/* Function: main_loop_threaded()
* Date: SRE, Wed Dec 1 12:43:09 1999 [St. Louis]
*
* Purpose: Given an HMM and parameters for synthesizing random
* sequences; return a histogram of scores.
* (Threaded version.)
*
* Args: hmm - an HMM to calibrate.
* seed - random number seed
* nsample - number of seqs to synthesize
* lenmean - mean length of random sequence
* lensd - std dev of random seq length
* fixedlen - if nonzero, override lenmean, always this len
* nthreads - number of threads to start
* ret_hist - RETURN: the score histogram
* ret_max - RETURN: highest score seen in simulation
* twatch - RETURN: accumulation of thread times
*
* Returns: (void)
* hist is alloc'ed here, and must be free'd by caller.
*/
static void
main_loop_threaded(struct plan7_s *hmm, int seed, int nsample,
float lenmean, float lensd, int fixedlen,
int nthreads,
struct histogram_s **ret_hist, float *ret_max,
Stopwatch_t *twatch)
{
struct histogram_s *hist;
float randomseq[MAXABET];
float p1;
struct workpool_s *wpool; /* pool of worker threads */
/* Initialize.
* We assume we've already set the alphabet (safe, because
* HMM input sets the alphabet).
*/
sre_srandom(seed);
P7Logoddsify(hmm, TRUE);
P7DefaultNullModel(randomseq, &p1);
hist = AllocHistogram(-200, 200, 100);
wpool = workpool_start(hmm, lenmean, lensd, fixedlen, randomseq, nsample,
hist, nthreads);
workpool_stop(wpool);
*ret_hist = hist;
*ret_max = wpool->max_score;
StopwatchInclude(twatch, &(wpool->watch));
workpool_free(wpool);
return;
}
/* Function: Plan9toPlan7()
*
* Purpose: Convert an old HMM into Plan7. Configures it in
* ls mode.
*
* Args: hmm - old ugly plan9 style HMM
* ret_plan7 - new wonderful Plan7 HMM
*
* Return: (void)
* Plan7 HMM is allocated here. Free w/ FreePlan7().
*/
void
Plan9toPlan7(struct plan9_s *hmm, struct plan7_s **ret_plan7)
{
struct plan7_s *plan7;
int k, x;
plan7 = AllocPlan7(hmm->M);
for (k = 1; k < hmm->M; k++)
{
plan7->t[k][TMM] = hmm->mat[k].t[MATCH];
plan7->t[k][TMD] = hmm->mat[k].t[DELETE];
plan7->t[k][TMI] = hmm->mat[k].t[INSERT];
plan7->t[k][TDM] = hmm->del[k].t[MATCH];
plan7->t[k][TDD] = hmm->del[k].t[DELETE];
plan7->t[k][TIM] = hmm->ins[k].t[MATCH];
plan7->t[k][TII] = hmm->ins[k].t[INSERT];
}
for (k = 1; k <= hmm->M; k++)
for (x = 0; x < Alphabet_size; x++)
plan7->mat[k][x] = hmm->mat[k].p[x];
for (k = 1; k < hmm->M; k++)
for (x = 0; x < Alphabet_size; x++)
plan7->ins[k][x] = hmm->ins[k].p[x];
plan7->tbd1 = hmm->mat[0].t[DELETE] / (hmm->mat[0].t[DELETE] + hmm->mat[0].t[MATCH]);
/* We have to make up the null transition p1; use default */
P7DefaultNullModel(plan7->null, &(plan7->p1));
for (x = 0; x < Alphabet_size; x++)
plan7->null[x] = hmm->null[x];
if (hmm->name != NULL)
Plan7SetName(plan7, hmm->name);
if (hmm->flags & HMM_REF) {
strcpy(plan7->rf, hmm->ref);
plan7->flags |= PLAN7_RF;
}
if (hmm->flags & HMM_CS) {
strcpy(plan7->cs, hmm->cs);
plan7->flags |= PLAN7_CS;
}
Plan7LSConfig(plan7); /* configure specials for ls-style alignment */
Plan7Renormalize(plan7); /* mainly to correct for missing ID and DI */
plan7->flags |= PLAN7_HASPROB; /* probabilities are valid */
plan7->flags &= ~PLAN7_HASBITS; /* scores are not valid */
*ret_plan7 = plan7;
}
/* Function: main_loop_serial()
* Date: SRE, Tue Aug 18 16:18:28 1998 [St. Louis]
*
* Purpose: Given an HMM and parameters for synthesizing random
* sequences; return a histogram of scores.
* (Serial version)
*
* Args: hmm - an HMM to calibrate.
