int
p7_hit_Compare(const P7_HIT *h1, const P7_HIT *h2, float tol)
{
int d;
int status;
if ( strcmp(h1->name, h2->name) != 0) return eslFAIL;
if (esl_strcmp(h1->acc, h2->acc) != 0) return eslFAIL;
if (esl_strcmp(h1->desc, h2->desc) != 0) return eslFAIL;
if ( h1->window_length != h2->window_length) return eslFAIL;
if ( h1->ndom != h2->ndom) return eslFAIL;
if ( h1->noverlaps != h2->noverlaps) return eslFAIL;
if ( h1->flags != h2->flags) return eslFAIL;
if ( h1->nreported != h2->nreported) return eslFAIL;
if ( h1->nincluded != h2->nincluded) return eslFAIL;
if ( h1->best_domain != h2->best_domain) return eslFAIL;
if ( h1->seqidx != h2->seqidx) return eslFAIL;
if ( h1->subseq_start != h2->subseq_start) return eslFAIL;
if ( h1->offset != h2->offset) return eslFAIL;
if ( esl_DCompare( h1->sortkey, h2->sortkey, tol ) != eslOK) return eslFAIL;
if ( esl_FCompare( h1->score, h2->score, tol ) != eslOK) return eslFAIL;
if ( esl_FCompare( h1->pre_score, h2->pre_score, tol ) != eslOK) return eslFAIL;
if ( esl_FCompare( h1->sum_score, h2->sum_score, tol ) != eslOK) return eslFAIL;
if ( esl_DCompare( h1->lnP, h2->lnP, tol ) != eslOK) return eslFAIL;
if ( esl_DCompare( h1->pre_lnP, h2->pre_lnP, tol ) != eslOK) return eslFAIL;
if ( esl_DCompare( h1->sum_lnP, h2->sum_lnP, tol ) != eslOK) return eslFAIL;
if ( esl_DCompare( h1->nexpected, h2->nexpected, tol ) != eslOK) return eslFAIL;
for (d = 0; d < h1->ndom; d++)
if (( status = p7_domain_Compare(&(h1->dcl[d]), &(h2->dcl[d]), tol)) != eslOK) return status;
return eslOK;
}
int
esl_vec_FCompare(const float *vec1, const float *vec2, int n, float tol)
{
int i;
for (i = 0; i < n; i++) if (esl_DCompare(vec1[i], vec2[i], tol) == eslFAIL) return eslFAIL;
return eslOK;
}
/* Function: esl_vec_DCompare()
* Synopsis: Return <eslOK> if two vectors are equal.
* Incept: SRE, Mon Nov 6 10:20:28 2006 [Janelia]
*
* Purpose: Compare <vec1> to <vec2> for equality, by
* comparing each cognate element pair. Both vectors
* are of size <n>. Equality of elements is
* defined by being $\leq$ fractional tolerance <tol>
* for floating point comparisons, and strict equality
* for integer comparisons. Return <eslOK>
* if the vectors are equal, and <eslFAIL> if not.
*
* <esl_vec_FCompare()> and <esl_vec_ICompare()> do the same,
* for float and integer vectors.
*/
int
esl_vec_DCompare(const double *vec1, const double *vec2, int n, double tol)
{
int i;
for (i = 0; i < n; i++) if (esl_DCompare(vec1[i], vec2[i], tol) == eslFAIL) return eslFAIL;
return eslOK;
}
/* The LogGamma() function is rate-limiting in hmmbuild, because it is
* used so heavily in mixture Dirichlet calculations.
* ./configure --with-gsl; [compile test driver]
* ./stats_utest -v
* runs a comparison of time/precision against GSL.
