEntropyCalculatorDiscrete.java

/*
 *  Java Information Dynamics Toolkit (JIDT)
 *  Copyright (C) 2012, Joseph T. Lizier
 *  
 *  This program is free software: you can redistribute it and/or modify
 *  it under the terms of the GNU General Public License as published by
 *  the Free Software Foundation, either version 3 of the License, or
 *  (at your option) any later version.
 *  
 *  This program is distributed in the hope that it will be useful,
 *  but WITHOUT ANY WARRANTY; without even the implied warranty of
 *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 *  GNU General Public License for more details.
 *  
 *  You should have received a copy of the GNU General Public License
 *  along with this program.  If not, see <http://www.gnu.org/licenses/>.
 */

package infodynamics.measures.discrete;

import infodynamics.utils.MathsUtils;
import infodynamics.utils.MatrixUtils;

/**
 * <p>Block entropy calculator for univariate discrete (int[]) data
 * (ie computes entropy over blocks of consecutive states in time).
 * Implemented separately from the single state entropy 
 *  calculator to allow the single state calculator 
 *  to have optimal performance.</p>
 *
 * <p>Usage of the class is intended to follow this paradigm:</p>
 * <ol>
 * 		<li>Construct the calculator: {@link #BlockEntropyCalculatorDiscrete(int, int)};</li>
 *		<li>Initialise the calculator using {@link #initialise()};</li>
 * 		<li>Provide the observations/samples for the calculator
 *      	to set up the PDFs, using one or more calls to
 * 			sets of {@link #addObservations(int[])} methods, then</li>
 * 		<li>Compute the required quantities, being one or more of:
 * 			<ul>
 * 				<li>the average entropy: {@link #computeAverageLocalOfObservations()};</li>
 * 				<li>local entropy values, such as {@link #computeLocal(int[])};</li>
 * 				<li>and variants of these.</li>
 * 			</ul>
 * 		</li>
 * 		<li>As an alternative to steps 3 and 4, the user may undertake
 * 			standalone computation from a single set of observations, via
 *  		e.g.: {@link #computeLocal(int[])},
 *  		{@link #computeAverageLocal(int[])} etc.</li>
 * 		<li>
 * 		Return to step 2 to re-use the calculator on a new data set.
 * 		</li>
 * 	</ol>
 * 
 * <p><b>References:</b><br/>
 * <ul>
 * 	<li>T. M. Cover and J. A. Thomas, 'Elements of Information
Theory' (John Wiley & Sons, New York, 1991).</li>
 * </ul>
 * 
 * @author Joseph Lizier (<a href="joseph.lizier at gmail.com">email</a>,
 * <a href="http://lizier.me/joseph/">www</a>)
 */
public class BlockEntropyCalculatorDiscrete extends EntropyCalculatorDiscrete {

	/**
	 * Number of consecutive time-steps to compute entropy over.
	 */
	protected int blocksize = 0; //  Need initialised to 0 for changedSizes
	protected int[] maxShiftedValue = null; // states * (base^(blocksize-1))

	protected int base_power_blocksize = 0;

	/**
	 * User was formerly forced to create new instances through this factory method.
	 * Retained for backwards compatibility.
	 * 
	 * @param blocksize
	 * @param base
	 * @deprecated
	 */
	public static EntropyCalculatorDiscrete newInstance(int blocksize, int base) {
		return new BlockEntropyCalculatorDiscrete(blocksize, base);
	}
	
	/**
	 * Construct an instance with default blocksize 1 and base 2
	 */
	public BlockEntropyCalculatorDiscrete() {
		this(1, 2);
	}
	
	/**
	 * Construct a new instance
	 * 
	 * @param blocksize Number of consecutive time-steps to compute entropy over.
	 * @param base number of quantisation levels for each variable.
	 *        E.g. binary variables are in base-2.
	 */
	public BlockEntropyCalculatorDiscrete(int blocksize, int base) {

		super(base);
		resetBlocksize(blocksize);
	}
	
	private void resetBlocksize(int blocksize) {
		this.blocksize = blocksize;
		base_power_blocksize = MathsUtils.power(base, blocksize);

		if (blocksize <= 1) {
			throw new RuntimeException("Blocksize " + blocksize + " is not > 1 for Block Entropy Calculator");
		}
		if (blocksize > Math.log(Integer.MAX_VALUE) / log_base) {
			throw new RuntimeException("Base and blocksize combination too large");
		}
		
