/* Viterbi score implementation; non-optimized. * * This Viterbi routine is pass through to the generic * Viterbi routine. * * Contents: * 1. Viterbi score implementation. * 2. Benchmark driver. * 3. Unit tests. * 4. Test driver. * 5. Example. * 6. Copyright and license information. * * MSF, Tue Nov 03, 2009 [Janelia] * SVN $Id: vitscore.c 2818 2009-06-03 12:31:02Z eddys $ */ #include "p7_config.h" #include #include #include "easel.h" #include "hmmer.h" #include "impl_dummy.h" /***************************************************************** * 1. Viterbi score implementation *****************************************************************/ /* Function: p7_ViterbiScore() * Synopsis: Calculates Viterbi score. * Incept: MSF, Tue Nov 03, 2009 [Janelia] * * Purpose: Calculates the Viterbi score for sequence of * length residues, using profile . Return the * Viterbi score (in nats) in . * * The score is calculated using the generic Viterbi * routine. * * Args: dsq - digital target sequence, 1..L * L - length of dsq in residues * om - optimized profile * ox - DP matrix * ret_sc - RETURN: Viterbi score (in nats) * * Returns: on success. */ int p7_ViterbiScore(const ESL_DSQ *dsq, int L, const P7_OPROFILE *om, P7_OMX *ox, float *ret_sc) { return p7_GViterbi(dsq, L, om, ox, ret_sc); } /*------------------ end, p7_ViterbiScore() ---------------------*/ /***************************************************************** * 2. Benchmark driver. *****************************************************************/ #ifdef p7VITSCORE_BENCHMARK /* -c, -x are used for debugging, testing. See msvfilter.c for * an explanation. Here -c and -x are the same: both compare * to p7_GViterbi() scores. */ /* gcc -o benchmark-vitscore -std=gnu99 -g -Wall -I.. -L.. -I../../easel -L../../easel -Dp7VITSCORE_BENCHMARK vitscore.c -lhmmer -leasel -lm icc -o benchmark-vitscore -O3 -static -I.. -L.. -I../../easel -L../../easel -Dp7VITSCORE_BENCHMARK vitscore.c -lhmmer -leasel -lm ./benchmark-vitscore runs benchmark ./benchmark-vitscore -N100 -c compare scores to generic impl ./benchmark-vitscore -N100 -x equate scores to exact emulation */ #include "p7_config.h" #include "easel.h" #include "esl_alphabet.h" #include "esl_getopts.h" #include "esl_random.h" #include "esl_randomseq.h" #include "esl_stopwatch.h" #include "hmmer.h" #include "impl_dummy.h" static ESL_OPTIONS options[] = { /* name type default env range toggles reqs incomp help docgroup*/ { "-h", eslARG_NONE, FALSE, NULL, NULL, NULL, NULL, NULL, "show brief help on version and usage", 0 }, { "-c", eslARG_NONE, FALSE, NULL, NULL, NULL, NULL, "-x", "compare scores to generic implementation (debug)", 0 }, { "-s", eslARG_INT, "42", NULL, NULL, NULL, NULL, NULL, "set random number seed to ", 0 }, { "-x", eslARG_NONE, FALSE, NULL, NULL, NULL, NULL, "-c", "equate scores to trusted implementation (debug)", 0 }, { "-L", eslARG_INT, "400", NULL, "n>0", NULL, NULL, NULL, "length of random target seqs", 0 }, { "-N", eslARG_INT, "50000", NULL, "n>0", NULL, NULL, NULL, "number of random target seqs", 0 }, { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 }, }; static char usage[] = "[-options] "; static char banner[] = "benchmark driver for non-optimized ViterbiScore()"; int main(int argc, char **argv) { ESL_GETOPTS *go = esl_getopts_CreateDefaultApp(options, 1, argc, argv, banner, usage); char *hmmfile = esl_opt_GetArg(go, 1); ESL_STOPWATCH *w = esl_stopwatch_Create(); ESL_RANDOMNESS *r = esl_randomness_CreateFast(esl_opt_GetInteger(go, "-s")); 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; int