Let save_/load_results() properly handles experiments dtypes (#280)

* bugfix for #277 : load_results now casts all columns in experiments to dtype as specified in metadata

bugfix for load_results. In parsing experiments, the dtype metadata is now correctly applied.  Also expands save)results unittest to cover all dtypes used in storing experiments.

* small typo in test

ema_workbench/util/utilities.py

@@ -64,6 +64,9 @@ def load_results(file_name):
             except TypeError:
                 dtype = pd.api.types.pandas_dtype(dtype)
 
+            if experiments[name].dtype is not dtype:
+                experiments[name] = experiments[name].astype(dtype)
+            # this check is for backward compatability with data stored with 2.4.
             if pd.api.types.is_object_dtype(dtype):
                 experiments[name] = experiments[name].astype("category")
 

test/test_util/test_utilities.py

@@ -41,9 +41,11 @@ def test_save_results(self):
 
         # test for 1d
         nr_experiments = 10000
-        experiments = pd.DataFrame(
-            index=np.arange(nr_experiments), columns={"x": float, "y": float}
+        cases = np.empty(
+            nr_experiments, dtype=[("x", float), ("y", int), ("z", bool), ("q", object)]
         )
+        experiments = pd.DataFrame.from_records(cases)
+        experiments["q"] = experiments["q"].astype("category")
         outcome_q = np.random.rand(nr_experiments, 1)
 
         outcomes = {ScalarOutcome("q").name: outcome_q}
@@ -85,19 +87,19 @@ def test_save_results(self):
 
 class LoadResultsTestCase(unittest.TestCase):
     def test_load_results(self):
-        # test for 1d
-
-        # test for 3d
-
         nr_experiments = 10000
 
         # test for 2d
-        experiments = pd.DataFrame(
-            index=np.arange(nr_experiments), columns={"x": float, "y": float}
+        cases = np.empty(
+            nr_experiments, dtype=[("x", float), ("y", int), ("z", bool), ("q", object)]
         )
+        experiments = pd.DataFrame.from_records(cases)
 
         experiments["x"] = np.random.rand(nr_experiments)
-        experiments["y"] = np.random.rand(nr_experiments)
+        experiments["y"] = np.random.randint(0, 10, size=nr_experiments)
+        experiments["z"] = np.random.randint(0, 1, size=nr_experiments, dtype=bool)
+        experiments["q"] = np.random.randint(0, 10, size=nr_experiments).astype(object)
+        experiments["q"] = experiments["q"].astype("category")
 
         outcome_a = np.zeros((nr_experiments, 1))
 
@@ -111,6 +113,12 @@ def test_load_results(self):
         self.assertTrue(np.all(np.allclose(experiments["x"], loaded_experiments["x"])))
         self.assertTrue(np.all(np.allclose(experiments["y"], loaded_experiments["y"])))
 
+        for name, dtype in experiments.dtypes.items():
+            self.assertTrue(
+                dtype == loaded_experiments[name].dtype,
+                msg=f"{name}, {dtype}, {loaded_experiments[name].dtype}",
+            )
+
         os.remove("../data/test.tar.gz")
 
         # test 3d