org.apache.datasketches.cpc.CpcSketch Maven / Gradle / Ivy
Show all versions of datasketches-java Show documentation
/*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
* regarding copyright ownership. The ASF licenses this file
* to you under the Apache License, Version 2.0 (the
* "License"); you may not use this file except in compliance
* with the License. You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing,
* software distributed under the License is distributed on an
* "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
* KIND, either express or implied. See the License for the
* specific language governing permissions and limitations
* under the License.
*/
package org.apache.datasketches.cpc;
import static java.lang.Math.log;
import static java.lang.Math.sqrt;
import static java.nio.charset.StandardCharsets.UTF_8;
import static org.apache.datasketches.Util.DEFAULT_UPDATE_SEED;
import static org.apache.datasketches.Util.checkSeedHashes;
import static org.apache.datasketches.Util.computeSeedHash;
import static org.apache.datasketches.Util.invPow2;
import static org.apache.datasketches.Util.zeroPad;
import static org.apache.datasketches.cpc.CpcUtil.bitMatrixOfSketch;
import static org.apache.datasketches.cpc.CpcUtil.checkLgK;
import static org.apache.datasketches.cpc.CpcUtil.countBitsSetInMatrix;
import static org.apache.datasketches.hash.MurmurHash3.hash;
import org.apache.datasketches.Family;
import org.apache.datasketches.memory.Memory;
import org.apache.datasketches.memory.WritableMemory;
/**
* This is a unique-counting sketch that implements the
* Compressed Probabilistic Counting (CPC, a.k.a FM85) algorithms developed by Kevin Lang in
* his paper
* Back to the Future: an Even More Nearly
* Optimal Cardinality Estimation Algorithm.
*
* This sketch is extremely space-efficient when serialized. In an apples-to-apples empirical
* comparison against compressed HyperLogLog sketches, this new algorithm simultaneously wins on
* the two dimensions of the space/accuracy tradeoff and produces sketches that are
* smaller than the entropy of HLL, so no possible implementation of compressed HLL can match its
* space efficiency for a given accuracy. As described in the paper this sketch implements a newly
* developed ICON estimator algorithm that survives unioning operations, another
* well-known estimator, the
* Historical Inverse Probability (HIP) estimator
* does not.
* The update speed performance of this sketch is quite fast and is comparable to the speed of HLL.
* The unioning (merging) capability of this sketch also allows for merging of sketches with
* different configurations of K.
*
*
For additional security this sketch can be configured with a user-specified hash seed.
*
* @author Lee Rhodes
* @author Kevin Lang
*/
public final class CpcSketch {
private static final String LS = System.getProperty("line.separator");
private static final double[] kxpByteLookup = new double[256];
/**
* The default Log_base2 of K
*/
public static final int DEFAULT_LG_K = 11;
final long seed;
//common variables
final int lgK;
long numCoupons; // The number of coupons collected so far.
boolean mergeFlag; // Is the sketch the result of merging?
int fiCol; // First Interesting Column. This is part of a speed optimization.
int windowOffset;
byte[] slidingWindow; //either null or size K bytes
PairTable pairTable; //for sparse and surprising values, either null or variable size
//The following variables are only valid in HIP varients
double kxp; //used with HIP
double hipEstAccum; //used with HIP
/**
* Constructor with default log_base2 of k
*/
public CpcSketch() {
this(DEFAULT_LG_K, DEFAULT_UPDATE_SEED);
}
/**
* Constructor with log_base2 of k.
* @param lgK the given log_base2 of k
*/
public CpcSketch(final int lgK) {
this(lgK, DEFAULT_UPDATE_SEED);
}
/**
* Constructor with log_base2 of k and seed.
