Upgrade unittest equality method (#1132)

### Summary

Current implementation of equality check, i.e.
`QiskitExperimentsTestCase.json_equiv`, is not readable and scalable
because it implements equality check logic for different types in a
single method. This PR adds new test module `test/extended_equality.py`
which implements new equality check dispatcher `is_equivalent`.

Developers no longer need to specify `check_func` in the
`assertRoundTripSerializable` and `assertRoundTripPickle` methods unless
they define custom class for a specific unittest. This simplifies
unittests and improves readability of equality check logic (and test
becomes more trustable).

This PR adds new software dependency in develop;
[multimethod](https://pypi.org/project/multimethod/)

Among several similar packages, this is chosen in favor of
- its license type (Apache License, Version 2.0) 
- syntax compatibility with `functools.singledispatch`
- support for subscripted generics in `typings`, e.g. `Union`

---------

Co-authored-by: Helena Zhang <[email protected]>

Author: nkanazawa1989

releasenotes/notes/add-test-equality-checker-dbe5762d2b6a967f.yaml

@@ -0,0 +1,27 @@
+---
+developer:
+  - |
+    Added the :meth:`QiskitExperimentsTestCase.assertEqualExtended` method for generic equality checks
+    of Qiskit Experiments class instances in unittests. This is a drop-in replacement of 
+    calling the assertTrue with :meth:`QiskitExperimentsTestCase.json_equiv`.
+    Note that some Qiskit Experiments classes may not officially implement equality check logic,
+    although objects may be compared during unittests. Extended equality check is used
+    for such situations.
+  - |
+    The following unittest test case methods will be deprecated:
+    
+      * :meth:`QiskitExperimentsTestCase.json_equiv`
+      * :meth:`QiskitExperimentsTestCase.ufloat_equiv`
+      * :meth:`QiskitExperimentsTestCase.analysis_result_equiv`
+      * :meth:`QiskitExperimentsTestCase.curve_fit_data_equiv`
+      * :meth:`QiskitExperimentsTestCase.experiment_data_equiv`
+    
+    One can now use the :func:`~test.extended_equality.is_equivalent` function instead.
+    This function internally dispatches the logic for equality check.
+  - |
+    The default behavior of :meth:`QiskitExperimentsTestCase.assertRoundTripSerializable` and 
+    :meth:`QiskitExperimentsTestCase.assertRoundTripPickle` when `check_func` is not 
+    provided was upgraded. These methods now compare the decoded instance with
+    :func:`~test.extended_equality.is_equivalent`, rather than 
+    delegating to the native `assertEqual` unittest method.
+    One writing a unittest for serialization no longer need to explicitly set checker function.

requirements-dev.txt

@@ -16,6 +16,7 @@ qiskit-aer>=0.11.0
 pandas>=1.1.5
 cvxpy>=1.1.15
 pylatexenc
+multimethod
 scikit-learn
 sphinx-copybutton
 # Pin versions below because of build errors

test/base.py

@@ -13,27 +13,22 @@
 Qiskit Experiments test case class
 """
 
-import dataclasses
 import json
 import pickle
 import warnings
 from typing import Any, Callable, Optional
 
-import numpy as np
 import uncertainties
-from lmfit import Model
 from qiskit.test import QiskitTestCase
-from qiskit_experiments.data_processing import DataAction, DataProcessor
-from qiskit_experiments.framework.experiment_data import ExperimentStatus
+from qiskit.utils.deprecation import deprecate_func
+
 from qiskit_experiments.framework import (
     ExperimentDecoder,
     ExperimentEncoder,
     ExperimentData,
-    BaseExperiment,
-    BaseAnalysis,
 )
-from qiskit_experiments.visualization import BaseDrawer
-from qiskit_experiments.curve_analysis.curve_data import CurveFitResult
+from qiskit_experiments.framework.experiment_data import ExperimentStatus
+from .extended_equality import is_equivalent
 
 
 class QiskitExperimentsTestCase(QiskitTestCase):
@@ -76,15 +71,52 @@ def assertExperimentDone(
             msg="All threads are executed but status is not DONE. " + experiment_data.errors(),
         )
 
-    def assertRoundTripSerializable(self, obj: Any, check_func: Optional[Callable] = None):
+    def assertEqualExtended(
+        self,
+        first: Any,
+        second: Any,
+        *,
+        msg: Optional[str] = None,
+        strict_type: bool = False,
+    ):
+        """Extended equality assertion which covers Qiskit Experiments classes.
+
+        .. note::
+            Some Qiskit Experiments class may intentionally avoid implementing
+            the equality dunder method, or may be used in some unusual situations.
+            These are mainly caused by to JSON round trip situation, and some custom classes
+            doesn't guarantee object equality after round trip.
+            This assertion function forcibly compares input two objects with
+            the custom equality checker, which is implemented for unittest purpose.
+
+        Args:
+            first: First object to compare.
+            second: Second object to compare.
+            msg: Optional. Custom error message issued when first and second object are not equal.
+            strict_type: Set True to enforce type check before comparison.
+        """
+        default_msg = f"{first} != {second}"
+
+        self.assertTrue(
+            is_equivalent(first, second, strict_type=strict_type),
+            msg=msg or default_msg,
+        )
+
+    def assertRoundTripSerializable(
+        self,
+        obj: Any,
+        *,
+        check_func: Optional[Callable] = None,
+        strict_type: bool = False,
+    ):
         """Assert that an object is round trip serializable.
 
         Args:
             obj: the object to be serialized.
             check_func: Optional, a custom function ``check_func(a, b) -> bool``
-                        to check equality of the original object with the decoded
-                        object. If None the ``__eq__`` method of the original
-                        object will be used.
+                to check equality of the original object with the decoded
+                object. If None :meth:`.assertEqualExtended` is called.
+            strict_type: Set True to enforce type check before comparison.
         """
         try:
             encoded = json.dumps(obj, cls=ExperimentEncoder)
@@ -94,20 +126,27 @@ def assertRoundTripSerializable(self, obj: Any, check_func: Optional[Callable] =
             decoded = json.loads(encoded, cls=ExperimentDecoder)
         except TypeError:
             self.fail("JSON deserialization raised unexpectedly.")
-        if check_func is None:
-            self.assertEqual(obj, decoded)
-        else:
+
+        if check_func is not None:
             self.assertTrue(check_func(obj, decoded), msg=f"{obj} != {decoded}")
+        else:
+            self.assertEqualExtended(obj, decoded, strict_type=strict_type)
 
-    def assertRoundTripPickle(self, obj: Any, check_func: Optional[Callable] = None):
+    def assertRoundTripPickle(
+        self,
+        obj: Any,
+        *,
+        check_func: Optional[Callable] = None,
+        strict_type: bool = False,
+    ):
         """Assert that an object is round trip serializable using pickle module.
 
         Args:
             obj: the object to be serialized.
             check_func: Optional, a custom function ``check_func(a, b) -> bool``
-                        to check equality of the original object with the decoded
-                        object. If None the ``__eq__`` method of the original
-                        object will be used.
+                to check equality of the original object with the decoded
+                object. If None :meth:`.assertEqualExtended` is called.
+            strict_type: Set True to enforce type check before comparison.
         """
         try:
             encoded = pickle.dumps(obj)
@@ -117,150 +156,63 @@ def assertRoundTripPickle(self, obj: Any, check_func: Optional[Callable] = None)
             decoded = pickle.loads(encoded)
         except TypeError:
             self.fail("pickle deserialization raised unexpectedly.")
-        if check_func is None:
-            self.assertEqual(obj, decoded)
-        else:
+
+        if check_func is not None:
             self.assertTrue(check_func(obj, decoded), msg=f"{obj} != {decoded}")
+        else:
+            self.assertEqualExtended(obj, decoded, strict_type=strict_type)
 
