//implementation of spring outlier
int TimeSeriesSetSlice::springOutlier(void)
{
double *timeSeries = set->getRawData(0, 0);
int seqLen = set->seqLength;
int seqCnt = set->seqCount;
int startT = slice.start;
int endT = slice.end;
//for each subsequence in this interval
//get each subsequence
double msf = 0; //max distance so far
int maxTS = -1; //Sequence
double tempDist = 0;
vector<double> temp1;
vector<double> temp2;
for (int i = 0; i < seqCnt; i++)
{
temp1.clear();
temp2.clear();
//copy the subsequence in temp
for (int p = startT; p <= endT; p++)
temp1.push_back(timeSeries[i * seqLen + p]);
//compute the distance of this subsequence with all sequences
for (int j = 0; j < seqCnt; j++)
{
if (j!=i)
{
for (int p = startT; p <= endT; p++)
temp2.push_back(timeSeries[j * seqLen + p]);
tempDist+=simpleDTW(temp2,temp1);
temp2.clear();
}
}
if (tempDist > msf)
{
msf=tempDist;
maxTS=i;
}
tempDist=0;
}
cout << "Outlier found by SPRING " << maxTS << " " << msf << endl;
return maxTS;
}
// the incoming file should be PAA'd already
DTWData detectOutliers(std::string dataFile, int length) {
std::ifstream dataStream(dataFile.c_str());
if (!dataStream) {
std::cout << "ERROR: ifstream failed on " << dataFile << ": " << strerror(errno) << std::endl;
return DTWData();
}
std::string dataTimePoints;
std::string queryData;
std::vector<double> dataVector;
std::vector<double> queryVector;
int dataPos;
int curQTimeSeries = -1;
double maxSum = 0;
int maxSeries;
int maxOffset;
int queryStart = 0, queryEnd = length;
int dataStart = 0, dataEnd = length;
// get first query time series
while (queryData.empty())
std::getline(dataStream, queryData);
// run through each time series as query
while (!queryData.empty()) {
queryStart = 0;
queryEnd = length;
curQTimeSeries++;
// turn time series into vector
queryVector = timeSeriesToVector(queryData);
// grab location in file
dataPos = dataStream.tellg();
// run through all chunks of the query
while (queryEnd <= (int) queryVector.size()) {
double curDist = 0;
std::vector<double> subQVec(queryVector.begin()+queryStart, queryVector.begin()+queryEnd);
// return to the beginning of the file
dataStream.clear();
dataStream.seekg(0, dataStream.beg);
// grab first data time series
while (dataTimePoints.empty() && dataStream.good())
std::getline(dataStream, dataTimePoints);
// run through the rest of the time series
while (!dataTimePoints.empty()) {
// turn time series into vector
dataVector = timeSeriesToVector(dataTimePoints);
// reset the data start and end for the next query chunk
dataStart = 0;
dataEnd = length;
// run through all chunks of the data
while (dataEnd <= (int) dataVector.size()) {
std::vector<double> subDVec(dataVector.begin()+dataStart, dataVector.begin()+dataEnd);
curDist += simpleDTW(subQVec, subDVec);
dataStart++;
dataEnd++;
// std::cout << "subV: " << subQVec << std::endl;
// std::cout << "subD: " << subDVec << std::endl;
}
dataVector.clear();
dataTimePoints = "";
// get next time series
while (dataTimePoints.empty() && dataStream.good())
std::getline(dataStream, dataTimePoints);
}
// check against worst so far
double oldMax = maxSum;
maxSum = std::max(maxSum, curDist);
if (maxSum != oldMax) {
maxSeries = curQTimeSeries;
maxOffset = queryStart;
}
queryStart++;
queryEnd++;
}
dataStream.clear();
dataStream.seekg(dataPos, dataStream.beg);
queryData = "";
dataTimePoints = "";
dataVector.clear();
queryVector.clear();
// turn "next" data time series into query vector
while (queryData.empty() && dataStream.good())
std::getline(dataStream, queryData);
}
return DTWData(maxSum, maxSeries, maxOffset, length);
}
//implementation of Spring
//query Type 1 mean k similar
vector<kBest> TimeSeriesSetSlice::springkSimilar(vector<double> tempQ, int queryType, int k)
{
vector<kBest> kbestArray;
double *timeSeries = set->getRawData(0, 0);
int N = set->seqCount;
int L = set->seqLength;
double bsf=INF;