* @param lgK the given log_base2 of k
* @param seed the given seed
*/
public CpcSketch(final int lgK, final long seed) {
checkLgK(lgK);
this.lgK = (byte) lgK;
this.seed = seed;
kxp = 1 << lgK;
reset();
}
/**
* Returns a copy of this sketch
* @return a copy of this sketch
*/
CpcSketch copy() {
final CpcSketch copy = new CpcSketch(lgK, seed);
copy.numCoupons = numCoupons;
copy.mergeFlag = mergeFlag;
copy.fiCol = fiCol;
copy.windowOffset = windowOffset;
copy.slidingWindow = (slidingWindow == null) ? null : slidingWindow.clone();
copy.pairTable = (pairTable == null) ? null : pairTable.copy();
copy.kxp = kxp;
copy.hipEstAccum = hipEstAccum;
return copy;
}
/**
* Returns the best estimate of the cardinality of the sketch.
* @return the best estimate of the cardinality of the sketch.
*/
public double getEstimate() {
if (mergeFlag) { return IconEstimator.getIconEstimate(lgK, numCoupons); }
return hipEstAccum;
}
/**
* Return the DataSketches identifier for this CPC family of sketches.
* @return the DataSketches identifier for this CPC family of sketches.
*/
public static Family getFamily() {
return Family.CPC;
}
/**
* Return the parameter LgK.
* @return the parameter LgK.
*/
public int getLgK() {
return lgK;
}
/**
* Returns the best estimate of the lower bound of the confidence interval given kappa,
* the number of standard deviations from the mean.
* @param kappa the given number of standard deviations from the mean: 1, 2 or 3.
* @return the best estimate of the lower bound of the confidence interval given kappa.
*/
public double getLowerBound(final int kappa) {
if (mergeFlag) {
return CpcConfidence.getIconConfidenceLB(lgK, numCoupons, kappa);
}
return CpcConfidence.getHipConfidenceLB(lgK, numCoupons, hipEstAccum, kappa);
}
/*
* These empirical values for the 99.9th percentile of size in bytes were measured using 100,000
* trials. The value for each trial is the maximum of 5*16=80 measurements that were equally
* spaced over values of the quantity C/K between 3.0 and 8.0. This table does not include the
* worst-case space for the preamble, which is added by the function.
*/
private static final int[] empiricalMaxBytes = {
24, // lgK = 4
36, // lgK = 5
56, // lgK = 6
100, // lgK = 7
180, // lgK = 8
344, // lgK = 9
660, // lgK = 10
1292, // lgK = 11
2540, // lgK = 12
5020, // lgK = 13
9968, // lgK = 14
19836, // lgK = 15
39532, // lgK = 16
78880, // lgK = 17
157516, // lgK = 18
314656 // lgK = 19
};
/**
* The actual size of a compressed CPC sketch has a small random variance, but the following
* empirically measured size should be large enough for at least 99.9 percent of sketches.
*
*
For small values of n the size can be much smaller.
*
* @param lgK the given value of lgK.
* @return the estimated maximum compressed serialized size of a sketch.
*/
public static int getMaxSerializedBytes(final int lgK) {
checkLgK(lgK);
if (lgK <= 19) { return empiricalMaxBytes[lgK - 4] + 40; }
final int k = 1 << lgK;
return (int) (0.6 * k) + 40; // 0.6 = 4.8 / 8.0
}
/**
* Returns the best estimate of the upper bound of the confidence interval given kappa,
* the number of standard deviations from the mean.
* @param kappa the given number of standard deviations from the mean: 1, 2 or 3.
* @return the best estimate of the upper bound of the confidence interval given kappa.
*/
public double getUpperBound(final int kappa) {
if (mergeFlag) {
return CpcConfidence.getIconConfidenceUB(lgK, numCoupons, kappa);
}
return CpcConfidence.getHipConfidenceUB(lgK, numCoupons, hipEstAccum, kappa);
}
/**
* Return the given Memory as a CpcSketch on the Java heap using the DEFAULT_UPDATE_SEED.
* @param mem the given Memory
* @return the given Memory as a CpcSketch on the Java heap.
*/
public static CpcSketch heapify(final Memory mem) {
return heapify(mem, DEFAULT_UPDATE_SEED);
}
/**
* Return the given byte array as a CpcSketch on the Java heap using the DEFAULT_UPDATE_SEED.
