/* ViterbiFilter() unit test * * We can check that scores are identical (within machine error) to * scores of generic DP with scores rounded the same way. Do this for * a random model of length <M>, for <N> test sequences of length <L>. * * We assume that we don't accidentally generate a high-scoring random * sequence that overflows ViterbiFilter()'s limited range. * */ static void utest_viterbi_filter(ESL_RANDOMNESS *r, ESL_ALPHABET *abc, P7_BG *bg, int M, int L, int N) { P7_HMM *hmm = NULL; P7_PROFILE *gm = NULL; P7_OPROFILE *om = NULL; ESL_DSQ *dsq = malloc(sizeof(ESL_DSQ) * (L+2)); P7_OMX *ox = p7_omx_Create(M, 0, 0); P7_GMX *gx = p7_gmx_Create(M, L); float sc1, sc2; p7_oprofile_Sample(r, abc, bg, M, L, &hmm, &gm, &om); p7_profile_SameAsVF(om, gm); /* round and scale the scores in <gm> the same as in <om> */ #if 0 p7_oprofile_Dump(stdout, om); // dumps the optimized profile p7_omx_SetDumpMode(stdout, ox, TRUE); // makes the fast DP algorithms dump their matrices #endif while (N--) { esl_rsq_xfIID(r, bg->f, abc->K, L, dsq); p7_ViterbiFilter(dsq, L, om, ox, &sc1); p7_GViterbi (dsq, L, gm, gx, &sc2); #if 0 p7_gmx_Dump(stdout, gx); // dumps a generic DP matrix #endif sc2 /= om->scale_w; sc2 -= 3.0; if (fabs(sc1-sc2) > 0.001) esl_fatal("viterbi filter unit test failed: scores differ (%.2f, %.2f)", sc1, sc2); } free(dsq); p7_hmm_Destroy(hmm); p7_omx_Destroy(ox); p7_gmx_Destroy(gx); p7_profile_Destroy(gm); p7_oprofile_Destroy(om); }
int main(int argc, char **argv) { ESL_GETOPTS *go = esl_getopts_CreateDefaultApp(options, 2, argc, argv, banner, usage); char *hmmfile = esl_opt_GetArg(go, 1); char *seqfile = esl_opt_GetArg(go, 2); ESL_ALPHABET *abc = NULL; ESL_RANDOMNESS *rng = esl_randomness_CreateFast(0); P7_HMMFILE *hfp = NULL; P7_HMM *hmm = NULL; P7_BG *bg = NULL; P7_PROFILE *gm = NULL; P7_OPROFILE *om = NULL; P7_GMX *gx = NULL; P7_OMX *fwd = NULL; P7_TRACE *tr = NULL; ESL_SQ *sq = NULL; ESL_SQFILE *sqfp = NULL; int format = eslSQFILE_UNKNOWN; int N = esl_opt_GetInteger(go, "-N"); int i; float vsc, fsc, tsc; char errbuf[eslERRBUFSIZE]; int status; /* Read in one HMM */ if (p7_hmmfile_Open(hmmfile, NULL, &hfp) != eslOK) p7_Fail("Failed to open HMM file %s", hmmfile); if (p7_hmmfile_Read(hfp, &abc, &hmm) != eslOK) p7_Fail("Failed to read HMM"); /* Read in one sequence */ sq = esl_sq_CreateDigital(abc); status = esl_sqfile_Open(seqfile, format, NULL, &sqfp); if (status == eslENOTFOUND) p7_Fail("No such file."); else if (status == eslEFORMAT) p7_Fail("Format unrecognized."); else if (status == eslEINVAL) p7_Fail("Can't autodetect stdin or .gz."); else if (status != eslOK) p7_Fail("Open failed, code %d.", status); if (esl_sqio_Read(sqfp, sq) != eslOK) p7_Fail("Failed to read sequence"); /* create default null model, then create and optimize profile */ bg = p7_bg_Create(abc); p7_bg_SetLength(bg, sq->n); gm = p7_profile_Create(hmm->M, abc); p7_ProfileConfig(hmm, bg, gm, sq->n, p7_LOCAL); om = p7_oprofile_Create(gm->M, abc); p7_oprofile_Convert(gm, om); if (esl_opt_GetBoolean(go, "-p")) p7_oprofile_Dump(stdout, om); fwd = p7_omx_Create(gm->M, sq->n, sq->n); gx = p7_gmx_Create(gm->M, sq->n); tr = p7_trace_Create(); if (esl_opt_GetBoolean(go, "-m") == TRUE) p7_omx_SetDumpMode(stdout, fwd, TRUE); p7_GViterbi(sq->dsq, sq->n, gm, gx, &vsc); p7_Forward (sq->dsq, sq->n, om, fwd, &fsc); for (i = 0; i < N; i++) { p7_StochasticTrace(rng, sq->dsq, sq->n, om, fwd, tr); p7_trace_Score(tr, sq->dsq, gm, &tsc); if (esl_opt_GetBoolean(go, "-t") == TRUE) p7_trace_Dump(stdout, tr, gm, sq->dsq); if (p7_trace_Validate(tr, abc, sq->dsq, errbuf) != eslOK) p7_Die("trace %d fails validation:\n%s\n", i, errbuf); printf("Sampled trace: %.4f nats\n", tsc); p7_trace_Reuse(tr); } printf("Forward score: %.4f nats\n", fsc); printf("Viterbi score: %.4f nats\n", vsc); /* cleanup */ esl_sq_Destroy(sq); esl_sqfile_Close(sqfp); p7_trace_Destroy(tr); p7_omx_Destroy(fwd); p7_gmx_Destroy(gx); p7_oprofile_Destroy(om); p7_profile_Destroy(gm); p7_bg_Destroy(bg); p7_hmm_Destroy(hmm); p7_hmmfile_Close(hfp); esl_randomness_Destroy(rng); esl_alphabet_Destroy(abc); esl_getopts_Destroy(go); return 0; }
int main(int argc, char **argv) { ESL_GETOPTS *go = p7_CreateDefaultApp(options, 2, argc, argv, banner, usage); char *hmmfile = esl_opt_GetArg(go, 1); char *seqfile = esl_opt_GetArg(go, 2); ESL_ALPHABET *abc = NULL; P7_HMMFILE *hfp = NULL; P7_HMM *hmm = NULL; P7_BG *bg = NULL; P7_PROFILE *gm = NULL; P7_OPROFILE *om = NULL; P7_OMX *ox = NULL; P7_GMX *gx = NULL; ESL_SQ *sq = NULL; ESL_SQFILE *sqfp = NULL; int format = eslSQFILE_UNKNOWN; float msvraw, nullsc, msvscore; float graw, gscore; double P, gP; int status; /* Read in one HMM */ if (p7_hmmfile_OpenE(hmmfile, NULL, &hfp, NULL) != eslOK) p7_Fail("Failed to open HMM file %s", hmmfile); if (p7_hmmfile_Read(hfp, &abc, &hmm) != eslOK) p7_Fail("Failed to read HMM"); /* Open sequence file for reading */ sq = esl_sq_CreateDigital(abc); status = esl_sqfile_Open(seqfile, format, NULL, &sqfp); if (status == eslENOTFOUND) p7_Fail("No such file."); else if (status == eslEFORMAT) p7_Fail("Format unrecognized."); else if (status == eslEINVAL) p7_Fail("Can't autodetect stdin or .gz."); else if (status != eslOK) p7_Fail("Open failed, code %d.", status); /* create default null model, then create and optimize profile */ bg = p7_bg_Create(abc); p7_bg_SetLength(bg, sq->n); gm = p7_profile_Create(hmm->M, abc); p7_ProfileConfig(hmm, bg, gm, sq->n, p7_LOCAL); om = p7_oprofile_Create(gm->M, abc); p7_oprofile_Convert(gm, om); /* allocate DP matrices, both a generic and an optimized one */ ox = p7_omx_Create(gm->M, 0, 0); /* one row version */ gx = p7_gmx_Create(gm->M, sq->n); /* Useful to place and compile in for debugging: p7_oprofile_Dump(stdout, om); dumps the optimized profile p7_omx_SetDumpMode(stdout, ox, TRUE); makes the fast DP algorithms dump their matrices p7_gmx_Dump(stdout, gx, p7_DEFAULT); dumps a generic DP matrix p7_oprofile_SameMSV(om, gm); */ p7_omx_SetDumpMode(stdout, ox, TRUE); while ((status = esl_sqio_Read(sqfp, sq)) == eslOK) { p7_oprofile_ReconfigLength(om, sq->n); p7_ReconfigLength(gm, sq->n); p7_bg_SetLength(bg, sq->n); p7_omx_GrowTo(ox, om->M, 0, sq->n); p7_gmx_GrowTo(gx, gm->M, sq->n); p7_MSVFilter (sq->dsq, sq->n, om, ox, &msvraw); p7_bg_NullOne (bg, sq->dsq, sq->n, &nullsc); msvscore = (msvraw - nullsc) / eslCONST_LOG2; P = esl_gumbel_surv(msvscore, om->evparam[p7_MMU], om->evparam[p7_MLAMBDA]); p7_GMSV(sq->dsq, sq->n, gm, gx, 2.0, &graw); gscore = (graw - nullsc) / eslCONST_LOG2; gP = esl_gumbel_surv(gscore, gm->evparam[p7_MMU], gm->evparam[p7_MLAMBDA]); if (esl_opt_GetBoolean(go, "-1")) { printf("%-30s %-20s %9.2g %7.2f %9.2g %7.2f\n", sq->name, hmm->name, P, msvscore, gP, gscore); } else if (esl_opt_GetBoolean(go, "-P")) { /* output suitable for direct use in profmark benchmark postprocessors: */ printf("%g %.2f %s %s\n", P, msvscore, sq->name, hmm->name); } else { printf("target sequence: %s\n", sq->name); printf("msv filter raw score: %.2f nats\n", msvraw); printf("null score: %.2f nats\n", nullsc); printf("per-seq score: %.2f bits\n", msvscore); printf("P-value: %g\n", P); printf("GMSV raw score: %.2f nats\n", graw); printf("GSMV per-seq score: %.2f bits\n", gscore); printf("GSMV P-value: %g\n", gP); } esl_sq_Reuse(sq); } /* cleanup */ esl_sq_Destroy(sq); esl_sqfile_Close(sqfp); p7_omx_Destroy(ox); p7_gmx_Destroy(gx); p7_oprofile_Destroy(om); p7_profile_Destroy(gm); p7_bg_Destroy(bg); p7_hmm_Destroy(hmm); p7_hmmfile_Close(hfp); esl_alphabet_Destroy(abc); esl_getopts_Destroy(go); return 0; }