L = esl_opt_GetInteger(go, "-L"); int N = esl_opt_GetInteger(go, "-N"); ESL_DSQ *dsq = malloc(sizeof(ESL_DSQ) * (L+2)); int i; float sc1, sc2; double base_time, bench_time, Mcs; 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"); bg = p7_bg_Create(abc); p7_bg_SetLength(bg, L); gm = p7_profile_Create(hmm->M, abc); p7_ProfileConfig(hmm, bg, gm, L, p7_LOCAL); om = p7_oprofile_Create(gm->M, abc); p7_oprofile_Convert(gm, om); p7_oprofile_ReconfigLength(om, L); p7_oprofile_Logify(om); ox = p7_omx_Create(gm->M, 0, 0); gx = p7_gmx_Create(gm->M, L); /* Get a baseline time: how long it takes just to generate the sequences */ esl_stopwatch_Start(w); for (i = 0; i < N; i++) esl_rsq_xfIID(r, bg->f, abc->K, L, dsq); esl_stopwatch_Stop(w); base_time = w->user; /* Run the benchmark */ esl_stopwatch_Start(w); for (i = 0; i < N; i++) { esl_rsq_xfIID(r, bg->f, abc->K, L, dsq); p7_ViterbiScore (dsq, L, om, ox, &sc1); if (esl_opt_GetBoolean(go, "-c") || esl_opt_GetBoolean(go, "-x")) { p7_GViterbi(dsq, L, gm, gx, &sc2); printf("%.4f %.4f\n", sc1, sc2); } } esl_stopwatch_Stop(w); bench_time = w->user - base_time; Mcs = (double) N * (double) L * (double) gm->M * 1e-6 / (double) bench_time; esl_stopwatch_Display(stdout, w, "# CPU time: "); printf("# M = %d\n", gm->M); printf("# %.1f Mc/s\n", Mcs); free(dsq); 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_stopwatch_Destroy(w); esl_randomness_Destroy(r); esl_getopts_Destroy(go); return 0; } #endif /*p7VITSCORE_BENCHMARK*/ /*------------------- end, benchmark driver ---------------------*/ /***************************************************************** * 3. Unit tests. *****************************************************************/ #ifdef p7VITSCORE_TESTDRIVE #include "esl_random.h" #include "esl_randomseq.h" /* ViterbiScore() unit test * * We can compare these scores to GViterbi() almost exactly; the only * differences should be negligible roundoff errors. Must convert * the optimized profile to lspace, though, rather than pspace. */ static void utest_viterbi_score(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_oprofile_Logify(om); while (N--) { esl_rsq_xfIID(r, bg->f, abc->K, L, dsq); p7_ViterbiScore(dsq, L, om, ox, &sc1); p7_GViterbi (dsq, L, gm, gx, &sc2); if (fabs(sc1-sc2) > 0.001) esl_fatal("viterbi score unit test failed: scores differ"); } free(dsq); p7_hmm_Destroy(hmm); p7_omx_Destroy(ox); p7_gmx_Destroy(gx); p7_profile_Destroy(gm); p7_oprofile_Destroy(om); } #endif /*p7VITSCORE_TESTDRIVE*/ /*-------------------- end, unit tests --------------------------*/ /***************************************************************** * 4. Test driver *****************************************************************/ #ifdef p7VITSCORE_TESTDRIVE /* gcc -g -Wall -std=gnu99 -I.. -L.. -I../../easel -L../../easel -o vitscore_utest -Dp7VITSCORE_TESTDRIVE vitscore.c -lhmmer -leasel -lm ./vitscore_utest */ #include "p7_config.h" #include "easel.h" #include "esl_alphabet.h" #include "esl_getopts.h" #include "hmmer.h" #include "impl_dummy.h" static ESL_OPTIONS options[] = { /* name type default env range toggles reqs incomp help docgroup*/ { "-h", eslARG_NONE, FALSE, NULL, NULL, NULL, NULL, NULL, "show brief help on version and usage", 0 }, { "-s", eslARG_INT, "42", NULL, NULL, NULL, NULL, NULL, "set random number seed to ", 0 }, { "-L", eslARG_INT, "200", NULL, NULL, NULL, NULL, NULL, "size of random sequences to sample", 0 }, { "-M", eslARG_INT, "145", NULL, NULL, NULL, NULL, NULL, "size of random models to sample", 0 }, { "-N", eslARG_INT, "100", NULL, NULL, NULL, NULL, NULL, "number of random sequences to sample", 0 }, { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 }, }; static char usage[] = "[-options]"; static char banner[] = "test driver for the non-optimized implementation"; int main(int argc, char **argv) { ESL_GETOPTS *go = esl_getopts_CreateDefaultApp(options, 0, argc, argv, banner, usage); ESL_RANDOMNESS *r = esl_randomness_CreateFast(esl_opt_GetInteger(go, "-s")); ESL_ALPHABET *abc = NULL; P7_BG *bg = NULL; int M = esl_opt_GetInteger(go, "-M"); int L = esl_opt_GetInteger(go, "-L"); int N = esl_opt_GetInteger(go, "-N"); /* First round of tests for DNA alphabets. */ if ((abc = esl_alphabet_Create(eslDNA)) == NULL) esl_fatal("failed to create alphabet"); if ((bg = p7_bg_Create(abc)) == NULL) esl_fatal("failed to create null model"); utest_viterbi_score(r, abc, bg, M, L, N); utest_viterbi_score(r, abc, bg, 1, L, 10); utest_viterbi_score(r, abc, bg, M, 1, 10); esl_alphabet_Destroy(abc); p7_bg_Destroy(bg); /* Second round of tests for amino alphabets. */ if ((abc = esl_alphabet_Create(eslAMINO)) == NULL) esl_fatal("failed to create alphabet"); if ((bg = p7_bg_Create(abc)) == NULL) esl_fatal("failed to create null model"); utest_viterbi_score(r, abc, bg, M, L, N); utest_viterbi_score(r, abc, bg, 1, L, 10); utest_viterbi_score(r, abc, bg, M, 1, 10); esl_alphabet_Destroy(abc); p7_bg_Destroy(bg); esl_getopts_Destroy(go); esl_randomness_Destroy(r); return eslOK; } #endif /*VITSCORE_TESTDRIVE*/ /*--------------------- end, test driver ------------------------*/ /***************************************************************** * 5. Example *****************************************************************/ #ifdef p7VITSCORE_EXAMPLE /* A minimal example. Also useful for debugging on small HMMs and sequences. gcc -g -Wall -std=gnu99 -I.. -L.. -I../../easel -L../../easel -o example -Dp7VITSCORE_EXAMPLE vitscore.c -lhmmer -leasel -lm ./example */ #include "p7_config.h" #include "easel.h" #include "esl_alphabet.h" #include "esl_sq.h" #include "esl_sqio.h" #include "hmmer.h" #include "impl_dummy.h" int main(int argc, char **argv) { char *hmmfile = argv[1]; char *seqfile = argv[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 sc; 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); p7_oprofile_Logify(om); /* allocate DP matrices, both a generic and an optimized one */ ox = p7_omx_Create(gm->M, 0, sq->n); 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(ox, TRUE); makes the fast DP algorithms dump their matrices p7_gmx_Dump(stdout, gx); dumps a generic DP matrix */ p7_ViterbiScore(sq->dsq, sq->n, om, ox, &sc); printf("viterbi (non-optimized): %.2f nats\n", sc); p7_GViterbi (sq->dsq, sq->n, gm, gx, &sc); printf("viterbi (generic): %.2f nats\n", sc); /* now in a real app, you'd need to convert raw nat scores to final bit * scores, by subtracting the null model score and rescaling. */ /* 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); return 0; } #endif /*p7VITSCORE_EXAMPLE*/ /*-------------------------- end, example ------------------------------*/ /***************************************************************** * HMMER - Biological sequence analysis with profile HMMs * Version 3.0; March 2010 * Copyright (C) 2010 Howard Hughes Medical Institute. * Other copyrights also apply. See the COPYRIGHT file for a full list. * * HMMER is distributed under the terms of the GNU General Public License * (GPLv3). See the LICENSE file for details. *****************************************************************/