Merge branch 'ff384' into 'master'

New flip-flop, size384

See merge request algorithm/flappie!8

CMakeLists.txt

@@ -23,7 +23,7 @@ set (CPACK_PACKAGE_DESCRIPTION_FILE "${PROJECT_SOURCE_DIR}/README.md")
 #set (CPACK_RESOURCE_FILE_LICENSE "${PROJECT_SOURCE_DIR}/LICENSE.md")
 set (CPACK_PACKAGE_VERSION_MAJOR 2)
 set (CPACK_PACKAGE_VERSION_MINOR 1)
-set (CPACK_PACKAGE_VERSION_PATCH 1)
+set (CPACK_PACKAGE_VERSION_PATCH 2)
 
 # Get the latest abbreviated commit hash of the working branch
 execute_process(
@@ -155,12 +155,13 @@ add_executable(flappie_unittest
 target_include_directories(flappie_unittest PUBLIC "src/test" "src")
 target_link_libraries(flappie_unittest flappie_static ${BLAS} ${HDF5} m cunit)
 
-set (READSDIR ${PROJECT_SOURCE_DIR}/reads)
+set (READSDIR ${PROJECT_SOURCE_DIR}/reads/single/)
 set (TESTREAD "single/de1508c4-755b-489e-9ffb-51af35c9a7e6.fast5")
 set (ENV{OPENBLAS_NUM_THREADS} 1)
 
 add_test(NAME unittest WORKING_DIRECTORY ${PROJECT_SOURCE_DIR}/src/test/ COMMAND flappie_unittest)
 add_test(test_flappie_call flappie ${READSDIR})
+add_test(test_flappie_vitonly flappie --viterbi ${READSDIR})
 add_test(test_flappie_licence flappie --licence)
 add_test(test_flappie_license flappie --license)
 add_test(test_flappie_help flappie --help)

RELEASES.md

@@ -2,10 +2,15 @@
 
 Flappie is a prototype basecaller for the sequence platforms produced by Oxford Nanopore Technologies.
 
+##  2.1.2 ()
+Further improvements to Flip-flop and RLE models
+- Accuracy improvement, typically +1.5% over Guppy Hac v3.4
+- Performance reduced
+
 ##  2.1.1 (_Solea solea_)
 Improved Flip-flop and RLE models
 - Accuracy improvements, typically > +1%
-- Performance improvements, approximated 2x faster
+- Performance improvements, approximately 2x faster
 
 ##  2.1.0 (_Lepidopsetta bilineata_)
 _Runnie(r)_, improved Run-Length Encoding.

src/flappie.c

@@ -51,6 +51,9 @@ static struct argp_option options[] = {
     {"licence", 10, 0, 0, "Print licensing information"},
     {"license", 11, 0, OPTION_ALIAS, "Print licensing information"},
     {"segmentation", 3, "chunk:percentile", 0, "Chunk size and percentile for variance based segmentation"},
+    {"viterbi", 'v', 0, 0, "Use viterbi decoding only"},
+    {"no-viterbi", 8, 0, OPTION_ALIAS, "Use forward-backward followed by viterbi"},
+    {"fb", 9, 0, OPTION_ALIAS, "Use forward-backward followed by viterbi"},
     {"hdf5-compression", 12, "level", 0,
      "Gzip compression level for HDF5 output (0:off, 1: quickest, 9: best)"},
     {"hdf5-chunk", 13, "size", 0, "Chunk size for HDF5 output"},
@@ -77,6 +80,7 @@ struct arguments {
     int trim_end;
     int varseg_chunk;
     float varseg_thresh;
+    bool viterbi_only;
     char ** files;
     bool uuid;
 };
@@ -95,6 +99,7 @@ static struct arguments args = {
     .trim_end = 10,
     .varseg_chunk = 100,
     .varseg_thresh = 0.0f,
+    .viterbi_only = false,
     .files = NULL,
     .uuid = true
 };
@@ -162,6 +167,9 @@ static error_t parse_arg(int key, char * arg, struct  argp_state * state){
     case 'T':
         args.trace = arg;
         break;
+    case 'v':
+        args.viterbi_only = true;
+        break;
     case 3:
         args.varseg_chunk = atoi(strtok(arg, ":"));
         next_tok = strtok(NULL, ":");
@@ -176,6 +184,10 @@ static error_t parse_arg(int key, char * arg, struct  argp_state * state){
 	args.temperature = atof(arg);
 	assert(isfinite(args.temperature) && args.temperature > 0.0f);
         break;
+    case 8:
+    case 9:
+        args.viterbi_only = false;
+        break;
     case 10:
     case 11:
         ret = fputs(flappie_licence_text, stdout);
@@ -241,7 +253,11 @@ static struct _raw_basecall_info calculate_post(char * filename, enum model_type
 
     float score = NAN;
 
-    flappie_matrix posterior = transpost_crf_flipflop(trans_weights, true);
+    flappie_matrix posterior = trans_weights;
+    if( ! args.viterbi_only){
+        posterior = transpost_crf_flipflop(trans_weights, true);
+        free(trans_weights);
+    }
     score = decode_crf_flipflop(posterior, false, path, qpath);
     size_t path_nidx = change_positions(path, nblock, path_idx);
 
@@ -259,7 +275,6 @@ static struct _raw_basecall_info calculate_post(char * filename, enum model_type
     free(qpath);
     free(path_idx);
     free(path);
-    trans_weights = free_flappie_matrix(trans_weights);
     const size_t basecall_length = strlen(basecall);
 
     return (struct _raw_basecall_info) {

src/models/flipflop5_r941native.h

@@ -1 +1 @@
-flipflop5_20200220_r941native_69216729.mdl
+flipflop5_202003_r941native_5b1c1eee.mdl