/* ---------------------------------------------------------------------
* Numenta Platform for Intelligent Computing (NuPIC)
* Copyright (C) 2014, Numenta, Inc. Unless you have an agreement
* with Numenta, Inc., for a separate license for this software code, the
* following terms and conditions apply:
*
* This program is free software: you can redistribute it and/or modify
* it under the terms of the GNU Affero Public License version 3 as
* published by the Free Software Foundation.
*
* 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 Affero Public License for more details.
*
* You should have received a copy of the GNU Affero Public License
* along with this program. If not, see http://www.gnu.org/licenses.
*
* http://numenta.org/licenses/
* ---------------------------------------------------------------------
*/
package org.numenta.nupic.algorithms;
import java.util.ArrayList;
import java.util.Arrays;
import java.util.List;
import org.joda.time.DateTime;
import org.numenta.nupic.algorithms.MovingAverage.Calculation;
import org.numenta.nupic.util.ArrayUtils;
import org.numenta.nupic.util.NamedTuple;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
import gnu.trove.list.TDoubleList;
import gnu.trove.list.array.TDoubleArrayList;
import gnu.trove.map.TObjectDoubleMap;
/**
* <p>The anomaly likelihood computer.</p>
*
* <p>From {@link Anomaly}:</p>
*
* This module analyzes and estimates the distribution of averaged anomaly scores
* from a CLA model. Given a new anomaly score `s`, estimates `P(score >= s)`.
*
* The number `P(score >= s)` represents the likelihood of the current state of
* predictability. For example, a likelihood of 0.01 or 1% means we see this much
* predictability about one out of every 100 records. The number is not as unusual
* as it seems. For records that arrive every minute, this means once every hour
* and 40 minutes. A likelihood of 0.0001 or 0.01% means we see it once out of
* 10,000 records, or about once every 7 days.
*
* <p>
* <em>
* USAGE FOR LOW-LEVEL FUNCTIONS
* -----------------------------
* </em>
* <pre>
* There are two primary interface routines:
*
* estimateAnomalyLikelihoods: batch routine, called initially and once in a
* while
* updateAnomalyLikelihoods: online routine, called for every new data point
*
* 1. Initially::
*
* {@link AnomalyLikelihoodMetrics} { likelihoods, avgRecordList, estimatorParams } =
* {@link #estimateAnomalyLikelihoods(List, int, int)}
*
* 2. Whenever you get new data::
*
* {@link AnomalyLikelihoodMetrics} { likelihoods, avgRecordList, estimatorParams } =
* {@link #updateAnomalyLikelihoods(List, NamedTuple)}
*
* 3. And again (make sure you use the new estimatorParams (a.k.a {@link NamedTuple}) returned
* in the above {@link AnomalyLikelihoodMetrics} call to updateAnomalyLikelihoods!)
*
* {@link AnomalyLikelihoodMetrics} { likelihoods, avgRecordList, estimatorParams } =
* {@link #updateAnomalyLikelihoods(List, NamedTuple)}
*
* 4. Every once in a while update estimator with a lot of recent data
*
* {@link AnomalyLikelihoodMetrics} { likelihoods, avgRecordList, estimatorParams } =
* {@link #estimateAnomalyLikelihoods(List, int, int)}
* </pre>
* </p>
*
* @author Numenta
* @author David Ray
* @see Anomaly
* @see AnomalyLikelihoodMetrics
* @see Sample
* @see AnomalyParams
* @see Statistic
* @see MovingAverage
*/
public class AnomalyLikelihood extends Anomaly {
private static final long serialVersionUID = 1L;
private static final Logger LOG = LoggerFactory.getLogger(AnomalyLikelihood.class);
private int claLearningPeriod = 300;
private int estimationSamples = 300;
private int probationaryPeriod;
private int iteration;
private int reestimationPeriod;
private boolean isWeighted;
private List<Sample> historicalScores = new ArrayList<>();
private AnomalyParams distribution;
public AnomalyLikelihood(boolean useMovingAvg, int windowSize, boolean isWeighted, int claLearningPeriod, int estimationSamples) {
super(useMovingAvg, windowSize);
this.isWeighted = isWeighted;
this.claLearningPeriod = claLearningPeriod == VALUE_NONE ? this.claLearningPeriod : claLearningPeriod;
this.estimationSamples = estimationSamples == VALUE_NONE ? this.estimationSamples : estimationSamples;
this.probationaryPeriod = this.claLearningPeriod + this.estimationSamples;
// How often we re-estimate the Gaussian distribution. The ideal is to
// re-estimate every iteration but this is a performance hit. In general the
// system is not very sensitive to this number as long as it is small
// relative to the total number of records processed.
this.reestimationPeriod = 100;
}
/**
* Compute a log scale representation of the likelihood value. Since the
* likelihood computations return low probabilities that often go into four 9's
* or five 9's, a log value is more useful for visualization, thresholding,
* etc.
