Handle estimator tags in wrapper classes

Allows setting some parameters of the SampleWrapper and CheckpointWrapper via estimator tags. bob.bio.base#143 (closed)

bob/pipelines/wrappers.py

@@ -495,7+495,7 @@
         estimator,
         fit_tag=None,
         transform_tag=None,
         **kwargs,
     ):
         super().__init__(**kwargs)
         self.estimator = estimator
@@ -610,7+610,7 @@
         return {"stateless": True, "requires_fit": False}
 
 
+def get_default_tags():
+    """Returns the default tags of a Transformer.
+
+    Relies on the tags API of sklearn.
+
+    Specify tags in ``Transformer._more_tags``:
+
+    .. code-block:: py
+
+        def _more_tags(self):
+            return {"bob_input": "annotations"}
+
+    Retrieve all the tags with ``Transformer._get_tags``.
+    """
+    return {
+        "bob_is_checkpointable": True,
+        "bob_is_daskable": True,
+        "bob_input": "data",
+        "bob_extra_input": [],
+        "bob_output": "data",
+        "bob_load_fct": bob.io.base.load,
+        "bob_save_fct": bob.io.base.save,
+    }
+
+
 def wrap(bases, estimator=None, **kwargs):
     """Wraps several estimators inside each other.
 
@@ -690,6 +715,7 @@ def wrap(bases, estimator=None, **kwargs):
     return estimator
 
 
+
 def dask_tags(estimator):
     """Recursively collects resource_tags in dasked estimators."""
     tags = {}