reduce test_reports tols where possible

Author: cattabiani

tests/integration-e2e/test_reports.py

@@ -224,14 +224,17 @@ def test_reports_compartment_vs_summation_reference_compartment_set(create_tmp_s
     reference_dir = V5_SONATA / "reference" / "reports" if is_v5_sonata else RINGTEST_DIR / "reference" / "reports"
 
     nd.run()
+    loose_tols = {"rtol": 1e-6, "atol": 1e-6}
 
     # compartment vs summation
     for var in ["v", "i_membrane", "pas"]:
         r_compartment = ReportReader(output_dir / f"compartment_{var}.h5")
         r_summation = ReportReader(output_dir / f"summation_{var}.h5")
 
         r_compartment.convert_to_summation()
-        assert r_compartment == r_summation, f"The summation-converted-compartment:\n{r_compartment}\ndiffers from the summation one:\n{r_summation}"
+        # summation after report printing loses accuracy due to truncation
+        # We use the loose tols in that case
+        assert r_compartment.allclose(r_summation, **loose_tols), f"The summation-converted-compartment:\n{r_compartment}\ndiffers from the summation one:\n{r_summation}"
 
     # summation vs summation. Variable reordering
     r_summation_i_membrane_IClamp = ReportReader(output_dir / f"summation_i_membrane_IClamp.h5")
@@ -244,15 +247,18 @@ def test_reports_compartment_vs_summation_reference_compartment_set(create_tmp_s
     r_summation_IClamp = ReportReader(output_dir / f"summation_IClamp.h5")
     r_compartment_i_membrane = ReportReader(output_dir / f"compartment_i_membrane.h5")
     r_summation_i_membrane_IClamp_manual = r_compartment_i_membrane+ r_summation_IClamp
-    assert r_summation_i_membrane_IClamp == r_summation_i_membrane_IClamp_manual, "Summation report does not match manual addition of compartment_i_membrane and summation_IClamp reports."
+    # summation after report printing loses accuracy due to truncation
+    # We use the loose tols in that case
+    assert r_summation_i_membrane_IClamp.allclose(r_summation_i_membrane_IClamp_manual, **loose_tols), "Summation report does not match manual addition of compartment_i_membrane and summation_IClamp reports."
 
     # Compare files to reference. Since the reference is fixed, this is also a comparison neuron vs coreneuron
     for ref_file in reference_dir.glob("*.h5"):
         r_reference = ReportReader(ref_file)   
         file = output_dir / ref_file.name 
         r = ReportReader(file)
 
-        assert r == r_reference, f"The reports differ:\n{file}\n{ref_file}"
+        # coreneuron does not have exactly the same results, we use the loose tols in that case
+        assert r.allclose(r_reference, **(loose_tols if SimConfig.use_coreneuron else {})), f"The reports differ:\n{file}\n{ref_file}"
 
     # compartment vs compartment_set
     # magic list of positions in the full compartment list. It was done by hand because there isn't a clear cut way

tests/utils.py

@@ -443,6 +443,14 @@ def __init__(self, file: str):
             self.populations[name] = (node_ids, df)
 
     def __eq__(self, other: object) -> bool:
+        return self.allclose(other)
+
+    def allclose(self, other: object, rtol=1e-16, atol=1e-16) -> bool:
+        """
+        Compare two ReportReader instances for approximate equality.
+        
+        Rtol and atol are the relative and absolute tolerances. The default values are
+        the standards for numpy.allclose."""
         if not isinstance(other, ReportReader):
             return NotImplemented
 
@@ -458,7 +466,7 @@ def __eq__(self, other: object) -> bool:
 
             # coreneuron has sometimes garbage for the first line
             # erro thresholds as for old bb5 itegration report tests
-            if not np.allclose(df1.values[1:], df2.values[1:], rtol=1e-6, atol=1e-6):
+            if not np.allclose(df1.values[1:], df2.values[1:], rtol=rtol, atol=atol):
                 return False
 
         return True