		// Create constants for tracking stateValues
		maxShiftedValue = new int[base];
		for (int v = 0; v < base; v++) {
			maxShiftedValue[v] = v * MathsUtils.power(base, blocksize-1);
		}
	}
	
	public void initialise(int blocksize, int base) {
		boolean baseOrBlocksizeChanged = (this.blocksize != blocksize) || (this.base != base);
		
		super.initialise(base);
		if (baseOrBlocksizeChanged) {
			resetBlocksize(blocksize);
		}
		
		if (baseOrBlocksizeChanged || (stateCount == null)) {
			// Create storage for counts of observations.
			// We're recreating stateCount, which was created in super()
			//  however the super() didn't create it for blocksize > 1
			try {
				stateCount = new int[MathsUtils.power(base, blocksize)];
			} catch (OutOfMemoryError e) {
				// Allow any Exceptions to be thrown, but catch and wrap
				//  Error as a Runtimexception
				throw new RuntimeException("Requested memory for the base " +
						base + " and k=" + blocksize+ " is too large for the JVM at this time", e);
			}
		} else {
			// Storage for sample counts exists and needs to be reset
			MatrixUtils.fill(stateCount, 0);
		}
	}
	
	@Override
	public void initialise() {
		initialise(blocksize, base);
	}
	
	@Override
	public void addObservations(int states[]) {
		int rows = states.length;
		// increment the count of observations:
		observations += (rows - blocksize + 1); 
		
		// Initialise and store the current previous value
		int stateVal = 0; 
		for (int p = 0; p < blocksize - 1; p++) {
			// Add the contribution from this observation
			stateVal += states[p];
			// And shift up
			stateVal *= base;
		}

		// 1. Now count the tuples observed from the next row onwards
		for (int r = blocksize - 1; r < rows; r++) {
			// Add the contribution from this observation
			stateVal += states[r];

			// Add to the count for this particular state:
			stateCount[stateVal]++;
			
			// Remove the oldest observation from the state value
			stateVal -= maxShiftedValue[states[r-blocksize+1]];
			stateVal *= base;
		}		
	}
	
	@Override
	public void addObservations(int states[][]) {
		int rows = states.length;
		int columns = states[0].length;
		// increment the count of observations:
		observations += (rows - blocksize + 1)*columns; 
		
		// Initialise and store the current previous value for each column
		int[] stateVal = new int[columns]; 
		for (int c = 0; c < columns; c++) {
			stateVal[c] = 0;
			for (int p = 0; p < blocksize - 1; p++) {
				// Add the contribution from this observation
				stateVal[c] += states[p][c];
				// And shift up
				stateVal[c] *= base;
			}
		}

		// 1. Now count the tuples observed from the next row onwards
		for (int r = blocksize - 1; r < rows; r++) {
			for (int c = 0; c < columns; c++) {
				// Add the contribution from this observation
				stateVal[c] += states[r][c];

				// Add to the count for this particular state:
				stateCount[stateVal[c]]++;
				
				// Remove the oldest observation from the state value
				stateVal[c] -= maxShiftedValue[states[r-blocksize+1][c]];
				stateVal[c] *= base;
			}
		}		
	}
	
	@Override
	public void addObservations(int states[][][]) {
		int timeSteps = states.length;
		if (timeSteps == 0) {
			return;
		}
		int agentRows = states[0].length;
		if (agentRows == 0) {
			return;
		}
		int agentColumns = states[0][0].length;
		// increment the count of observations:
		observations += (timeSteps - blocksize + 1) * agentRows * agentColumns; 
		
		// Initialise and store the current previous value for each column
		int[][] stateVal = new int[agentRows][agentColumns];
		for (int r = 0; r < agentRows; r++) {
			for (int c = 0; c < agentColumns; c++) {
				stateVal[r][c] = 0;
				for (int p = 0; p < blocksize - 1; p++) {
					// Add the contribution from this observation
					stateVal[r][c] += states[p][r][c];
					// And shift up
					stateVal[r][c] *= base;
				}
			}
		}

		// 1. Now count the tuples observed from the next row onwards
		for (int t = blocksize - 1; t < timeSteps; t++) {
			for (int r = 0; r < agentRows; r++) {
				for (int c = 0; c < agentColumns; c++) {
					// Add the contribution from this observation
					stateVal[r][c] += states[t][r][c];
	
					// Add to the count for this particular state:
					stateCount[stateVal[r][c]]++;
					
					// Remove the oldest observation from the state value
					stateVal[r][c] -= maxShiftedValue[states[t-blocksize+1][r][c]];
					stateVal[r][c] *= base;
				}
			}
		}		
	}

	@Override
	public void addObservations(int states[][], int col) {
		int rows = states.length;
		// increment the count of observations:
		observations += rows - blocksize + 1; 
		
		// Initialise and store the current previous value for each column
		int stateVal = 0;
		for (int p = 0; p < blocksize - 1; p++) {
			// Add the contribution from this observation
			stateVal += states[p][col];
			// And shift up
			stateVal *= base;
		}

		// 1. Count the tuples observed
		for (int r = blocksize - 1; r < rows; r++) {
			// Add the contribution from this observation
			stateVal += states[r][col];

			// Add to the count for this particular state:
			stateCount[stateVal]++;
			
			// Remove the oldest observation from the state value
			stateVal -= maxShiftedValue[states[r-blocksize+1][col]];
			stateVal *= base;
		}
	}

	@Override
	public void addObservations(int states[][][], int agentIndex1, int agentIndex2) {
		int timeSteps = states.length;
		// increment the count of observations:
		observations += timeSteps - blocksize + 1; 
		
		// Initialise and store the current previous value for each column
		int stateVal = 0;
		for (int p = 0; p < blocksize - 1; p++) {
			// Add the contribution from this observation
			stateVal += states[p][agentIndex1][agentIndex2];
			// And shift up
			stateVal *= base;
		}

		// 1. Count the tuples observed
		for (int t = blocksize - 1; t < timeSteps; t++) {
			// Add the contribution from this observation
			stateVal += states[t][agentIndex1][agentIndex2];

			// Add to the count for this particular state:
			stateCount[stateVal]++;
			
			// Remove the oldest observation from the state value
			stateVal -= maxShiftedValue[states[t-blocksize+1][agentIndex1][agentIndex2]];
			stateVal *= base;
		}
	}

	@Override
	public double computeAverageLocalOfObservations() {
		double ent = 0.0;
		double entCont = 0.0;

		max = 0;
		min = 0;
		for (int stateVal = 0; stateVal < base_power_blocksize; stateVal++) {
			// compute p_state
			double p_state = (double) stateCount[stateVal] / (double) observations;
			if (p_state > 0.0) {
				// Entropy takes the negative log:
				double localValue = - Math.log(p_state) / log_2;
				entCont = p_state * localValue;
				if (localValue > max) {
					max = localValue;
				} else if (localValue < min) {
					min = localValue;
				}
			} else {
				entCont = 0.0;
			}
			ent += entCont;
		}
		
		average = ent;
		return ent;
	}
	
	@Override
	public double[][] computeLocalFromPreviousObservations(int states[][]){
		int rows = states.length;
		int columns = states[0].length;
		
		// Allocate for all rows even though we may not be assigning all of them
		double[][] localEntropy = new double[rows][columns];
		average = 0;
		max = 0;
		min = 0;
		
		// Initialise and store the current previous value for each column
		int[] stateVal = new int[columns]; 
		for (int c = 0; c < columns; c++) {
			stateVal[c] = 0;
			for (int p = 0; p < blocksize - 1; p++) {
				stateVal[c] += states[p][c];
				stateVal[c] *= base;
			}
		}
		// StateVal just needs the next value put in before processing

		for (int r = blocksize - 1; r < rows; r++) {
			for (int c = 0; c < columns; c++) {
				// Add the most recent observation to the state count
				stateVal[c] += states[r][c];
				// Process the stateVal:
				double p_state = (double) stateCount[stateVal[c]] / (double) observations;
				// Entropy takes the negative log:
				localEntropy[r][c] = - Math.log(p_state) / log_2;
				average += localEntropy[r][c];
				if (localEntropy[r][c] > max) {
					max = localEntropy[r][c];
				} else if (localEntropy[r][c] < min) {
					min = localEntropy[r][c];
				}
				// Subtract out the oldest part of the state value:
				stateVal[c] -= maxShiftedValue[states[r-blocksize+1][c]];
				// And shift all upwards
				stateVal[c] *= base;
			}
		}
		average = average/(double) (columns * (rows - blocksize + 1));
		
		return localEntropy;
		
	}
	
	@Override
	public double[][][] computeLocalFromPreviousObservations(int states[][][]){
		int timeSteps = states.length;
		int agentRows, agentColumns;
		if (timeSteps == 0) {
			agentRows = 0;
			agentColumns = 0;
		} else {
			agentRows = states[0].length;
			if (agentRows == 0) {
				agentColumns = 0;
			} else {
				agentColumns = states[0][0].length;
			}
		}
		
		// Allocate for all rows even though we may not be assigning all of them
		double[][][] localEntropy = new double[timeSteps][agentRows][agentColumns];
		average = 0;
		max = 0;
		min = 0;
		