* SRE, Sat May 23 10:04:41 2009, on home Mac:
* LogGamma = 1.29u / N=1e8 = 13 nsec/call
* gsl_sf_lngamma = 1.43u / N=1e8 = 14 nsec/call
*/
static void
utest_LogGamma(ESL_RANDOMNESS *r, int N, int be_verbose)
{
char *msg = "esl_stats_LogGamma() unit test failed";
ESL_STOPWATCH *w = esl_stopwatch_Create();
double *x = malloc(sizeof(double) * N);
double *lg = malloc(sizeof(double) * N);
double *lg2 = malloc(sizeof(double) * N);
int i;
for (i = 0; i < N; i++)
x[i] = esl_random(r) * 100.;
esl_stopwatch_Start(w);
for (i = 0; i < N; i++)
if (esl_stats_LogGamma(x[i], &(lg[i])) != eslOK) esl_fatal(msg);
esl_stopwatch_Stop(w);
if (be_verbose) esl_stopwatch_Display(stdout, w, "esl_stats_LogGamma() timing: ");
#ifdef HAVE_LIBGSL
esl_stopwatch_Start(w);
for (i = 0; i < N; i++) lg2[i] = gsl_sf_lngamma(x[i]);
esl_stopwatch_Stop(w);
if (be_verbose) esl_stopwatch_Display(stdout, w, "gsl_sf_lngamma() timing: ");
for (i = 0; i < N; i++)
if (esl_DCompare(lg[i], lg2[i], 1e-2) != eslOK) esl_fatal(msg);
#endif
free(lg2);
free(lg);
free(x);
esl_stopwatch_Destroy(w);
}
static void
utest_pvectors(void)
{
char *msg = "pvector unit test failed";
double p1[4] = { 0.25, 0.25, 0.25, 0.25 };
double p2[4];
double p3[4];
float p1f[4];
float p2f[4] = { 0.0, 0.5, 0.5, 0.0 };
float p3f[4];
int n = 4;
double result;
esl_vec_D2F(p1, n, p1f);
esl_vec_F2D(p2f, n, p2);
if (esl_vec_DValidate(p1, n, 1e-12, NULL) != eslOK) esl_fatal(msg);
if (esl_vec_FValidate(p1f, n, 1e-7, NULL) != eslOK) esl_fatal(msg);
result = esl_vec_DEntropy(p1, n); if (esl_DCompare(2.0, result, 1e-9) != eslOK) esl_fatal(msg);
result = esl_vec_FEntropy(p1f, n); if (esl_DCompare(2.0, result, 1e-9) != eslOK) esl_fatal(msg);
result = esl_vec_DEntropy(p2, n); if (esl_DCompare(1.0, result, 1e-9) != eslOK) esl_fatal(msg);
result = esl_vec_FEntropy(p2f, n); if (esl_DCompare(1.0, result, 1e-9) != eslOK) esl_fatal(msg);
result = esl_vec_DRelEntropy(p2, p1, n); if (esl_DCompare(1.0, result, 1e-9) != eslOK) esl_fatal(msg);
result = esl_vec_FRelEntropy(p2f, p1f, n); if (esl_DCompare(1.0, result, 1e-9) != eslOK) esl_fatal(msg);
result = esl_vec_DRelEntropy(p1, p2, n); if (result != eslINFINITY) esl_fatal(msg);
result = esl_vec_FRelEntropy(p1f, p2f, n); if (result != eslINFINITY) esl_fatal(msg);
esl_vec_DLog(p2, n);
if (esl_vec_DLogValidate(p2, n, 1e-12, NULL) != eslOK) esl_fatal(msg);
esl_vec_DExp(p2, n);
if (p2[0] != 0.) esl_fatal(msg);
esl_vec_FLog(p2f, n);
if (esl_vec_FLogValidate(p2f, n, 1e-7, NULL) != eslOK) esl_fatal(msg);
esl_vec_FExp(p2f, n);
if (p2f[0] != 0.) esl_fatal(msg);
esl_vec_DCopy(p2, n, p3);
esl_vec_DScale(p3, n, 10.);
esl_vec_DNorm(p3, n);
if (esl_vec_DCompare(p2, p3, n, 1e-12) != eslOK) esl_fatal(msg);
esl_vec_DLog(p3, n);
result = esl_vec_DLogSum(p3, n); if (esl_DCompare(0.0, result, 1e-12) != eslOK) esl_fatal(msg);
esl_vec_DIncrement(p3, n, 2.0);
esl_vec_DLogNorm(p3, n);
if (esl_vec_DCompare(p2, p3, n, 1e-12) != eslOK) esl_fatal(msg);
esl_vec_FCopy(p2f, n, p3f);
esl_vec_FScale(p3f, n, 10.);
esl_vec_FNorm(p3f, n);
if (esl_vec_FCompare(p2f, p3f, n, 1e-7) != eslOK) esl_fatal(msg);
esl_vec_FLog(p3f, n);
result = esl_vec_FLogSum(p3f, n); if (esl_DCompare(0.0, result, 1e-7) != eslOK) esl_fatal(msg);
esl_vec_FIncrement(p3f, n, 2.0);
esl_vec_FLogNorm(p3f, n);
if (esl_vec_FCompare(p2f, p3f, n, 1e-7) != eslOK) esl_fatal(msg);
return;
}
int
main(int argc, char **argv)
{
ESL_GETOPTS *go;
char *msafile;
ESLX_MSAFILE *afp;
ESL_MSA *msa;
float *sqd;
int status;
int nbad;
int nali = 0;
int nbadali = 0;
int nwgt = 0;
int nbadwgt = 0;
int i;
int be_quiet;
int do_gsc;
int do_pb;
int do_blosum;
double maxid;
double tol;
int maxN;
/* Process command line
*/
go = esl_getopts_Create(options);
if (esl_opt_ProcessCmdline(go, argc, argv) != eslOK) esl_fatal("failed to parse cmd line: %s\n", go->errbuf);
if (esl_opt_VerifyConfig(go) != eslOK) esl_fatal("failed to parse cmd line: %s\n", go->errbuf);
if (esl_opt_GetBoolean(go, "-h") == TRUE) {
puts(usage);
puts("\n where options are:");
esl_opt_DisplayHelp(stdout, go, 0, 2, 80); /* 0=all docgroups; 2=indentation; 80=width */
return 0;
}
be_quiet = esl_opt_GetBoolean(go, "-q");
do_blosum = esl_opt_GetBoolean(go, "--blosum");
do_gsc = esl_opt_GetBoolean(go, "--gsc");
do_pb = esl_opt_GetBoolean(go, "--pb");
maxid = esl_opt_GetReal (go, "--id");
tol = esl_opt_GetReal (go, "--tol");
maxN = esl_opt_GetInteger(go, "--maxN");
if (esl_opt_ArgNumber(go) != 1) {
puts("Incorrect number of command line arguments.");
puts(usage);
return 1;
}
msafile = esl_opt_GetArg(go, 1);
esl_getopts_Destroy(go);
/* Weight one or more alignments from input file
*/
if ((status = eslx_msafile_Open(NULL, msafile, NULL, eslMSAFILE_UNKNOWN, NULL, &afp)) != eslOK)
eslx_msafile_OpenFailure(afp, status);
while ( (status = eslx_msafile_Read(afp, &msa)) != eslEOF)
{
if (status != eslOK) eslx_msafile_ReadFailure(afp, status);
if (maxN > 0 && msa->nseq > maxN) { esl_msa_Destroy(msa); continue; }
nali++;
nwgt += msa->nseq;
ESL_ALLOC(sqd, sizeof(float) * msa->nseq);
if (do_gsc) {
esl_msaweight_GSC(msa);
GSCWeights(msa->aseq, msa->nseq, msa->alen, sqd);
} else if (do_pb) {
esl_msaweight_PB(msa);
PositionBasedWeights(msa->aseq, msa->nseq, msa->alen, sqd);
} else if (do_blosum) {
esl_msaweight_BLOSUM(msa, maxid);
BlosumWeights(msa->aseq, msa->nseq, msa->alen, maxid, sqd);
/* workaround SQUID bug: BLOSUM weights weren't renormalized to sum to nseq. */
esl_vec_FNorm (sqd, msa->nseq);
esl_vec_FScale(sqd, msa->nseq, (float) msa->nseq);
}
if (! be_quiet) {
for (i = 0; i < msa->nseq; i++)
fprintf(stdout, "%-20s %.3f %.3f\n",
msa->sqname[i], msa->wgt[i], sqd[i]);
}
nbad = 0;
for (i = 0; i < msa->nseq; i++)
if (esl_DCompare((double) sqd[i], msa->wgt[i], tol) != eslOK)
nbad++;
if (nbad > 0) nbadali++;
nbadwgt += nbad;
if (nbad > 0) printf("%-20s :: alignment shows %d weights that differ (out of %d) \n",
msa->name, nbad, msa->nseq);
esl_msa_Destroy(msa);
free(sqd);
}
eslx_msafile_Close(afp);
if (nbadali == 0)
printf("OK: all weights identical between squid and Easel in %d alignment(s)\n", nali);
else {
printf("%d of %d weights mismatched at (> %f fractional difference)\n",
nbadwgt, nwgt, tol);
printf("involving %d of %d total alignments\n", nbadali, nali);
}
return eslOK;
ERROR:
return status;
}
/* Function: cm_tr_penalties_Validate()
* Date: EPN, Fri Jan 27 14:57:04 2012
*
* Purpose: Validate a CM_TR_PENALTIES object by checking that
* all possible fragments in local mode sum to 1.0
* for the three scenarios: 5' and 3' truncation,
* 5' truncation only and 3' truncation only.
