Update mlflow with using resolve_tags (#6746)

* Update mlflow.py

#6745 adds additional info about the run, as in the native API

* Update mlflow.py

trying to fix some backward compatibility issues with `resolve_tags`

* wip on backward compatibility

added a default for `getattr` in case the `registry` object exists, but has no proper attribute (weird case but who knows...)

* fix pep

* impoert

* fix registry import

* try fix failing tests

removed the first if statement, so that `resolve_tags` would be defined either case

* fix formatting

Co-authored-by: Jirka Borovec <jirka.borovec@seznam.cz>

pytorch_lightning/loggers/mlflow.py

@@ -30,10 +30,23 @@
 try:
     import mlflow
     from mlflow.tracking import MlflowClient
+    from mlflow.tracking import context
 # todo: there seems to be still some remaining import error with Conda env
 except ImportError:
     _MLFLOW_AVAILABLE = False
-    mlflow, MlflowClient = None, None
+    mlflow, MlflowClient, context = None, None, None
+
+
+# before v1.1.0
+if hasattr(context, 'resolve_tags'):
+    from mlflow.tracking.context import resolve_tags
+# since v1.1.0
+elif hasattr(context, 'registry'):
+    from mlflow.tracking.context.registry import resolve_tags
+else:
+
+    def resolve_tags(tags=None):
+        return tags
 
 
 class MLFlowLogger(LightningLoggerBase):
@@ -140,7 +153,7 @@ def experiment(self) -> MlflowClient:
                 )
 
         if self._run_id is None:
-            run = self._mlflow_client.create_run(experiment_id=self._experiment_id, tags=self.tags)
+            run = self._mlflow_client.create_run(experiment_id=self._experiment_id, tags=resolve_tags(self.tags))
             self._run_id = run.info.run_id
         return self._mlflow_client