*
* @param likelihood
* @return
*/
public static double computeLogLikelihood(double likelihood) {
return Math.log(1.0000000001 - likelihood) / -23.02585084720009;
}
/**
* Return the probability that the current value plus anomaly score represents
* an anomaly given the historical distribution of anomaly scores. The closer
* the number is to 1, the higher the chance it is an anomaly.
*
* Given the current metric value, plus the current anomaly score, output the
* anomalyLikelihood for this record.
*
* @param value input value
* @param anomalyScore current anomaly score
* @param timestamp (optional) timestamp
* @return Given the current metric value, plus the current anomaly score, output the
* anomalyLikelihood for this record.
*/
public double anomalyProbability(double value, double anomalyScore, DateTime timestamp) {
if(timestamp == null) {
timestamp = new DateTime();
}
Sample dataPoint = new Sample(timestamp, value, anomalyScore);
double likelihoodRetval;
if(historicalScores.size() < probationaryPeriod) {
likelihoodRetval = 0.5;
}else{
if(distribution == null || iteration % reestimationPeriod == 0) {
this.distribution = estimateAnomalyLikelihoods(
historicalScores, 10, claLearningPeriod).getParams();
}
AnomalyLikelihoodMetrics metrics = updateAnomalyLikelihoods(Arrays.asList(dataPoint), this.distribution);
this.distribution = metrics.getParams();
likelihoodRetval = 1.0 - metrics.getLikelihoods()[0];
}
historicalScores.add(dataPoint);
this.iteration += 1;
return likelihoodRetval;
}
/**
* Given a series of anomaly scores, compute the likelihood for each score. This
* function should be called once on a bunch of historical anomaly scores for an
* initial estimate of the distribution. It should be called again every so often
* (say every 50 records) to update the estimate.
*
* @param anomalyScores
* @param averagingWindow
* @param skipRecords
* @return
*/
public AnomalyLikelihoodMetrics estimateAnomalyLikelihoods(List<Sample> anomalyScores, int averagingWindow, int skipRecords) {
if(anomalyScores.size() == 0) {
throw new IllegalArgumentException("Must have at least one anomaly score.");
}
// Compute averaged anomaly scores
AveragedAnomalyRecordList records = anomalyScoreMovingAverage(anomalyScores, averagingWindow);
// Estimate the distribution of anomaly scores based on aggregated records
Statistic distribution;
if(records.averagedRecords.size() <= skipRecords) {
distribution = nullDistribution();
}else{
TDoubleList samples = records.getMetrics();
final int numRecordsToCopy = samples.size() - skipRecords;
distribution = estimateNormal(samples.toArray(skipRecords, numRecordsToCopy), true);
/* Taken from the Python Documentation
# HACK ALERT! The CLA model currently does not handle constant metric values
# very well (time of day encoder changes sometimes lead to unstable SDR's
# even though the metric is constant). Until this is resolved, we explicitly
# detect and handle completely flat metric values by reporting them as not
# anomalous.
*/
samples = records.getSamples();
Statistic metricDistribution = estimateNormal(samples.toArray(skipRecords, numRecordsToCopy), false);
if(metricDistribution.variance < 1.5e-5) {
distribution = nullDistribution();
}
}
// Estimate likelihoods based on this distribution
int i = 0;
double[] likelihoods = new double[records.averagedRecords.size()];
for(Sample sample : records.averagedRecords) {
likelihoods[i++] = normalProbability(sample.score, distribution);
}
// Filter likelihood values
double[] filteredLikelihoods = filterLikelihoods(likelihoods);
int len = likelihoods.length;
AnomalyParams params = new AnomalyParams(
new String[] { "distribution", "movingAverage", "historicalLikelihoods" },
distribution,
new MovingAverage(records.historicalValues, records.total, averagingWindow),
len > 0 ?