*
* This is an expensive test and was written only to test
* the code that determines fragment probability (really
* only for local mode) in cm_tr_penalties_Create(). It can
* only be run if the <ignore_inserts> flag was set to TRUE
* when cm_tr_penalties_Create() was called. However, in
* real life that inserts should not be ignored, so this
* test should never actually be run except during testing
* (it also is helpful for understanding the logic behind
* the derivation of the truncated begin
* penalties/probabilities).
*
* Returns: eslOK if all checks pass within tolerance level.
* eslFAIL if any check fails, errbuf is filled.
*/
int
cm_tr_penalties_Validate(CM_TR_PENALTIES *trp, CM_t *cm, double tol, char *errbuf)
{
if(! trp->ignored_inserts) ESL_FAIL(eslFAIL, errbuf, "cm_tr_penalties_Validate(), trp->ignored_inserts flag is not TRUE");
/* This is an expensive test of the trp->l_ptyAA values, the truncation
* penalties for local mode alignment. We test each of the three arrays
* in trp->ptyAA, one each for the following three scenarios:
*
* 1. trp->l_ptyAA[TRPENALTY_5P_AND_3P][0..v..M-1]: penalty for state v
* when 5' and 3' truncation are allowed.
* 2. trp->l_ptyAA[TRPENALTY_5P_ONLY][0..v..M-1]: penalty for state v when
* only 5' truncation is allowed.
* 3. trp->l_ptyAA[TRPENALTY_3P_ONLY][0..v..M-1]: penalty for state v when
* only 3' truncation is allowed.
*
* The test is to enumerate all possible g,h fragments in the
* consensus yield 1..clen, for those that can possibly be generated
* in the scenario (^), determine the state t with the smallest
* subtree yield that contains g..h. This is the state at which an
* alignment of a g..h fragment would be rooted. We then add the
* probability of a truncated parsetree rooted at v (that is,
* exp_2(trp->l_ptyAA[][t])) to a growing sum. After all fragments
* are considered the sum should be 1.0. If it is then our
* penalties are valid, if not they're invalid and we computed them
* incorrectly.
*
* (^): When 5' and 3' truncation are both allowed, all fragments can be
* generated, but not all fragments (for most models) can be generated if
* only 5' or 3' truncation is allowed.
*
*/
double sump = 0.; /* the sum, should be 1.0 after all fragments are considered */
int lpos, rpos; /* left and right consensus positions of a parsetree */
int g, h; /* fragment start/stop */
int keep_going; /* break the loop when this is set to FALSE */
int nd, v;
/* test 1: trp->l_ptyAA[TRPENALTY_5P_AND_3P]: */
for(g = 1; g <= cm->clen; g++) {
for(h = g; h <= cm->clen; h++) {
/* determine which node a truncated parsetree from [a..b] would align to,
* this will be lowest node in the model whose subtree spans a..b
*/
nd = cm->nodes-1;
keep_going = TRUE;
while(keep_going) {
if(nd == 0) ESL_FAIL(eslFAIL, errbuf, "cm_tr_penalties_Validate: 5' and 3' test, unable to find node that spans %d..%d\n", g, h);
lpos = cm->emap->lpos[nd];
rpos = cm->emap->rpos[nd];
if(cm->ndtype[nd] != MATP_nd && cm->ndtype[nd] != MATL_nd) lpos++; /* lpos was one less than what we want */
if(cm->ndtype[nd] != MATP_nd && cm->ndtype[nd] != MATR_nd) rpos--; /* rpos was one more than what we want */
if((cm->ndtype[nd] == BIF_nd || cm->ndtype[nd] == MATP_nd || cm->ndtype[nd] == MATL_nd || cm->ndtype[nd] == MATR_nd) &&
(lpos <= g && rpos >= h)) {
keep_going = FALSE;
}
else { nd--; }
}
v = cm->nodemap[nd];
sump += sreEXP2(trp->l_ptyAA[TRPENALTY_5P_AND_3P][v]);
/* printf("LRBOTH g: %3d h: %3d nd: %3d adding %10.5f (%10.5f)\n", g, h, nd, trp->l_ptyAA[TRPENALTY_5P_AND_3P][v], sump); */
}
}
printf("L and R sump: %.5f\n", sump);
if(esl_DCompare(1.0, sump, tol) != eslOK) ESL_FAIL(eslFAIL, errbuf, "cm_tr_penalties_Validate(), 5' and 3' truncation test failed (%g != 1.0)", sump);
/* test 2: trp->l_ptyAA[TRPENALTY_5P_ONLY]: */
sump = 0.;
for(g = 1; g <= cm->clen; g++) {
for(h = g; h <= cm->clen; h++) {
/* determine which node a truncated parsetree from [g..h] would align to,
* this will be lowest node in the model whose subtree spans g..h.
