tests: test_call_after_construction_with_learning_pathway: fix missin…

…g assert (#2647)

np.allclose call was not asserted since this test was added. The
results for this call changed in
0946af0 (#1639) due to a change in
default for argument loss_spec (formerly loss_function) from None to
MSE in add_linear_learning_pathway and
_create_backpropagation_learning_pathway.

As of this commit, changing these defaults back to None produces the
numeric results that were present when this test was created.

The allclose call also misinterpreted the return value for
Composition.run as producing multiple trial outputs that are instead in
Composition.results

tests/composition/test_composition.py

@@ -122,7 +122,7 @@ def test_call_after_construction_with_learning_pathway(self):
         # Run with learning
         target = c.pathways['LEARNING_PATHWAY'].target
         result = c(inputs={A:[[1],[100]],B:[[2],[1]],target:[[3],[300]]})
-        np.allclose(result, [[[1.], [0.73105858]], [[100.], [0.62507661]]])
+        np.testing.assert_allclose(c.results[-2:], [[[1.], [0.73105858]], [[100.], [0.62768661]]])
 
 
 class TestAddMechanism: