make min nb models arg, small fixes

argparse_dicts.py

@@ -291,6 +291,12 @@ def get_compile_model_parser():
         'default': ''  # need it to be existent; passed to production pipeline
     })
 
+    min_nb_models = ('--min-nb-models', {
+        'type': int,
+        'default': 25,
+        'help': 'Minimum number of kmer models to include in the multi-network'
+    })
+
     for arg in (kmer_list, target_16S, nn_directory, out_model):
         parser.add_argument(arg[0], **arg[1])
     parser.add_argument('--train-required', action='store_true',

inference/InferenceModel.py

@@ -25,7 +25,7 @@ class InferenceModel(object):
 
     def predict(self, read, kmer):
         model_output = self._model_dict[kmer](read)
-        return np.any(model_output > 5.0)  # todo make threshold feature?
+        return np.any(model_output > 6.2)  # todo make threshold feature?
 
         # # For fixed batch sizes
         # read, last_idx = read

inference/compile_model.py

@@ -114,7 +114,7 @@ def main(args):
         kmer_size = 8
         kmer_candidates_dict = fa2kmers(args.target_16S, kmer_size)  # Collect k-mers per sequence in target fasta
         if args.train_required:
-            kmers = get_kmer_candidates_16S(kmer_candidates_dict, 10, 0.0001)
+            kmers = get_kmer_candidates_16S(kmer_candidates_dict, args.min_nb_models, 0.0001)
             kmers_no_models = [km for km in kmers if km not in available_mod_kmers]
             if len(kmers_no_models):  # train additional models, if required
                 args.kmer_list = kmers_no_models
@@ -127,7 +127,7 @@ def main(args):
                         warnings.warn(f'model generation failed for {km}, see {args.out_dir}logs. Continuing compilation without it.')
                         kmers.remove(km)
         else:  # filter out k-mers for which no stored model exists
-            kmers = get_kmer_candidates_16S(kmer_candidates_dict, 10, 0.0001, filter_list=available_mod_kmers)
+            kmers = get_kmer_candidates_16S(kmer_candidates_dict, args.min_nb_models, 0.0001, filter_list=available_mod_kmers)
             if not len(kmers):
                 raise ValueError('Sequences do not contain any of available models!')
 

threshold_search/optimize_thresholds.sf

@@ -48,7 +48,7 @@ rule compile_model:
         species_seq_fasta='{{target_fastas_dir}}{species}.fasta'
     output:
         model='{{model_dir}}{species}/compiled.mod'
-    threads: workflow.cores
+    threads: max(workflow.cores // len(species_list), 4)
     shell:
         """
         python {__baseless__} compile_model \