Merge pull request pycaret#4029 from kondziolka9ld/fix_parallel_mode_…

…no_model_trained

Fix: In parallel mode return empty list of models when no models was trained.

pycaret/parallel/fugue_backend.py

@@ -163,7 +163,12 @@ def compare_models(
             as_fugue=True,
             as_local=True,
         ).as_array()
-        res = pd.concat(cloudpickle.loads(x[0]) for x in outputs)
+
+        pd_dataframe_for_models = [cloudpickle.loads(x[0]) for x in outputs]
+        if all(pd_dataframe_for_model.empty for pd_dataframe_for_model in pd_dataframe_for_models):
+            return []
+
+        res = pd.concat(pd_dataframe_for_models)
         res = res.sort_values(sort_col, ascending=asc)
         top = res.head(self._params.get("n_select", 1))
         instance._display_container.append(res.iloc[:, :-1])

tests/test_classification_parallel.py

@@ -1,14 +1,13 @@
 import pycaret.classification as pc
 from pycaret.datasets import get_data
+from pycaret.parallel import FugueBackend
 
 
 def _score_dummy(y_true, y_prob, axis=0):
     return 0.0
 
 
 def test_classification_parallel():
-    from pycaret.parallel import FugueBackend
-
     pc.setup(
         data_func=lambda: get_data("juice", verbose=False),
         target="Purchase",
@@ -46,3 +45,24 @@ def test_classification_parallel():
 
     pc.compare_models(n_select=2, sort="DUMMY", parallel=be)
     pc.pull()
+
+
+def test_classification_parallel_returns_empty_models_list_when_no_model_is_trained():
+    pc.setup(
+        data_func=lambda: get_data("juice", verbose=False),
+        target="Purchase",
+        session_id=0,
+        n_jobs=1,
+        verbose=False,
+        html=False,
+    )
+
+    fconf = {
+        "fugue.rpc.server": "fugue.rpc.flask.FlaskRPCServer",
+        "fugue.rpc.flask_server.host": "localhost",
+        "fugue.rpc.flask_server.port": "3333",
+        "fugue.rpc.flask_server.timeout": "2 sec",
+    }
+
+    res = pc.compare_models(include=[], parallel=FugueBackend("dask", fconf, display_remote=True, batch_size=3, top_only=False))
+    assert (len(res) == 0)