* @param byteArray the given byte array
* @return the given byte array as a CpcSketch on the Java heap.
*/
public static CpcSketch heapify(final byte[] byteArray) {
return heapify(byteArray, DEFAULT_UPDATE_SEED);
}
/**
* Return the given Memory as a CpcSketch on the Java heap.
* @param mem the given Memory
* @param seed the seed used to create the original sketch from which the Memory was derived.
* @return the given Memory as a CpcSketch on the Java heap.
*/
public static CpcSketch heapify(final Memory mem, final long seed) {
final CompressedState state = CompressedState.importFromMemory(mem);
return uncompress(state, seed);
}
/**
* Return the given byte array as a CpcSketch on the Java heap.
* @param byteArray the given byte array
* @param seed the seed used to create the original sketch from which the byte array was derived.
* @return the given byte array as a CpcSketch on the Java heap.
*/
public static CpcSketch heapify(final byte[] byteArray, final long seed) {
final Memory mem = Memory.wrap(byteArray);
return heapify(mem, seed);
}
/**
* Return true if this sketch is empty
* @return true if this sketch is empty
*/
public boolean isEmpty() {
return numCoupons == 0;
}
/**
* Resets this sketch to empty but retains the original LgK and Seed.
*/
public final void reset() {
numCoupons = 0;
mergeFlag = false;
fiCol = 0;
windowOffset = 0;
slidingWindow = null;
pairTable = null;
kxp = 1 << lgK;
hipEstAccum = 0;
}
/**
* Return this sketch as a compressed byte array.
* @return this sketch as a compressed byte array.
*/
public byte[] toByteArray() {
final CompressedState state = CompressedState.compress(this);
final long cap = state.getRequiredSerializedBytes();
final WritableMemory wmem = WritableMemory.allocate((int) cap);
state.exportToMemory(wmem);
return (byte[]) wmem.getArray();
}
/**
* Present the given long as a potential unique item.
*
* @param datum The given long datum.
*/
public void update(final long datum) {
final long[] data = { datum };
final long[] arr = hash(data, seed);
hashUpdate(arr[0], arr[1]);
}
/**
* Present the given double (or float) datum as a potential unique item.
* The double will be converted to a long using Double.doubleToLongBits(datum),
* which normalizes all NaN values to a single NaN representation.
* Plus and minus zero will be normalized to plus zero.
* The special floating-point values NaN and +/- Infinity are treated as distinct.
*
* @param datum The given double datum.
*/
public void update(final double datum) {
final double d = (datum == 0.0) ? 0.0 : datum; // canonicalize -0.0, 0.0
final long[] data = { Double.doubleToLongBits(d) };// canonicalize all NaN forms
final long[] arr = hash(data, seed);
hashUpdate(arr[0], arr[1]);
}
/**
* Present the given String as a potential unique item.
* The string is converted to a byte array using UTF8 encoding.
* If the string is null or empty no update attempt is made and the method returns.
*
*
Note: About 2X faster performance can be obtained by first converting the String to a
* char[] and updating the sketch with that. This bypasses the complexity of the Java UTF_8
* encoding. This, of course, will not produce the same internal hash values as updating directly
* with a String. So be consistent! Unioning two sketches, one fed with strings and the other
* fed with char[] will be meaningless.
*
*
* @param datum The given String.
*/
public void update(final String datum) {
if ((datum == null) || datum.isEmpty()) { return; }
final byte[] data = datum.getBytes(UTF_8);
final long[] arr = hash(data, seed);
hashUpdate(arr[0], arr[1]);
}
/**
* Present the given byte array as a potential unique item.
* If the byte array is null or empty no update attempt is made and the method returns.
*
* @param data The given byte array.
*/
public void update(final byte[] data) {
if ((data == null) || (data.length == 0)) { return; }
final long[] arr = hash(data, seed);
hashUpdate(arr[0], arr[1]);
}
/**
* Present the given char array as a potential unique item.
* If the char array is null or empty no update attempt is made and the method returns.
*
* Note: this will not produce the same output hash values as the {@link #update(String)}
* method but will be a little faster as it avoids the complexity of the UTF8 encoding.
*
* @param data The given char array.
*/
public void update(final char[] data) {
if ((data == null) || (data.length == 0)) { return; }
final long[] arr = hash(data, seed);
hashUpdate(arr[0], arr[1]);
}
/**
* Present the given integer array as a potential unique item.
* If the integer array is null or empty no update attempt is made and the method returns.
*
* @param data The given int array.
*/
public void update(final int[] data) {
if ((data == null) || (data.length == 0)) { return; }
final long[] arr = hash(data, seed);
hashUpdate(arr[0], arr[1]);
}
/**
* Present the given long array as a potential unique item.
* If the long array is null or empty no update attempt is made and the method returns.
*
* @param data The given long array.
*/
public void update(final long[] data) {
if ((data == null) || (data.length == 0)) { return; }
final long[] arr = hash(data, seed);
hashUpdate(arr[0], arr[1]);
}
/**
* Convience function that this Sketch is valid. This is a troubleshooting tool
* for sketches that have been heapified from serialized images.
*
* If you are starting with a serialized image as a byte array, first heapify
* the byte array to a sketch, which performs a number of checks. Then use this
* function as one additional check on the sketch.
*
* @return true if this sketch is validated.
*/
public boolean validate() {
final long[] bitMatrix = bitMatrixOfSketch(this);
final long matrixCoupons = countBitsSetInMatrix(bitMatrix);
return matrixCoupons == numCoupons;
}
/**
* Returns the current Flavor of this sketch.
* @return the current Flavor of this sketch.
*/
Flavor getFlavor() {
return CpcUtil.determineFlavor(lgK, numCoupons);
}
/**
* Returns the Format of the serialized form of this sketch.
* @return the Format of the serialized form of this sketch.
*/
Format getFormat() {
final int ordinal;
final Flavor f = getFlavor();
if ((f == Flavor.HYBRID) || (f == Flavor.SPARSE)) {
ordinal = 2 | ( mergeFlag ? 0 : 1 ); //Hybrid is serialized as SPARSE
} else {
ordinal = ((slidingWindow != null) ? 4 : 0)
| (((pairTable != null) && (pairTable.getNumPairs() > 0)) ? 2 : 0)
| ( mergeFlag ? 0 : 1 );
}
return Format.ordinalToFormat(ordinal);
}
private static void promoteEmptyToSparse(final CpcSketch sketch) {
assert sketch.numCoupons == 0;
assert sketch.pairTable == null;
sketch.pairTable = new PairTable(2, 6 + sketch.lgK);
}
//In terms of flavor, this promotes SPARSE to HYBRID.
private static void promoteSparseToWindowed(final CpcSketch sketch) {
final int lgK = sketch.lgK;
final int k = (1 << lgK);
final long c32 = sketch.numCoupons << 5;
assert ((c32 == (3 * k)) || ((lgK == 4) && (c32 > (3 * k))));
final byte[] window = new byte[k];
final PairTable newTable = new PairTable(2, 6 + lgK);
final PairTable oldTable = sketch.pairTable;
final int[] oldSlots = oldTable.getSlotsArr();
final int oldNumSlots = (1 << oldTable.getLgSizeInts());
assert (sketch.windowOffset == 0);
for (int i = 0; i < oldNumSlots; i++) {
final int rowCol = oldSlots[i];
if (rowCol != -1) {
final int col = rowCol & 63;
if (col < 8) {
final int row = rowCol >>> 6;
window[row] |= (1 << col);
}
else {
// cannot use Table.mustInsert(), because it doesn't provide for growth
final boolean isNovel = PairTable.maybeInsert(newTable, rowCol);
assert (isNovel == true);
}
}
}
assert (sketch.slidingWindow == null);
sketch.slidingWindow = window;
sketch.pairTable = newTable;
}
/**
* The KXP register is a double with roughly 50 bits of precision, but
* it might need roughly 90 bits to track the value with perfect accuracy.
