make_dataset.py extracts more time segments at the ends of tracks

make_dataset.py passes past_future_extensions=False to taggingbackends.explorer.BackendExplorer.generate_dataset, which results in more time segments sampled near the beginning and end of a track.

Bugfix: finetune_model.py supports dict-like label definitions (with color information) from clf_config.json files.

src/maggotuba/data/make_dataset.py

@@ -52,6 +52,8 @@ def make_dataset(backend, labels_expected=False, trxmat_only=False,
                         # note kwargs['labels'] may be defined, but we dismiss
                         # the input argument, because we need to preserve the
                         # order of the labels (the class indices)
+                        if isinstance(labels, dict):
+                            labels = labels['names']
                         kwargs['labels'] = labels
                         break
 
@@ -59,6 +61,7 @@ def make_dataset(backend, labels_expected=False, trxmat_only=False,
             # generate a larva_dataset_*.hdf5 file in data/interim/{instance}/
             balance = isinstance(balancing_strategy, str) and balancing_strategy.lower() == 'maggotuba'
             out = backend.generate_dataset(backend.raw_data_dir(),
+                                           past_future_extensions=False,
                                            balance=balance, **kwargs)
             print(f"larva_dataset file generated: {out}")