     @classmethod
+    @deprecate_func(
+        since="0.6",
+        additional_msg="Use test.extended_equality.is_equivalent instead.",
+        pending=True,
+        package_name="qiskit-experiments",
+    )
     def json_equiv(cls, data1, data2) -> bool:
         """Check if two experiments are equivalent by comparing their configs"""
-        # pylint: disable = too-many-return-statements
-        configurable_type = (BaseExperiment, BaseAnalysis, BaseDrawer)
-        compare_repr = (DataAction, DataProcessor)
-        list_type = (list, tuple, set)
-        skipped = tuple()
-
-        if isinstance(data1, skipped) and isinstance(data2, skipped):
-            warnings.warn(f"Equivalence check for data {data1.__class__.__name__} is skipped.")
-            return True
-        elif isinstance(data1, configurable_type) and isinstance(data2, configurable_type):
-            return cls.json_equiv(data1.config(), data2.config())
-        elif dataclasses.is_dataclass(data1) and dataclasses.is_dataclass(data2):
-            # not using asdict. this copies all objects.
-            return cls.json_equiv(data1.__dict__, data2.__dict__)
-        elif isinstance(data1, dict) and isinstance(data2, dict):
-            if set(data1) != set(data2):
-                return False
-            return all(cls.json_equiv(data1[k], data2[k]) for k in data1.keys())
-        elif isinstance(data1, np.ndarray) or isinstance(data2, np.ndarray):
-            return np.allclose(data1, data2)
-        elif isinstance(data1, list_type) and isinstance(data2, list_type):
-            return all(cls.json_equiv(e1, e2) for e1, e2 in zip(data1, data2))
-        elif isinstance(data1, uncertainties.UFloat) and isinstance(data2, uncertainties.UFloat):
-            return cls.ufloat_equiv(data1, data2)
-        elif isinstance(data1, Model) and isinstance(data2, Model):
-            return cls.json_equiv(data1.dumps(), data2.dumps())
-        elif isinstance(data1, CurveFitResult) and isinstance(data2, CurveFitResult):
-            return cls.curve_fit_data_equiv(data1, data2)
-        elif isinstance(data1, compare_repr) and isinstance(data2, compare_repr):
-            # otherwise compare instance representation
-            return repr(data1) == repr(data2)
-
-        return data1 == data2
+        return is_equivalent(data1, data2)
 
     @staticmethod
+    @deprecate_func(
+        since="0.6",
+        additional_msg="Use test.extended_equality.is_equivalent instead.",
+        pending=True,
+        package_name="qiskit-experiments",
+    )
     def ufloat_equiv(data1: uncertainties.UFloat, data2: uncertainties.UFloat) -> bool:
         """Check if two values with uncertainties are equal. No correlation is considered."""
-        return data1.n == data2.n and data1.s == data2.s
+        return is_equivalent(data1, data2)
 
     @classmethod
+    @deprecate_func(
+        since="0.6",
+        additional_msg="Use test.extended_equality.is_equivalent instead.",
+        pending=True,
+        package_name="qiskit-experiments",
+    )
     def analysis_result_equiv(cls, result1, result2):
         """Test two analysis results are equivalent"""
-        # Check basic attributes skipping service which is not serializable
-        for att in [
-            "name",
-            "value",
-            "extra",
-            "device_components",
-            "result_id",
-            "experiment_id",
-            "chisq",
-            "quality",
-            "verified",
-            "tags",
-            "auto_save",
-            "source",
-        ]:
-            if not cls.json_equiv(getattr(result1, att), getattr(result2, att)):
-                return False
-        return True
+        return is_equivalent(result1, result2)
 
     @classmethod
+    @deprecate_func(
+        since="0.6",
+        additional_msg="Use test.extended_equality.is_equivalent instead.",
+        pending=True,
+        package_name="qiskit-experiments",
+    )
     def curve_fit_data_equiv(cls, data1, data2):
         """Test two curve fit result are equivalent."""
-        for att in [
-            "method",
-            "model_repr",
-            "success",
-            "nfev",
-            "message",
-            "dof",
-            "init_params",
-            "chisq",
-            "reduced_chisq",
-            "aic",
-            "bic",
-            "params",
-            "var_names",
-            "x_data",
-            "y_data",
-            "covar",
-        ]:
-            if not cls.json_equiv(getattr(data1, att), getattr(data2, att)):
-                return False
-        return True
+        return is_equivalent(data1, data2)
 