* Therefore we recalculate KXP occasionally from the sketch's full bitMatrix
* so that it will reflect changes that were previously outside the mantissa.
* @param sketch the given sketch
* @param bitMatrix the given bit Matrix
*/
//Also used in test
static void refreshKXP(final CpcSketch sketch, final long[] bitMatrix) {
final int k = (1 << sketch.lgK);
// for improved numerical accuracy, we separately sum the bytes of the U64's
final double[] byteSums = new double[8];
for (int j = 0; j < 8; j++) { byteSums[j] = 0.0; }
for (int i = 0; i < k; i++) {
long row = bitMatrix[i];
for (int j = 0; j < 8; j++) {
final int byteIdx = (int) (row & 0XFFL);
byteSums[j] += kxpByteLookup[byteIdx];
row >>>= 8;
}
}
double total = 0.0;
for (int j = 7; j-- > 0; ) { // the reverse order is important
final double factor = invPow2(8 * j); // pow(256, -j) == pow(2, -8 * j);
total += factor * byteSums[j];
}
sketch.kxp = total;
}
/**
* This moves the sliding window
* @param sketch the given sketch
* @param newOffset the new offset, which must be oldOffset + 1
*/
private static void modifyOffset(final CpcSketch sketch, final int newOffset) {
assert ((newOffset >= 0) && (newOffset <= 56));
assert (newOffset == (sketch.windowOffset + 1));
assert (newOffset == CpcUtil.determineCorrectOffset(sketch.lgK, sketch.numCoupons));
assert (sketch.slidingWindow != null);
assert (sketch.pairTable != null);
final int k = 1 << sketch.lgK;
// Construct the full-sized bit matrix that corresponds to the sketch
final long[] bitMatrix = CpcUtil.bitMatrixOfSketch(sketch);
// refresh the KXP register on every 8th window shift.
if ((newOffset & 0x7) == 0) { refreshKXP(sketch, bitMatrix); }
sketch.pairTable.clear();
final PairTable table = sketch.pairTable;
final byte[] window = sketch.slidingWindow;
final long maskForClearingWindow = (0XFFL << newOffset) ^ -1L;
final long maskForFlippingEarlyZone = (1L << newOffset) - 1L;
long allSurprisesORed = 0;
for (int i = 0; i < k; i++) {
long pattern = bitMatrix[i];
window[i] = (byte) ((pattern >>> newOffset) & 0XFFL);
pattern &= maskForClearingWindow;
// The following line converts surprising 0's to 1's in the "early zone",
// (and vice versa, which is essential for this procedure's O(k) time cost).
pattern ^= maskForFlippingEarlyZone;
allSurprisesORed |= pattern; // a cheap way to recalculate fiCol
while (pattern != 0) {
final int col = Long.numberOfTrailingZeros(pattern);
pattern = pattern ^ (1L << col); // erase the 1.
final int rowCol = (i << 6) | col;
final boolean isNovel = PairTable.maybeInsert(table, rowCol);
assert isNovel == true;
}
}
sketch.windowOffset = newOffset;
sketch.fiCol = Long.numberOfTrailingZeros(allSurprisesORed);
if (sketch.fiCol > newOffset) {
sketch.fiCol = newOffset; // corner case
}
}
/**
* Call this whenever a new coupon has been collected.
* @param sketch the given sketch
* @param rowCol the given row / column
*/
private static void updateHIP(final CpcSketch sketch, final int rowCol) {
final int k = 1 << sketch.lgK;
final int col = rowCol & 63;
final double oneOverP = k / sketch.kxp;
sketch.hipEstAccum += oneOverP;
sketch.kxp -= invPow2(col + 1); // notice the "+1"
}
private static void updateSparse(final CpcSketch sketch, final int rowCol) {
final int k = 1 << sketch.lgK;
final long c32pre = sketch.numCoupons << 5;
assert (c32pre < (3L * k)); // C < 3K/32, in other words, flavor == SPARSE
assert (sketch.pairTable != null);
final boolean isNovel = PairTable.maybeInsert(sketch.pairTable, rowCol);
if (isNovel) {
sketch.numCoupons += 1;
updateHIP(sketch, rowCol);
final long c32post = sketch.numCoupons << 5;
if (c32post >= (3L * k)) { promoteSparseToWindowed(sketch); } // C >= 3K/32
}
}
/**
* The flavor is HYBRID, PINNED, or SLIDING.
