uint64_t extrapolate_distinct_kmers_wrapped(unsigned long nbytes_memory, Bank *Reads)
{
unsigned long size_linearCounter = nbytes_memory * 8L; // alloc 8 bits * nbytes for counting
LinearCounter *linearCounter = new LinearCounter(size_linearCounter);
int stops = 100000;
// variant of bloom_pass_reads
int64_t NbRead = 0;
int64_t NbInsertedKmers = 0;
Reads->rewind_all();
char * rseq;
long i;
kmer_type kmer, graine, graine_revcomp;
long nb_distinct_kmers = 0;
long previous_nb_distinct_kmers = 0;
uint64_t estimated_nb_reads = Reads->estimate_nb_reads();
bool stop = false;
while (Reads->get_next_seq(&rseq,&readlen))
{
if (stop)
break;
for (i=0; i<readlen-sizeKmer+1; i++)
{
kmer = extractKmerFromRead(rseq,i,&graine,&graine_revcomp);
linearCounter->add(kmer);
NbInsertedKmers++;
if (NbInsertedKmers % stops == 0 && NbRead != 0)
{
previous_nb_distinct_kmers = nb_distinct_kmers;
nb_distinct_kmers = linearCounter->count()*estimated_nb_reads/NbRead;
//printf("estimated now: %ld\n",nb_distinct_kmers);
// the following condition will grossly over-estimate the number of distinct kmers
// I expect the correct result to be in the same order of magnitude
// 5% error
if (abs((int)(nb_distinct_kmers-previous_nb_distinct_kmers)) < previous_nb_distinct_kmers/20)
stop = true;
if (!linearCounter->is_accurate())
stop = true;
}
}
NbRead++;
if ((NbRead%10000)==0) fprintf (stderr,(char*)"%cExtrapolating number of distinct kmers %lld",13,NbRead);
}
if (!linearCounter->is_accurate())
{
printf("Inaccurate estimation, restarting with %d MB RAM\n",(2*nbytes_memory)/1024/1024);
delete linearCounter;
return extrapolate_distinct_kmers_wrapped(2*nbytes_memory, Reads);
}
nb_distinct_kmers = linearCounter->count()*estimated_nb_reads/NbRead; // this is a very rough estimation
printf("Linear estimation: ~%ld M distinct kmers are in the reads\n",nb_distinct_kmers/1000000L);
delete linearCounter;
return nb_distinct_kmers;
}