     @classmethod
+    @deprecate_func(
+        since="0.6",
+        additional_msg="Use test.extended_equality.is_equivalent instead.",
+        pending=True,
+        package_name="qiskit-experiments",
+    )
     def experiment_data_equiv(cls, data1, data2):
         """Check two experiment data containers are equivalent"""
-
-        # Check basic attributes
-        # Skip non-compatible backend
-        for att in [
-            "experiment_id",
-            "experiment_type",
-            "parent_id",
-            "tags",
-            "job_ids",
-            "figure_names",
-            "share_level",
-            "metadata",
-        ]:
-            if not cls.json_equiv(getattr(data1, att), getattr(data2, att)):
-                return False
-
-        # Check length of data, results, child_data
-        # check for child data attribute so this method still works for
-        # DbExperimentData
-        if hasattr(data1, "child_data"):
-            child_data1 = data1.child_data()
-        else:
-            child_data1 = []
-        if hasattr(data2, "child_data"):
-            child_data2 = data2.child_data()
-        else:
-            child_data2 = []
-
-        if (
-            len(data1.data()) != len(data2.data())
-            or len(data1.analysis_results()) != len(data2.analysis_results())
-            or len(child_data1) != len(child_data2)
-        ):
-            return False
-
-        # Check data
-        if not cls.json_equiv(data1.data(), data2.data()):
-            return False
-
-        # Check analysis results
-        for result1, result2 in zip(data1.analysis_results(), data2.analysis_results()):
-            if not cls.analysis_result_equiv(result1, result2):
-                return False
-
-        # Check child data
-        for child1, child2 in zip(child_data1, child_data2):
-            if not cls.experiment_data_equiv(child1, child2):
-                return False
-
-        return True
+        return is_equivalent(data1, data2)

test/calibration/test_calibrations.py

@@ -1754,7 +1754,7 @@ def test_serialization(self):
         cals = Calibrations.from_backend(backend, libraries=[library])
         cals.add_parameter_value(0.12345, "amp", 3, "x")
 
-        self.assertRoundTripSerializable(cals, self.json_equiv)
+        self.assertRoundTripSerializable(cals)
 
     def test_equality(self):
         """Test the equal method on calibrations."""

test/curve_analysis/test_baseclass.py

@@ -85,7 +85,7 @@ class TestCurveAnalysis(CurveAnalysisTestCase):
     def test_roundtrip_serialize(self):
         """A testcase for serializing analysis instance."""
         analysis = CurveAnalysis(models=[ExpressionModel(expr="par0 * x + par1", name="test")])
-        self.assertRoundTripSerializable(analysis, check_func=self.json_equiv)
+        self.assertRoundTripSerializable(analysis)
 
     def test_parameters(self):
         """A testcase for getting fit parameters with attribute."""

test/data_processing/test_data_processing.py

@@ -387,14 +387,14 @@ def test_json_single_node(self):
         """Check if the data processor is serializable."""
         node = MinMaxNormalize()
         processor = DataProcessor("counts", [node])
-        self.assertRoundTripSerializable(processor, check_func=self.json_equiv)
+        self.assertRoundTripSerializable(processor)
 
     def test_json_multi_node(self):
         """Check if the data processor with multiple nodes is serializable."""
         node1 = MinMaxNormalize()
         node2 = AverageData(axis=2)
         processor = DataProcessor("counts", [node1, node2])
-        self.assertRoundTripSerializable(processor, check_func=self.json_equiv)
+        self.assertRoundTripSerializable(processor)
 
     def test_json_trained(self):
         """Check if trained data processor is serializable and still work."""
@@ -405,7 +405,7 @@ def test_json_trained(self):
             main_axes=np.array([[1, 0]]), scales=[1.0], i_means=[0.0], q_means=[0.0]
         )
         processor = DataProcessor("memory", data_actions=[node])
-        self.assertRoundTripSerializable(processor, check_func=self.json_equiv)
+        self.assertRoundTripSerializable(processor)
 
         serialized = json.dumps(processor, cls=ExperimentEncoder)
         loaded_processor = json.loads(serialized, cls=ExperimentDecoder)