Suppress more numpy warnings from uncertainties (#1350)

This PR is a follow up to
#1070. It
suppresses numpy warnings during uncertainties array creation in more
places in the code. Something changed recently in the numpy code so that
it generates warnings when it used to suppress them, in particular when
using `numpy.vectorize` which wraps user-level Python code inside of
numpy C code and seems to lose context about warning state. See
numpy/numpy#21416 for more information.

Author: wshanks

qiskit_experiments/curve_analysis/curve_analysis.py

@@ -231,7 +231,11 @@ def _run_data_processing(
             )
         processed_values = self.options.data_processor(to_process)
         source["yval"] = unp.nominal_values(processed_values).flatten()
-        source["yerr"] = unp.std_devs(processed_values).flatten()
+        with np.errstate(invalid="ignore"):
+            # For averaged data, the processed std dev will be NaN.
+            # Setting std_devs to NaN will trigger floating point exceptions
+            # which we can ignore. See https://stackoverflow.com/q/75656026
+            source["yerr"] = unp.std_devs(processed_values).flatten()
         source["category"] = category
 
         table = ScatterTable(data=source)

qiskit_experiments/curve_analysis/curve_data.py

@@ -235,10 +235,13 @@ def ufloat_params(self) -> Dict[str, uncertainties.UFloat]:
                 )
             else:
                 # Invalid covariance matrix. Std dev is set to nan, i.e. not computed.
-                ufloat_fitvals = uarray(
-                    nominal_values=[self.params[name] for name in self.var_names],
-                    std_devs=np.full(len(self.var_names), np.nan),
-                )
+                with np.errstate(invalid="ignore"):
+                    # Setting std_devs to NaN will trigger floating point exceptions
+                    # which we can ignore. See https://stackoverflow.com/q/75656026
+                    ufloat_fitvals = uarray(
+                        nominal_values=[self.params[name] for name in self.var_names],
+                        std_devs=np.full(len(self.var_names), np.nan),
+                    )
             # Combine fixed params and fitting variables into a single dictionary
             # Fixed parameter has zero std_dev
             ufloat_params = {}

qiskit_experiments/data_processing/nodes.py

@@ -85,7 +85,10 @@ def _process(self, data: np.ndarray) -> np.ndarray:
 
         reduced_array = np.mean(data, axis=ax)
         nominals = unp.nominal_values(reduced_array)
-        errors = unp.std_devs(reduced_array)
+        with np.errstate(invalid="ignore"):
+            # Setting std_devs to NaN will trigger floating point exceptions
+            # which we can ignore. See https://stackoverflow.com/q/75656026
+            errors = unp.std_devs(reduced_array)
 
         if np.any(np.isnan(errors)):
             # replace empty elements with SEM
@@ -781,7 +784,10 @@ def _process(self, data: np.ndarray) -> np.ndarray:
             p_mean = alpha_posterior[0] / alpha_sum
             p_var = p_mean * (1 - p_mean) / (alpha_sum + 1)
 
-            probabilities[idx] = ufloat(nominal_value=p_mean, std_dev=np.sqrt(p_var))
+            with np.errstate(invalid="ignore"):
+                # Setting std_devs to NaN will trigger floating point exceptions
+                # which we can ignore. See https://stackoverflow.com/q/75656026
+                probabilities[idx] = ufloat(nominal_value=p_mean, std_dev=np.sqrt(p_var))
 
         return probabilities
 

test/data_processing/test_data_processing.py

@@ -418,10 +418,13 @@ def test_json_trained(self):
             unp.nominal_values(loaded_out),
         )
 
-        np.testing.assert_array_almost_equal(
-            unp.std_devs(ref_out),
-            unp.std_devs(loaded_out),
-        )
+        with np.errstate(invalid="ignore"):
+            # Setting std_devs to NaN will trigger floating point exceptions
+            # which we can ignore. See https://stackoverflow.com/q/75656026
+            np.testing.assert_array_almost_equal(
+                unp.std_devs(ref_out),
+                unp.std_devs(loaded_out),
+            )
 
 
 class TestIQSingleAvg(BaseDataProcessorTest):

test/data_processing/test_nodes.py

@@ -43,7 +43,10 @@ class TestAveraging(BaseDataProcessorTest):
 
     def test_simple(self):
         """Simple test of averaging. Standard error of mean is generated."""
-        datum = unp.uarray([[1, 2], [3, 4], [5, 6]], np.full((3, 2), np.nan))
+        with np.errstate(invalid="ignore"):
+            # Setting std_devs to NaN will trigger floating point exceptions
+            # which we can ignore. See https://stackoverflow.com/q/75656026
+            datum = unp.uarray([[1, 2], [3, 4], [5, 6]], np.full((3, 2), np.nan))
 
         node = AverageData(axis=1)
         processed_data = node(data=datum)
@@ -85,16 +88,19 @@ def test_with_error(self):
 
     def test_with_error_partly_non_error(self):
         """Compute error propagation. Some elements have no error."""
-        datum = unp.uarray(
-            [
-                [1, 2, 3, 4, 5, 6],
-                [1, 2, 3, 4, 5, 6],
-            ],
-            [
-                [0.1, 0.2, 0.3, 0.4, 0.5, 0.6],
-                [np.nan, 0.2, 0.3, 0.4, 0.5, 0.6],
-            ],
-        )
+        with np.errstate(invalid="ignore"):
+            # Setting std_devs to NaN will trigger floating point exceptions
+            # which we can ignore. See https://stackoverflow.com/q/75656026
+            datum = unp.uarray(
+                [
+                    [1, 2, 3, 4, 5, 6],
+                    [1, 2, 3, 4, 5, 6],
+                ],
+                [
+                    [0.1, 0.2, 0.3, 0.4, 0.5, 0.6],
+                    [np.nan, 0.2, 0.3, 0.4, 0.5, 0.6],
+                ],
+            )
 
         node = AverageData(axis=1)
         processed_data = node(data=datum)
@@ -130,7 +136,10 @@ def test_iq_averaging(self):
         )
         iq_std = np.full_like(iq_data, np.nan)
 
-        self.create_experiment_data(unp.uarray(iq_data, iq_std), single_shot=True)
+        with np.errstate(invalid="ignore"):
+            # Setting std_devs to NaN will trigger floating point exceptions
+            # which we can ignore. See https://stackoverflow.com/q/75656026
+            self.create_experiment_data(unp.uarray(iq_data, iq_std), single_shot=True)
 
         avg_iq = AverageData(axis=0)
         processed_data = avg_iq(data=np.asarray(self.iq_experiment.data(0)["memory"]))
@@ -188,11 +197,14 @@ def test_simple(self):
             decimal=-8,
         )
 
-        np.testing.assert_array_almost_equal(
-            unp.std_devs(processed),
-            unp.std_devs(expected),
-            decimal=-8,
-        )
+        with np.errstate(invalid="ignore"):
+            # Setting std_devs to NaN will trigger floating point exceptions
+            # which we can ignore. See https://stackoverflow.com/q/75656026
+            np.testing.assert_array_almost_equal(
+                unp.std_devs(processed),
+                unp.std_devs(expected),
+                decimal=-8,
+            )
 
 
 class TestNormalize(QiskitExperimentsTestCase):