Fix converting xgboost with user given feature_names

[odedzewi]

No description provided.

[TobyRoseman]

Thanks for the pull request @odedzewi. Please also include a unit test, which fails without your fix but pass with your fix.

[odedzewi]

Hey @TobyRoseman. I added a test to my PR.

Trying it, I ran into this exception:
Unable to load libmodelpackage. Cannot make save spec.

It might be an issue with macOS > 10.6 (I'm running 12.1), as similar tests are skipped in test_boosted_trees_regeression.py:
@unittest.skipIf(_macos_version() >= (10, 16), "rdar://problem/84898245")
Or maybe it's related to this, although I'm running python 3.8, on an intel silicon macbook.

I'm kind of at the edge of my dev capabilities, so hoping you can take it from here.

[TobyRoseman]

You'll need the following change to your unit test in order to get it to run:

 -217,7 +216,8 @@ class BoostedTreeRegressorXGboostTest(unittest.TestCase):
         )
         model = xgboost.train({}, dtrain, 1)
 
-        spec = xgb_converter.convert(model, feature_names=self.feature_names)
+        mlmodel = xgb_converter.convert(model, feature_names=self.feature_names)
+        spec = mlmodel.get_spec()
 
         self.assertEqual(
             sorted(self.feature_names),

However your unit test passes even without your change. So I don't think this is the right test.