* @param sketch the given sketch
* @param rowCol the given rowCol
*/
private static void updateWindowed(final CpcSketch sketch, final int rowCol) {
assert ((sketch.windowOffset >= 0) && (sketch.windowOffset <= 56));
final int k = 1 << sketch.lgK;
final long c32pre = sketch.numCoupons << 5;
assert c32pre >= (3L * k); // C < 3K/32, in other words flavor >= HYBRID
final long c8pre = sketch.numCoupons << 3;
final int w8pre = sketch.windowOffset << 3;
assert c8pre < ((27L + w8pre) * k); // C < (K * 27/8) + (K * windowOffset)
boolean isNovel = false; //novel if new coupon
final int col = rowCol & 63;
if (col < sketch.windowOffset) { // track the surprising 0's "before" the window
isNovel = PairTable.maybeDelete(sketch.pairTable, rowCol); // inverted logic
}
else if (col < (sketch.windowOffset + 8)) { // track the 8 bits inside the window
assert (col >= sketch.windowOffset);
final int row = rowCol >>> 6;
final byte oldBits = sketch.slidingWindow[row];
final byte newBits = (byte) (oldBits | (1 << (col - sketch.windowOffset)));
if (newBits != oldBits) {
sketch.slidingWindow[row] = newBits;
isNovel = true;
}
}
else { // track the surprising 1's "after" the window
assert col >= (sketch.windowOffset + 8);
isNovel = PairTable.maybeInsert(sketch.pairTable, rowCol); // normal logic
}
if (isNovel) {
sketch.numCoupons += 1;
updateHIP(sketch, rowCol);
final long c8post = sketch.numCoupons << 3;
if (c8post >= ((27L + w8pre) * k)) {
modifyOffset(sketch, sketch.windowOffset + 1);
assert (sketch.windowOffset >= 1) && (sketch.windowOffset <= 56);
final int w8post = sketch.windowOffset << 3;
assert c8post < ((27L + w8post) * k); // C < (K * 27/8) + (K * windowOffset)
}
}
}
//also used in test
static CpcSketch uncompress(final CompressedState source, final long seed) {
checkSeedHashes(computeSeedHash(seed), source.seedHash);
final CpcSketch sketch = new CpcSketch(source.lgK, seed);
sketch.numCoupons = source.numCoupons;
sketch.windowOffset = source.getWindowOffset();
sketch.fiCol = source.fiCol;
sketch.mergeFlag = source.mergeFlag;
sketch.kxp = source.kxp;
sketch.hipEstAccum = source.hipEstAccum;
sketch.slidingWindow = null;
sketch.pairTable = null;
CpcCompression.uncompress(source, sketch);
return sketch;
}
//Used here and for testing
void hashUpdate(final long hash0, final long hash1) {
int col = Long.numberOfLeadingZeros(hash1);
if (col < fiCol) { return; } // important speed optimization
if (col > 63) { col = 63; } // clip so that 0 <= col <= 63
final long c = numCoupons;
if (c == 0) { promoteEmptyToSparse(this); }
final long k = 1L << lgK;
final int row = (int) (hash0 & (k - 1L));
int rowCol = (row << 6) | col;
// Avoid the hash table's "empty" value which is (2^26 -1, 63) (all ones) by changing it
// to the pair (2^26 - 2, 63), which effectively merges the two cells.
// This case is *extremely* unlikely, but we might as well handle it.