* seed - random number seed
* nsample - number of seqs to synthesize
* lenmean - mean length of random sequence
* lensd - std dev of random seq length
* fixedlen - if nonzero, override lenmean, always this len
* ret_hist - RETURN: the score histogram
* ret_max - RETURN: highest score seen in simulation
*
* Returns: (void)
* hist is alloc'ed here, and must be free'd by caller.
*/
static void
main_loop_serial(struct plan7_s *hmm, int seed, int nsample,
float lenmean, float lensd, int fixedlen,
struct histogram_s **ret_hist, float *ret_max)
{
struct histogram_s *hist;
float randomseq[MAXABET];
float p1;
float max;
char *seq;
char *dsq;
float score;
int sqlen;
int idx;
/* Initialize.
* We assume we've already set the alphabet (safe, because
* HMM input sets the alphabet).
*/
sre_srandom(seed);
P7Logoddsify(hmm, TRUE);
P7DefaultNullModel(randomseq, &p1);
hist = AllocHistogram(-200, 200, 100);
max = -FLT_MAX;
for (idx = 0; idx < nsample; idx++)
{
/* choose length of random sequence */
if (fixedlen) sqlen = fixedlen;
else do sqlen = (int) Gaussrandom(lenmean, lensd); while (sqlen < 1);
/* generate it */
seq = RandomSequence(Alphabet, randomseq, Alphabet_size, sqlen);
dsq = DigitizeSequence(seq, sqlen);
if (P7ViterbiSize(sqlen, hmm->M) <= RAMLIMIT)
score = P7Viterbi(dsq, sqlen, hmm, NULL);
else
score = P7SmallViterbi(dsq, sqlen, hmm, NULL);
AddToHistogram(hist, score);
if (score > max) max = score;
free(dsq);
free(seq);
}
*ret_hist = hist;
*ret_max = max;
return;
}
void HMMCreateWPoolTask::runUnsafe() {
const UHMMCalibrateSettings& settings = pt->getSettings();
WorkPool_s* wpool = pt->getWorkPool();
SetAlphabet(wpool->hmm->atype);
sre_srandom(settings.seed);
wpool->fixedlen = settings.fixedlen;
wpool->hist = AllocHistogram(-200, 200, 100);
wpool->lenmean = settings.lenmean;
wpool->lensd = settings.lensd;
wpool->nsample = settings.nsample;
wpool->nseq = 0;
wpool->randomseq.resize(MAXABET);
wpool->max_score = -FLT_MAX;
float p1;
P7Logoddsify(wpool->hmm, TRUE);
P7DefaultNullModel(wpool->randomseq.data(), &p1);
}
static void main_loop_serial(struct plan7_s *hmm, int seed, int nsample,
float lenmean, float lensd, int fixedlen,
struct histogram_s **ret_hist, float *ret_max, int& cancelFlag, int& progress)
{
struct histogram_s *hist;
struct dpmatrix_s *mx;
float randomseq[MAXABET];
float p1;
float max;
char *seq;
unsigned char *dsq;
float score;
int sqlen;
int idx;
// Initialize.
// We assume we've already set the alphabet (safe, because
// HMM input sets the alphabet).
sre_srandom(seed);
//get HMMERTaskLocalData
HMMERTaskLocalData *tls = getHMMERTaskLocalData();
alphabet_s &al = tls->al;
SetAlphabet(hmm->atype);
P7Logoddsify(hmm, TRUE);
P7DefaultNullModel(randomseq, &p1);
hist = AllocHistogram(-200, 200, 100);
mx = CreatePlan7Matrix(1, hmm->M, 25, 0);
max = -FLT_MAX;
progress = 0;
int pStub;
for (idx = 0; idx < nsample && !cancelFlag; idx++) {
// choose length of random sequence
if (fixedlen) {
sqlen = fixedlen;
} else {
do sqlen = (int) Gaussrandom(lenmean, lensd); while (sqlen < 1);
}
// generate it
seq = RandomSequence(al.Alphabet, randomseq, al.Alphabet_size, sqlen);
dsq = DigitizeSequence(seq, sqlen);
if (P7ViterbiSpaceOK(sqlen, hmm->M, mx)) {
score = P7Viterbi(dsq, sqlen, hmm, mx, NULL);
} else {
score = P7SmallViterbi(dsq, sqlen, hmm, mx, NULL, pStub);
}
AddToHistogram(hist, score);
max = qMax(score, max);
progress = int(100*idx/float(nsample));
free(dsq);
free(seq);
}
FreePlan7Matrix(mx);
*ret_hist = hist;
*ret_max = max;
}
int
main(void)
{
int master_tid; /* PVM TID of our master */
int slaveidx; /* my slave index (0..nslaves-1) */
struct plan7_s *hmm; /* HMM to calibrate, sent from master */
struct histogram_s *hist; /* score histogram */
int hmmidx; /* index of this HMM */
char *seq; /* synthetic random sequence */
char *dsq; /* digitized seq */
int len; /* length of seq */
float sc; /* score of seq aligned to HMM */
float max; /* maximum score seen in sample */
int seed; /* random number seed */
int nsample; /* number of seqs to sample */
int fixedlen; /* if nonzero, fixed length of seq */
float lenmean; /* Gaussian mean length of seq */
float lensd; /* Gaussian length std. dev. for seq */
int fitok; /* TRUE if EVD fit was OK */
float randomseq[MAXABET]; /* iid frequencies of residues */
float p1;
int alphatype; /* alphabet type, hmmAMINO or hmmNUCLEIC */
int idx;
int code;
/* Register leave_pvm() cleanup function so any exit() call
* first calls pvm_exit().