Arrays.copyOfRange(likelihoods, len - Math.min(averagingWindow, len), len) :
new double[0]);
if(LOG.isDebugEnabled()) {
LOG.debug(
"Discovered params={} Number of likelihoods:{} First 20 likelihoods:{}",
params, len, Arrays.copyOfRange(filteredLikelihoods, 0, 20));
}
return new AnomalyLikelihoodMetrics(filteredLikelihoods, records, params);
}
/**
* Compute updated probabilities for anomalyScores using the given params.
*
* @param anomalyScores a list of records. Each record is a list with a {@link Sample} containing the
following three elements: [timestamp, value, score]
* @param params Associative {@link NamedTuple} returned by the {@link AnomalyLikelihoodMetrics} from
* {@link #estimateAnomalyLikelihoods(List, int, int)}
* @return
*/
public AnomalyLikelihoodMetrics updateAnomalyLikelihoods(List<Sample> anomalyScores, NamedTuple params) {
int anomalySize = anomalyScores.size();
if(LOG.isDebugEnabled()) {
LOG.debug("in updateAnomalyLikelihoods");
LOG.debug("Number of anomaly scores: {}", anomalySize);
LOG.debug("First 20: {}", anomalyScores.subList(0, Math.min(20, anomalySize)));
LOG.debug("Params: {}", params);
}
if(anomalyScores.size() == 0) {
throw new IllegalArgumentException("Must have at least one anomaly score.");
}
if(!isValidEstimatorParams(params)) {
throw new IllegalArgumentException("\"params\" is not a valid parameter structure");
}
double[] histLikelihoods;
if(!params.hasKey("historicalLikelihoods") ||
(histLikelihoods = (double[])params.get("historicalLikelihoods")) == null ||
histLikelihoods.length == 0) {
params = new NamedTuple(
new String[] { "distribution", "movingAverage", "historicalLikelihoods" },
params.get("distribution"),
params.get("movingAverage"),
histLikelihoods = new double[] { 1 });
}
// Compute moving averages of these new scores using the previous values
// as well as likelihood for these scores using the old estimator
MovingAverage mvgAvg = (MovingAverage)params.get("movingAverage");
TDoubleList historicalValues = mvgAvg.getSlidingWindow();
double total = mvgAvg.getTotal();
int windowSize = mvgAvg.getWindowSize();
List<Sample> aggRecordList = new ArrayList<>(anomalySize);
double[] likelihoods = new double[anomalySize];
int i = 0;
for(Sample sample : anomalyScores) {
Calculation calc = MovingAverage.compute(historicalValues, total, sample.score, windowSize);
aggRecordList.add(
new Sample(
sample.date,
sample.value,
calc.getAverage()));
total = calc.getTotal();
likelihoods[i++] = normalProbability(calc.getAverage(), (Statistic)params.get("distribution"));
}
// Filter the likelihood values. First we prepend the historical likelihoods
// to the current set. Then we filter the values. We peel off the likelihoods
// to return and the last windowSize values to store for later.
double[] likelihoods2 = ArrayUtils.concat(histLikelihoods, likelihoods);
double[] filteredLikelihoods = filterLikelihoods(likelihoods2);
likelihoods = Arrays.copyOfRange(filteredLikelihoods, filteredLikelihoods.length - likelihoods.length, filteredLikelihoods.length);
double[] historicalLikelihoods = Arrays.copyOf(likelihoods2, likelihoods2.length - Math.min(windowSize, likelihoods2.length));
// Update the estimator
AnomalyParams newParams = new AnomalyParams(
new String[] { "distribution", "movingAverage", "historicalLikelihoods" },
params.get("distribution"),
new MovingAverage(historicalValues, total, windowSize),
historicalLikelihoods);
return new AnomalyLikelihoodMetrics(
likelihoods,
new AveragedAnomalyRecordList(aggRecordList, historicalValues, total),
newParams);
}
/**
* Filter the list of raw (pre-filtered) likelihoods so that we only preserve
* sharp increases in likelihood. 'likelihoods' can be a numpy array of floats or
* a list of floats.
*
* @param likelihoods
* @return
*/
public double[] filterLikelihoods(double[] likelihoods) {
return filterLikelihoods(likelihoods, 0.99999, 0.999);
}
/**
* Filter the list of raw (pre-filtered) likelihoods so that we only preserve
* sharp increases in likelihood. 'likelihoods' can be an array of floats or
* a list of floats.
*
* @param likelihoods
* @param redThreshold
* @param yellowThreshold
* @return
*/
public double[] filterLikelihoods(double[] likelihoods, double redThreshold, double yellowThreshold) {
redThreshold = 1.0 - redThreshold;
yellowThreshold = 1.0 - yellowThreshold;
// The first value is untouched
double[] filteredLikelihoods = new double[likelihoods.length];
filteredLikelihoods[0] = likelihoods[0];
for(int i = 0;i < likelihoods.length - 1;i++) {
double v = likelihoods[i + 1];
if(v <= redThreshold) {
// If value is in redzone
if(likelihoods[i] > redThreshold) {
// Previous value is not in redzone, so leave as-is
filteredLikelihoods[i + 1] = v;
}else{
filteredLikelihoods[i + 1] = yellowThreshold;
}
}else{
// Value is below the redzone, so leave as-is
filteredLikelihoods[i + 1] = v;
}
}
return filteredLikelihoods;
}
/**
* Given a list of anomaly scores return a list of averaged records.
* anomalyScores is assumed to be a list of records of the form:
* <pre>
* Sample:
* dt = Tuple(2013, 8, 10, 23, 0) --> Date Fields
* sample = (double) 6.0
* metric(avg) = (double) 1.0
* </pre>
*
* @param anomalyScores List of {@link Sample} objects (described contents above)
* @param windowSize Count of historical items over which to compute the average
*
* @return Each record in the returned list contains [datetime field, value, averaged score]
*/
public AveragedAnomalyRecordList anomalyScoreMovingAverage(List<Sample> anomalyScores, int windowSize) {
TDoubleList historicalValues = new TDoubleArrayList();
double total = 0.0;
List<Sample> averagedRecordList = new ArrayList<Sample>();
for(Sample record : anomalyScores) {
////////////////////////////////////////////////////////////////////////////////////////////
// Python version has check for malformed records here, but can't happen in java version. //
////////////////////////////////////////////////////////////////////////////////////////////
Calculation calc = MovingAverage.compute(historicalValues, total, record.score, windowSize);
Sample avgRecord = new Sample(
record.date,
record.value,
calc.getAverage());
averagedRecordList.add(avgRecord);
total = calc.getTotal();
if(LOG.isDebugEnabled()) {
LOG.debug("Aggregating input record: {}, Result: {}", record, averagedRecordList.get(averagedRecordList.size() - 1));
}
}
return new AveragedAnomalyRecordList(averagedRecordList, historicalValues, total);
}
/**
* A Map containing the parameters of a normal distribution based on
* the sampleData.
*
* @param sampleData
* @param performLowerBoundCheck
* @return
*/
public Statistic estimateNormal(double[] sampleData, boolean performLowerBoundCheck) {
double d = ArrayUtils.average(sampleData);
double v = ArrayUtils.variance(sampleData, d);
if(performLowerBoundCheck) {
if(d < 0.03) {
d = 0.03;
}
if(v < 0.0003) {
v = 0.0003;
}
}
// Compute standard deviation
double s = v > 0 ? Math.sqrt(v) : 0.0;
return new Statistic(d, v, s);
}
/**
* Returns a distribution that is very broad and makes every anomaly
* score between 0 and 1 pretty likely
*
* @return
*/
public Statistic nullDistribution() {
if(LOG.isDebugEnabled()) {
LOG.debug("Returning nullDistribution");
}
return new Statistic(0.5, 1e6, 1e3);
}
/**
* Given the normal distribution specified in distributionParams, return
* the probability of getting samples > x
* This is essentially the Q-function
*
* @param x
* @param named
* @return
*/
public double normalProbability(double x, TObjectDoubleMap<String> named) {
return normalProbability(x,
new Statistic(named.get(KEY_MEAN), named.get(KEY_VARIANCE), named.get(KEY_STDEV)));
}
/**
* Given the normal distribution specified in distributionParams, return
* the probability of getting samples > x
* This is essentially the Q-function
*
* @param x
* @param named
* @return
*/
public double normalProbability(double x, Statistic s) {