int bsfIndex = -1; //index of bsf time series
double currentDist=INF;
int bestIntervalS = -1;
int bestIntervalE = -1;
vector<double> temp; //temporary subsequence
int kbestCount=0;
kBest tempBest;
if(queryType==1)
{
if(k==1) //most similar
{
//get each subsequence
for(int j=1;j<L;j++)
{
for (int l=0, m=l+j; m<L; l++, m++)
{
for (int i=0;i<N;i++)
{
//copy the subsequence in temp
for(int p=0;p<=m-l;p++)
temp.push_back(timeSeries[i*L + l+p]);
currentDist=simpleDTW(temp,tempQ);
if(currentDist<bsf)
{
bsf=currentDist;
bsfIndex=i; //best time series index
bestIntervalS=l; //record the best interval
bestIntervalE=m;
}
//clear the temp array
temp.clear();
}
}
}
cout<<"TS "<<bsfIndex<<" Interval "<<bestIntervalS<<" "<<bestIntervalE<<" Dist "<<bsf<<endl;
}
else
{
//have to find more than 1 similar time series
//get each subsequence
for(int j=1;j<L;j++)
{
for (int l=0, m=l+j; m<L-1; l++, m++)
{
for (int i=0;i<N;i++)
{
//copy the subsequence in temp
for(int p=0;p<m-l;p++)
temp.push_back(timeSeries[i*L + l+p]);
currentDist=simpleDTW(tempQ,temp);
if(kbestCount<k)
{
//add this TS to k best
kbestCount++;
tempBest.dist=currentDist;
tempBest.id=i;
kbestArray.push_back(tempBest);
}
else
{
sort(kbestArray.begin(),kbestArray.end(), _sortByDist);
double tempD=kbestArray[kbestCount].dist; //getting the last distance
if(tempD>currentDist)
{
tempBest.dist=currentDist;
tempBest.id=i;
kbestArray[kbestCount]=tempBest;
}
}
temp.clear();
}
}
}
}
}
return kbestArray;
}
// both incoming files should be PAA'd already
DTWData DTWaFile(std::string dataFile, std::string queryFile) {
std::ifstream data(getPAAFilename(dataFile).c_str()); // bigger file
std::ifstream query(getPAAFilename(queryFile).c_str()); // smaller file
if (!data) {
std::cout << "ERROR: ifstream failed on " << dataFile << ": " << strerror(errno) << std::endl;
return DTWData();
}
if (!query) {
std::cout << "ERROR: ifstream failed on " << queryFile << ": " << strerror(errno) << std::endl;
return DTWData();
}
std::string dataTimePoints;
std::string queryData;
std::vector<double> dataVector;
std::vector<double> queryVector;
int curTimeSeries = -1;
int bestMatchTimeSeries = 1;
int bestMatchIdx = 1;
int bestMatchBlkSz = 2;
double bestMatchDistance = std::numeric_limits<double>::max();
// turn query into vector
std::getline(query, queryData);
std::size_t prev = 0, pos;
while ((pos = queryData.find_first_of(" ,", prev)) != std::string::npos) {
if (pos > prev)
queryVector.push_back(stod(queryData.substr(prev, pos-prev)));
prev = pos+1;
}
if (prev < queryData.length())
queryVector.push_back(stod(queryData.substr(prev, std::string::npos)));
while (dataTimePoints.empty() && data.good())
std::getline(data, dataTimePoints);
while (data.good()) {
// get the next data timeseries
curTimeSeries++;
// split the timeseries numbers on space or comma
std::size_t prev = 0, pos;
while ((pos = dataTimePoints.find_first_of(" ,", prev)) != std::string::npos) {
if (pos > prev)
dataVector.push_back(stod(dataTimePoints.substr(prev, pos-prev)));
prev = pos+1;
}
if (prev < dataTimePoints.length())
dataVector.push_back(stod(dataTimePoints.substr(prev, std::string::npos)));
// run through all combinations from query
for (int blkSz = 2; blkSz <= (int)dataVector.size(); blkSz++) {
for (int startIdx = 0; startIdx+blkSz <= (int)dataVector.size(); startIdx++) {
std::vector<double> subVec(dataVector.begin()+startIdx, dataVector.begin()+startIdx+blkSz);
double newBest = std::min(simpleDTW(queryVector, subVec), bestMatchDistance);
if (newBest != bestMatchDistance) {
bestMatchDistance = newBest;
bestMatchIdx = startIdx;
bestMatchBlkSz = blkSz;
bestMatchTimeSeries = curTimeSeries;
}
}
}
// get the next data timeseries
dataVector.clear();
dataTimePoints = "";
while (dataTimePoints.empty() && data.good())
std::getline(data, dataTimePoints);
}
return DTWData(bestMatchDistance, bestMatchTimeSeries, bestMatchIdx, bestMatchBlkSz);
}