Merge branch 'dev'

Project.toml

 name = "TaggingBackends"
 uuid = "e551f703-3b82-4335-b341-d497b48d519b"
 authors = ["François Laurent", "Institut Pasteur"]
-version = "0.7.1"
+version = "0.7.2"
 
 [deps]
 Dates = "ade2ca70-3891-5945-98fb-dc099432e06a"

pyproject.toml

 [tool.poetry]
 name = "TaggingBackends"
-version = "0.7.1"
+version = "0.7.2"
 description = "Backbone for LarvaTagger.jl tagging backends"
 authors = ["François Laurent"]
 

src/taggingbackends/data/dataset.py

@@ -101,12 +101,13 @@ class LarvaDataset:
         train, val, test = torch.utils.data.random_split(TorchDataset(),
                 [ntrain, nval, ntest],
                 generator=self.generator)
-        self._training_set = iter(itertools.cycle(
-            torch.utils.data.DataLoader(train,
-                batch_size=self.batch_size,
-                shuffle=True,
-                generator=g_train,
-                drop_last=True)))
+        if 0 < ntrain:
+            self._training_set = iter(itertools.cycle(
+                torch.utils.data.DataLoader(train,
+                    batch_size=self.batch_size,
+                    shuffle=True,
+                    generator=g_train,
+                    drop_last=True)))
         self._validation_set = iter(
             torch.utils.data.DataLoader(val,
                 batch_size=self.batch_size))
@@ -186,14 +187,14 @@ def subset_size(ntot, train_share, val_share, test_share):
     while ndelta < 0:
         if 0 < val_share:
             nval += 1
-            ndelta -= 1
+            ndelta += 1
         if ndelta < 0:
             if 0 < test_share:
                 ntest += 1
-                ndelta -= 1
+                ndelta += 1
             if ndelta < 0:
                 ntrain += 1
-                ndelta -= 1
+                ndelta += 1
     while 0 < ndelta:
         if 0 < train_share:
             ntrain -= 1

src/taggingbackends/data/labels.py

@@ -8,6 +8,54 @@ TIME_PRECISION = 0.001
 
 labels_file_extension = (".nyxlabel", ".label", ".labels")
 
+"""
+Python interface to JSON *.label* files.
+
+It is a poor parent of Julia struct `Dataset` from `PlanarLarvae.jl`.
+
+The following example function opens such a *.label* file and loads the
+associated track data assumed to be available as a *trx.mat* file.
+It iterates over the tracks defined in the *.label* file. The same tracks are
+expected to be found also in the *trx.mat* file.
+
+```
+import numpy as np
+from sklearn.metrics import confusion_matrix
+
+def retagged_trxmat_confusion_matrix(label_file):
+    # load the .label and associated trx.mat files
+    predicted_labels = Labels(label_file)
+    labels = predicted_labels.labelspec
+    track_data_file = TrxMat(label_file.parent / 'trx.mat')
+    expected_labels = track_data_file.read(labels)
+
+    # iterate the tracks
+    cm = None
+    for run, larva in predicted_labels:
+
+        ypred = np.array([labels.index(label)
+                          for label in predicted_labels[(run, larva)].values()])
+        yexp = np.stack([expected_labels[label][run][larva]
+                         for label in labels])
+        iexp, yexp = np.nonzero(yexp.T==1) # the order of dimensions matters
+        ypred = ypred[iexp] # if the trx.mat file mentions more tags than the
+                            # .label file, some timesteps may appear as untagged
+
+        cm_ = confusion_matrix(yexp, ypred, labels=np.arange(len(labels)))
+        cm = cm_ if cm is None else cm + cm_
+    return cm
+```
+
+Note that unlike the `TrxMat` class, keys in `Labels` objects are
+(runid, trackid) pairs.
+A `Labels` object will typically describe a single run, but can technically
+represent several runs.
+
+The `labelspec` attribute is assumed to be a list, which is valid for *.label*
+generated by automatic tagging.
+Manual tagging will store label names as `Labels.labelspec['names']`, because
+label colors are also stored in the `labelspec` attribute.
+"""
 class Labels:
 
     def __init__(self, labels=None, labelspec=None, metadata=None, units=None,
@@ -23,8 +71,12 @@ class Labels:
         self.labelspec, self.units = labelspec, units
         self._tracking = tracking
         self._input_labels = None
+        #
+        self.filepath = None # unused attribute; may help to manage data
+                             # dependencies in the future
         if isinstance(labels, (str, pathlib.Path)):
-            self.load(labels)
+            self.filepath = pathlib.Path(labels) if isinstance(labels, str) else labels
+            self.load()
 
     @property
     def tracking(self):
@@ -182,10 +234,12 @@ class Labels:
         else:
             json.dump(self, file, cls=LabelEncoder, indent=indent, **kwargs)
 
-    def load(self, file, format="json"):
+    def load(self, file=None, format="json"):
         if format != "json":
             raise ArgumentError(f"unsupported format: {format}")
-        if isinstance(file, str):
+        if file is None:
+            file = self.filepath
+        elif isinstance(file, str):
             file = pathlib.Path(file)
         if isinstance(file, pathlib.Path):
             if file.is_dir():

src/taggingbackends/explorer.py

@@ -181,6 +181,11 @@ Cannot find any Python package in project root directory:
         # Julia logging
         elif line and line[0] in "[┌│└":
             print(line)
+        # Torch logs
+        elif line == '  return torch._C._cuda_getDeviceCount() > 0':
+            # typically follows:
+            # UserWarning: CUDA initialization: CUDA unknown error
+            print(line)
         # tensorflow logs
         elif 26 < len(line):
             # assume line[:26] to be e.g.: 2022-05-17 18:48:15.120981