// Distribution is symmetrical around mean
if(x < s.mean) {
double xp = 2*s.mean - x ;
return 1.0 - normalProbability(xp, s);
}
// How many standard deviations above the mean are we - scaled by 10X for table
double xs = 10*(x - s.mean) / s.stdev;
xs = Math.round(xs);
if(xs > 70) {
return 0.0;
}
return Q[(int)xs];
}
/**
* {@inheritDoc}
*/
@Override
public double compute(int[] activeColumns, int[] predictedColumns, double inputValue, long timestamp) {
if(inputValue == 0) {
throw new IllegalArgumentException("Selected anomaly mode Mode.LIKELIHOOD requires an \"inputValue\" to " +
"the compute() method.");
}
DateTime time = timestamp > 0 ? new DateTime(timestamp) : new DateTime();
// First compute raw anomaly score
double retVal = computeRawAnomalyScore(activeColumns, predictedColumns);
// low likelihood -> high anomaly
double probability = anomalyProbability(inputValue, retVal, time);
// Apply weighting if configured
retVal = isWeighted ? retVal * (1 - probability) : 1 - probability;
// Last, do moving-average if windowSize was specified
if(useMovingAverage) {
retVal = movingAverage.next(retVal);
}
return retVal;
}
/**
* Returns a flag indicating whether the specified params are valid.
* true if so, false if not
*
* @param params a {@link NamedTuple} containing { distribution, movingAverage, historicalLikelihoods }
* @return
*/
public boolean isValidEstimatorParams(NamedTuple params) {
if(params.get("distribution") == null || params.get("movingAverage") == null) {
return false;
}
Statistic stat = (Statistic)params.get("distribution");
if(stat.mean == 0 ||
stat.variance == 0 ||
stat.stdev == 0) {
return false;
}
return true;
}
/////////////////////////////////////////////////////////////////////////////
// AnomalyParams Class Definition //
/////////////////////////////////////////////////////////////////////////////
/**
* <p>
* Extends the dictionary key lookup functionality of the parent class to add
* definite typing for parameter retrieval. Also handles output formatting of
* the contents to a serializable JSON format.
* </p>
* <p>
* <pre>
* {
* "distribution": # describes the distribution is-a {@link Statistic}
* {
* "name": STRING, # name of the distribution, such as 'normal'
* "mean": SCALAR, # mean of the distribution
* "variance": SCALAR, # variance of the distribution
*
* # There may also be some keys that are specific to the distribution
* },
*
* "historicalLikelihoods": [] # Contains the last windowSize likelihood
* # values returned
*
* "movingAverage": # stuff needed to compute a rolling average is-a {@link MovingAverage}
* # of the anomaly scores
* {
* "windowSize": SCALAR, # the size of the averaging window
* "historicalValues": [], # list with the last windowSize anomaly
* # scores
* "total": SCALAR, # the total of the values in historicalValues
* },
*
* }
* </pre>
* </p>
*
* @author David Ray
*/
public static class AnomalyParams extends NamedTuple {
private static final long serialVersionUID = 1L;
private final Statistic distribution;
private final MovingAverage movingAverage;
private final double[] historicalLikelihoods;
private final int windowSize;
/**
* Constructs a new {@code AnomalyParams}
* @param keys
* @param values
*/
public AnomalyParams(String[] keys, Object... values) {
super(keys, values);
if(keys.length != 3 || values.length != 3) {
throw new IllegalArgumentException("AnomalyParams must have \"distribution\", \"movingAverage\", and \"historicalLikelihoods\"" +
" parameters. keys.length != 3 or values.length != 3");
}
this.distribution = (Statistic)get(KEY_DIST);
this.movingAverage = (MovingAverage)get(KEY_MVG_AVG);
this.historicalLikelihoods = (double[])get(KEY_HIST_LIKE);
this.windowSize = movingAverage.getWindowSize();
}
/**
* Returns the {@link Statistic} containing point calculations.
* @return
*/
public Statistic distribution() {
return distribution;
}
/**
* Returns the {@link MovingAverage} object
* @return
*/
public MovingAverage movingAverage() {
return movingAverage;
}
/**
* Returns the array of computed likelihoods
* @return
*/
public double[] historicalLikelihoods() {
return historicalLikelihoods;
}
/**
* Returns the window size of the moving average.
