Add NaN/Inf checks into check_numpy_array_features (#1145)

* Add NaN/Inf checks into check_numpy_array_features

For normal scenario (and provide a killer switch to turn off the check), all numpy arrays should not have NaN or Inf.

* fix E501

also put report_failure to the end to match the style

* Update doc

course/page/code.py

@@ -1263,7 +1263,8 @@ class PythonCodeQuestion(CodeQuestion, PageBaseWithoutHumanGrading):
 
           feedback.check_numpy_array_sanity(name, num_axes, data)
 
-          feedback.check_numpy_array_features(name, ref, data, report_failure=True)
+          feedback.check_numpy_array_features(name, ref, data, check_finite=True,
+              report_failure=True)
 
           feedback.check_numpy_array_allclose(name, ref, data,
                   accuracy_critical=True, rtol=1e-5, atol=1e-8,

course/page/code_feedback.py

@@ -64,7 +64,9 @@ def check_numpy_array_sanity(self, name, num_axes, data):
                     0, f"'{name}' does not consist of floating point numbers--"
                     f"got: '{data.dtype}'")
 
-    def check_numpy_array_features(self, name, ref, data, report_failure=True):
+    def check_numpy_array_features(self, name, ref, data, check_finite=True,
+            report_failure=True):
+
         import numpy as np
         assert isinstance(ref, np.ndarray)
 
@@ -91,6 +93,12 @@ def bad(msg):
                     f"'{name}' does not have correct data type--"
                     f"got: '{data.dtype}', expected: '{ref.dtype}'")
 
+        if check_finite:
+            if np.any(np.isnan(data)):
+                return bad(f"'{name}' contains NaN")
+            if np.any(np.isinf(data)):
+                return bad(f"'{name}' contains Inf")
+
         return True
 
     def check_numpy_array_allclose(self, name, ref, data, accuracy_critical=True,