* Since we're only truncating on the left, an alignment from
* g..h may be impossible, only those fragments for which a node exists with
* lpos <= g and rpos==h will be possible.
*/
nd = cm->nodes-1;
keep_going = TRUE;
while(keep_going && nd > 0) {
lpos = cm->emap->lpos[nd];
rpos = cm->emap->rpos[nd];
if(cm->ndtype[nd] != MATP_nd && cm->ndtype[nd] != MATL_nd) lpos++; /* lpos was one less than what we want */
if(cm->ndtype[nd] != MATP_nd && cm->ndtype[nd] != MATR_nd) rpos--; /* rpos was one more than what we want */
if((cm->ndtype[nd] == BIF_nd || cm->ndtype[nd] == MATP_nd || cm->ndtype[nd] == MATL_nd || cm->ndtype[nd] == MATR_nd) &&
(lpos <= g && rpos == h)) {
keep_going = FALSE;
}
else { nd--; }
}
if(keep_going == FALSE) {
v = cm->nodemap[nd];
sump += sreEXP2(trp->l_ptyAA[TRPENALTY_5P_ONLY][v]);
}
}
}
printf("L only sump: %.5f\n", sump);
if(esl_DCompare(1.0, sump, tol) != eslOK) ESL_FAIL(eslFAIL, errbuf, "cm_tr_penalties_Validate(), 5' only truncation test failed (%g != 1.0)", sump);
/* test 3: trp->l_ptyAA[TRPENALTY_3P_ONLY]: */
sump = 0.;
for(g = 1; g <= cm->clen; g++) {
for(h = g; h <= cm->clen; h++) {
/* determine which node a truncated parsetree from [g..h] would align to,
* this will be lowest node in the model whose subtree spans g..h
* since we're only truncating on the right, an alignment from
* g..h may be impossible, only those for which a node exists with
* lpos==g and rpos >= h will be possible.
*/
nd = cm->nodes-1;
keep_going = TRUE;
while(keep_going && nd > 0) {
lpos = cm->emap->lpos[nd];
rpos = cm->emap->rpos[nd];
if(cm->ndtype[nd] != MATP_nd && cm->ndtype[nd] != MATL_nd) lpos++; /* lpos was one less than what we want */
if(cm->ndtype[nd] != MATP_nd && cm->ndtype[nd] != MATR_nd) rpos--; /* rpos was one more than what we want */
if((cm->ndtype[nd] == BIF_nd || cm->ndtype[nd] == MATP_nd || cm->ndtype[nd] == MATL_nd || cm->ndtype[nd] == MATR_nd) &&
(lpos == g && rpos >= h)) {
keep_going = FALSE;
}
else { nd--; }
}
if(keep_going == FALSE) {
v = cm->nodemap[nd];
sump += sreEXP2(trp->l_ptyAA[TRPENALTY_3P_ONLY][v]);
}
}
}
printf("R only sump: %.5f\n", sump);
if(esl_DCompare(1.0, sump, tol) != eslOK) ESL_FAIL(eslFAIL, errbuf, "cm_tr_penalties_Validate(), 3' only truncation test failed (%g != 1.0)", sump);
return eslOK;
}