* @return
*/
public int windowSize() {
return windowSize;
}
/* (non-Javadoc)
* @see java.lang.Object#hashCode()
*/
@Override
public int hashCode() {
final int prime = 31;
int result = super.hashCode();
result = prime * result + ((distribution == null) ? 0 : distribution.hashCode());
result = prime * result + Arrays.hashCode(historicalLikelihoods);
result = prime * result + ((movingAverage == null) ? 0 : movingAverage.hashCode());
result = prime * result + windowSize;
return result;
}
/* (non-Javadoc)
* @see java.lang.Object#equals(java.lang.Object)
*/
@Override
public boolean equals(Object obj) {
if(this == obj)
return true;
if(!super.equals(obj))
return false;
if(getClass() != obj.getClass())
return false;
AnomalyParams other = (AnomalyParams)obj;
if(distribution == null) {
if(other.distribution != null)
return false;
} else if(!distribution.equals(other.distribution))
return false;
if(!Arrays.equals(historicalLikelihoods, other.historicalLikelihoods))
return false;
if(movingAverage == null) {
if(other.movingAverage != null)
return false;
} else if(!movingAverage.equals(other.movingAverage))
return false;
if(windowSize != other.windowSize)
return false;
return true;
}
}
// Table lookup for Q function, from wikipedia
// http://en.wikipedia.org/wiki/Q-function
private static double[] Q;
static {
Q = new double[71];
Q[0] = 0.500000000;
Q[1] = 0.460172163;
Q[2] = 0.420740291;
Q[3] = 0.382088578;
Q[4] = 0.344578258;
Q[5] = 0.308537539;
Q[6] = 0.274253118;
Q[7] = 0.241963652;
Q[8] = 0.211855399;
Q[9] = 0.184060125;
Q[10] = 0.158655254;
Q[11] = 0.135666061;
Q[12] = 0.115069670;
Q[13] = 0.096800485;
Q[14] = 0.080756659;
Q[15] = 0.066807201;
Q[16] = 0.054799292;
Q[17] = 0.044565463;
Q[18] = 0.035930319;
Q[19] = 0.028716560;
Q[20] = 0.022750132;
Q[21] = 0.017864421;
Q[22] = 0.013903448;
Q[23] = 0.010724110;
Q[24] = 0.008197536;
Q[25] = 0.006209665;
Q[26] = 0.004661188;
Q[27] = 0.003466974;
Q[28] = 0.002555130;
Q[29] = 0.001865813;
Q[30] = 0.001349898;
Q[31] = 0.000967603;
Q[32] = 0.000687138;
Q[33] = 0.000483424;
Q[34] = 0.000336929;
Q[35] = 0.000232629;
Q[36] = 0.000159109;
Q[37] = 0.000107800;
Q[38] = 0.000072348;
Q[39] = 0.000048096;
Q[40] = 0.000031671;
// From here on use the approximation in http://cnx.org/content/m11537/latest/
Q[41] = 0.000021771135897;
Q[42] = 0.000014034063752;
Q[43] = 0.000008961673661;
Q[44] = 0.000005668743475;
Q[45] = 0.000003551942468;
Q[46] = 0.000002204533058;
Q[47] = 0.000001355281953;
Q[48] = 0.000000825270644;
Q[49] = 0.000000497747091;
Q[50] = 0.000000297343903;
Q[51] = 0.000000175930101;
Q[52] = 0.000000103096834;
Q[53] = 0.000000059836778;
Q[54] = 0.000000034395590;
Q[55] = 0.000000019581382;
Q[56] = 0.000000011040394;
Q[57] = 0.000000006164833;
Q[58] = 0.000000003409172;
Q[59] = 0.000000001867079;
Q[60] = 0.000000001012647;
Q[61] = 0.000000000543915;
Q[62] = 0.000000000289320;
Q[63] = 0.000000000152404;
Q[64] = 0.000000000079502;
Q[65] = 0.000000000041070;
Q[66] = 0.000000000021010;
Q[67] = 0.000000000010644;
Q[68] = 0.000000000005340;
Q[69] = 0.000000000002653;
Q[70] = 0.000000000001305;
}
}