// It can't happen at all if lgK (or maxLgK) < 26.
if (rowCol == -1) { rowCol ^= (1 << 6); } //set the LSB of row to 0
if ((c << 5) < (3L * k)) { updateSparse(this, rowCol); }
else { updateWindowed(this, rowCol); }
}
//Used by union and in testing
void rowColUpdate(final int rowCol) {
final int col = rowCol & 63;
if (col < fiCol) { return; } // important speed optimization
final long c = numCoupons;
if (c == 0) { promoteEmptyToSparse(this); }
final long k = 1L << lgK;
if ((c << 5) < (3L * k)) { updateSparse(this, rowCol); }
else { updateWindowed(this, rowCol); }
}
/**
* Return a human-readable string summary of this sketch
*/
@Override
public String toString() {
return toString(false);
}
/**
* Return a human-readable string summary of this sketch
* @param detail include data detail
* @return a human-readable string summary of this sketch
*/
public String toString(final boolean detail) {
final int numPairs = (pairTable == null) ? 0 : pairTable.getNumPairs();
final int seedHash = Short.toUnsignedInt(computeSeedHash(seed));
final double errConst = mergeFlag ? log(2) : sqrt(log(2) / 2.0);
final double rse = errConst / Math.sqrt(1 << lgK);
final StringBuilder sb = new StringBuilder();
sb.append("### CPD SKETCH - PREAMBLE:").append(LS);
sb.append(" Flavor : ").append(getFlavor()).append(LS);
sb.append(" LgK : ").append(lgK).append(LS);
sb.append(" Merge Flag : ").append(mergeFlag).append(LS);
sb.append(" Error Const : ").append(errConst).append(LS);
sb.append(" RSE : ").append(rse).append(LS);
sb.append(" Seed Hash : ").append(Integer.toHexString(seedHash))
.append(" | ").append(seedHash).append(LS);
sb.append(" Num Coupons : ").append(numCoupons).append(LS);
sb.append(" Num Pairs (SV) : ").append(numPairs).append(LS);
sb.append(" First Inter Col: ").append(fiCol).append(LS);
sb.append(" Valid Window : ").append(slidingWindow != null).append(LS);
sb.append(" Valid PairTable: ").append(pairTable != null).append(LS);
sb.append(" Window Offset : ").append(windowOffset).append(LS);
sb.append(" KxP : ").append(kxp).append(LS);
sb.append(" HIP Accum : ").append(hipEstAccum).append(LS);
if (detail) {
sb.append(LS).append("### CPC SKETCH - DATA").append(LS);
if (pairTable != null) {
sb.append(pairTable.toString(true));
}
if (slidingWindow != null) {
sb.append("SlidingWindow : ").append(LS);
sb.append(" Index Bits (lsb ->)").append(LS);
for (int i = 0; i < slidingWindow.length; i++) {
final String bits = zeroPad(Integer.toBinaryString(slidingWindow[i] & 0XFF), 8);
sb.append(String.format("%9d %8s" + LS, i, bits));
}
}
}
sb.append("### END CPC SKETCH");
return sb.toString();
}
/**
* Returns a human readable string of the preamble of a byte array image of a CpcSketch.
* @param byteArr the given byte array
* @param detail if true, a dump of the compressed window and surprising value streams will be
* included.
* @return a human readable string of the preamble of a byte array image of a CpcSketch.
*/
public static String toString(final byte[] byteArr, final boolean detail) {
return PreambleUtil.toString(byteArr, detail);
}
/**
* Returns a human readable string of the preamble of a Memory image of a CpcSketch.
* @param mem the given Memory
* @param detail if true, a dump of the compressed window and surprising value streams will be
* included.
* @return a human readable string of the preamble of a Memory image of a CpcSketch.
*/
public static String toString(final Memory mem, final boolean detail) {
return PreambleUtil.toString(mem, detail);
}
private static void fillKxpByteLookup() { //called from static initializer
for (int b = 0; b < 256; b++) {
double sum = 0;
for (int col = 0; col < 8; col++) {
final int bit = (b >>> col) & 1;
if (bit == 0) { // note the inverted logic
sum += invPow2(col + 1); //note the "+1"
}
}
kxpByteLookup[b] = sum;
}
}
static {
fillKxpByteLookup();
}
}