		// Initialise and store the current previous value for each column
		int[][] stateVal = new int[agentRows][agentColumns]; 
		for (int r = 0; r < agentRows; r++) {
			for (int c = 0; c < agentColumns; c++) {
				stateVal[r][c] = 0;
				for (int p = 0; p < blocksize - 1; p++) {
					stateVal[r][c] += states[p][r][c];
					stateVal[r][c] *= base;
				}
			}
		}
		// StateVal just needs the next value put in before processing

		for (int t = blocksize - 1; t < timeSteps; t++) {
			for (int r = 0; r < agentRows; r++) {
				for (int c = 0; c < agentColumns; c++) {
					// Add the most recent observation to the state count
					stateVal[r][c] += states[t][r][c];
					// Process the stateVal:
					double p_state = (double) stateCount[stateVal[r][c]] / (double) observations;
					// Entropy takes the negative log:
					localEntropy[t][r][c] = - Math.log(p_state) / log_2;
					average += localEntropy[t][r][c];
					if (localEntropy[t][r][c] > max) {
						max = localEntropy[t][r][c];
					} else if (localEntropy[t][r][c] < min) {
						min = localEntropy[t][r][c];
					}
					// Subtract out the oldest part of the state value:
					stateVal[r][c] -= maxShiftedValue[states[t-blocksize+1][r][c]];
					// And shift all upwards
					stateVal[r][c] *= base;
				}
			}
		}
		average = average/(double) (agentRows * agentColumns * (timeSteps - blocksize + 1));
		
		return localEntropy;
		
	}

	@Override
	public double[] computeLocalFromPreviousObservations(int states[][], int col){
		int rows = states.length;
		//int columns = states[0].length;
		// First row to consider the destination cell at:
		int startRow = blocksize - 1;

		// Allocate for all rows even though we'll leave the first ones as zeros
		double[] localEntropy = new double[rows];
		average = 0;
		max = 0;
		min = 0;
		
		// Initialise and store the current previous value for each column
		int stateVal = 0; 
		for (int p = 0; p < blocksize - 1; p++) {
			stateVal += states[p][col];
			stateVal *= base;
		}
		// StateVal just needs the next value put in before processing

		for (int r = startRow; r < rows; r++) {
			// Add the most recent observation to the state count
			stateVal += states[r][col];
			// Process the stateVal:
			double p_state = (double) stateCount[stateVal] / (double) observations;
			// Entropy takes the negative log:
			localEntropy[r] = - Math.log(p_state) / log_2;
			average += localEntropy[r];
			if (localEntropy[r] > max) {
				max = localEntropy[r];
			} else if (localEntropy[r] < min) {
				min = localEntropy[r];
			}
			// Subtract out the oldest part of the state value:
			stateVal -= maxShiftedValue[states[r-blocksize+1][col]];
			// And shift all upwards
			stateVal *= base;
		}
		average = average/(double) (rows - blocksize + 1);
		
		return localEntropy;
		
	}
	
	@Override
	public double[] computeLocalFromPreviousObservations(int states[][][], int agentIndex1, int agentIndex2){
		int timeSteps = states.length;
		//int columns = states[0].length;
		// First time to consider the destination cell at:
		int startTime = blocksize - 1;

		// Allocate for all rows even though we'll leave the first ones as zeros
		double[] localEntropy = new double[timeSteps];
		average = 0;
		max = 0;
		min = 0;
		
		// Initialise and store the current previous value for each column
		int stateVal = 0; 
		for (int p = 0; p < blocksize - 1; p++) {
			stateVal += states[p][agentIndex1][agentIndex2];
			stateVal *= base;
		}
		// StateVal just needs the next value put in before processing

		for (int t = startTime; t < timeSteps; t++) {
			// Add the most recent observation to the state count
			stateVal += states[t][agentIndex1][agentIndex2];
			// Process the stateVal:
			double p_state = (double) stateCount[stateVal] / (double) observations;
			// Entropy takes the negative log:
			localEntropy[t] = - Math.log(p_state) / log_2;
			average += localEntropy[t];
			if (localEntropy[t] > max) {
				max = localEntropy[t];
			} else if (localEntropy[t] < min) {
				min = localEntropy[t];
			}
			// Subtract out the oldest part of the state value:
			stateVal -= maxShiftedValue[states[t-blocksize+1][agentIndex1][agentIndex2]];
			// And shift all upwards
			stateVal *= base;
		}
		average = average/(double) (timeSteps - blocksize + 1);
		
		return localEntropy;
		
	}
}