*/
if (atexit(leave_pvm) != 0) { pvm_exit(); Die("slave couldn't register leave_pvm()"); }
/*****************************************************************
* initialization.
* Master broadcasts the problem to us: parameters of the
* HMM calibration.
******************************************************************/
master_tid = pvm_parent(); /* who's our master? */
pvm_recv(master_tid, HMMPVM_INIT);
pvm_upkint(&nsample, 1, 1);
pvm_upkint(&fixedlen, 1, 1);
pvm_upkfloat(&lenmean, 1, 1);
pvm_upkfloat(&lensd, 1, 1);
/* tell the master we're OK and ready to go (or not)
*/
code = HMMPVM_OK;
pvm_initsend(PvmDataDefault);
pvm_pkint(&code, 1, 1);
pvm_send(master_tid, HMMPVM_RESULTS);
/*****************************************************************
* Main loop.
* Receive a random number seed, then an HMM to search against.
* If we receive a -1 seed, we shut down.
*****************************************************************/
slaveidx = -1;
for (;;)
{
pvm_recv(master_tid, HMMPVM_WORK);
pvm_upkint(&seed, 1, 1);
if (seed == -1) break; /* shutdown signal */
pvm_upkint(&hmmidx, 1, 1);
pvm_upkint(&alphatype,1, 1);
SetAlphabet(alphatype);
hmm = PVMUnpackHMM();
if (hmm == NULL) Die("oh no, the HMM never arrived");
if (slaveidx == -1) slaveidx = hmmidx;
P7DefaultNullModel(randomseq, &p1);
sre_srandom(seed);
P7Logoddsify(hmm, TRUE);
hist = AllocHistogram(-200, 200, 100);
max = -FLT_MAX;
for (idx = 0; idx < nsample; idx++)
{
/* choose length of random sequence */
if (fixedlen) len = fixedlen;
else do len = (int) Gaussrandom(lenmean, lensd); while (len < 1);
/* generate it */
seq = RandomSequence(Alphabet, randomseq, Alphabet_size, len);
dsq = DigitizeSequence(seq, len);
if (P7ViterbiSize(len, hmm->M) <= RAMLIMIT)
sc = P7Viterbi(dsq, len, hmm, NULL);
else
sc = P7SmallViterbi(dsq, len, hmm, NULL);
AddToHistogram(hist, sc);
if (sc > max) max = sc;
free(seq);
free(dsq);
}
/* Fit an EVD to the observed histogram.
* The TRUE left-censors and fits only the right slope of the histogram.
* The 9999. is an arbitrary high number that means we won't trim outliers
* on the right.
*/
fitok = ExtremeValueFitHistogram(hist, TRUE, 9999.);
/* Return output to master.
* Currently we don't send the histogram back, but we could.
*/
pvm_initsend(PvmDataDefault);
pvm_pkint(&slaveidx, 1, 1);
pvm_pkint(&hmmidx, 1, 1);
PVMPackString(hmm->name);
pvm_pkint(&fitok, 1, 1);
pvm_pkfloat(&(hist->param[EVD_MU]), 1, 1);
pvm_pkfloat(&(hist->param[EVD_LAMBDA]), 1, 1);
pvm_pkfloat(&max, 1, 1);
pvm_send(master_tid, HMMPVM_RESULTS);
/* cleanup
*/
FreeHistogram(hist);
FreePlan7(hmm);
}
/***********************************************
* Cleanup, return.
***********************************************/
return 0; /* pvm_exit() is called by atexit() registration. */
}
