void fragFind(struct seqList *goodSeq, char *badName, int fragSize, int mismatchesAllowed,
boolean considerRc, double profile[16][4])
/* Do fast finding of patterns that are in FA file "goodName", but not in FA file
* "badName." BadName can be null. Pass in the size of the pattern (fragSize) and
* the number of mismatches to pattern you're willing to tolerate (mismatchesAllowed).
* It returns the pattern in profile. */
{
int *goodTable, *badTable = NULL;
int goodCount, badCount = 0;
int goodIx;
long startTime;
DNA unpacked[17];
if (mismatchesAllowed > 3)
errAbort("Sorry, fragFind can only handle 0-3 mismatches.");
if (fragSize > 10)
errAbort("Sorry, fragFind can only handle fragments up to 10 bases.");
startTime = clock1000();
makeOligoHistogram(NULL, goodSeq, fragSize, &goodTable, &goodCount);
if (badName)
makeOligoHistogram(badName, NULL, fragSize, &badTable, &badCount);
if (badName)
{
normalizeTable(goodTable, fragSize);
normalizeTable(badTable, fragSize);
diffTables(goodTable, badTable, fragSize);
goodIx = fuzzVal(badTable, fragSize, mismatchesAllowed, considerRc);
}
else
{
goodIx = fuzzVal(goodTable, fragSize, mismatchesAllowed, considerRc);
}
freez(&goodTable);
freez(&badTable);
unpackVal(goodIx, fragSize, unpacked);
makeProfile(unpacked, fragSize, mismatchesAllowed, goodSeq, considerRc, profile);
}
void tmodel<COUNT, PROB>::normalizeTable(const vcbList&engl, const vcbList&french, int iter)
// normalize conditional probability P(fj/ei):
// i.e. make sure that Sum over all j of P(fj/e) = 1
// this method reads the counts portion of the table and normalize into
// the probability portion. Then the counts are cleared (i.e. zeroed)
// if the resulting probability of an entry is below a threshold, then
// remove it .
{
if( iter==2 )
{
total2.resize(engl.uniqTokens());for(unsigned int i=0;i<total2.size();i++)total2[i]=0.0;
}
nFrench.resize(engl.uniqTokens());for(unsigned int i=0;i<nFrench.size();i++)nFrench[i]=0;
nEng.resize(french.uniqTokens());for(unsigned int i=0;i<nEng.size();i++)nEng[i]=0;
Vector<double> total(engl.uniqTokens(),0.0);
//Vector<int> nFrench(engl.uniqTokens(), 0);
//Vector<int> nEng(french.uniqTokens(), 0);
typename hash_map<wordPairIds, CPPair, hashpair, equal_to<wordPairIds> >::const_iterator i;
for(i = ef.begin(); i != ef.end(); i++){ // for all possible source words e
if( iter==2 )
total2[((*i).first).first] += (*i).second.count;
total[((*i).first).first] += (*i).second.count;
nFrench[((*i).first).first]++;
nEng[((*i).first).second]++;
}
for(unsigned int k=0;k<engl.uniqTokens();++k)
if( nFrench[k] )
{
double probMass=(french.uniqTokensInCorpus()-nFrench[k])*PROB_SMOOTH;
if( probMass<0.0 )
cout << k << " french.uniqTokensInCorpus(): " << french.uniqTokensInCorpus() << " nFrench[k]:"<< nFrench[k] << '\n';
total[k]+= total[k]*probMass/(1-probMass);
}
typename hash_map<wordPairIds, CPPair, hashpair, equal_to<wordPairIds> >::iterator j, k;
PROB p ;
int nParams=0;
int nor = 0;
for(j = ef.begin(); j != ef.end(); ){
k = j;
k++ ;
if( (total[((*j).first).first])>0.0 )
p = ((((*j).second).count) /(total[((*j).first).first])) ;
else
p= 0.0;
if (p > PROB_CUTOFF)
{
if( iter>0 )
{
if(useWord2Vec && word2vec.Method == 2){
p = word2vec.L * word2vec.getW2VProb(((*j).first).first, ((*j).first).second) + (1. - word2vec.L) * p;
nor = 1;
}
((*j).second).prob = 0 ;
((*j).second).count = p ;
}
else
{
((*j).second).prob = p ;
((*j).second).count = 0 ;
}
nParams++;
}
else {
erase(((*j).first).first, ((*j).first).second);
}
j = k ;
}
if(nor)
cout << "probabilities Normalized in iteration = " << iter << endl;
if( iter>0 )
return normalizeTable(engl, french, iter-1);
else
{
}
}
