Benchmarks: Code Revision - change 'reduce' to 'reduce_op' (#156)

**Description**
Change the field name `reduce` to `reduce_op`.

docs/design-docs/benchmarks.md

@@ -44,7 +44,7 @@ This chapter will describe the design details of all the components in `benchmar
 
 The E2E model benchmarks have 4-layer inheritance relationship.
 
-#### Training 
+#### Training
 
 The general process of model training is:
 
@@ -66,7 +66,7 @@ Compared with training, it just gets rid of create_optimizer operation.
 
 ### Micro Benchmarks
 
-The micro-benchmarks have 3-layer Inheritance Relationship. There are two base classes for micro-benchmark: 
+The micro-benchmarks have 3-layer Inheritance Relationship. There are two base classes for micro-benchmark:
 `MicroBenchmark` is pure-python benchmark.
 `MicroBenchmarkWithInvoke` is benchmark depending on third-party executable program.
 
@@ -219,7 +219,7 @@ result = {
              ...
         'metricsM': List[Number],
     },
-    'reduce': {
+    'reduce_op': {
         'metrics1': ReduceType,
              ...
         'metricsM': ReduceType,
@@ -249,7 +249,7 @@ result = {
             'throughput-inference-float32': [avg_throughput1, ..., avg_throughputN],
             'throughput-inference-float16': [avg_throughput1, ..., avg_throughputN],
     },
-    'reduce': {
+    'reduce_op': {
         'throughput-train-float32': 'min',
         'throughput-train-float16': 'min',
         'throughput-inference-float32': None,
@@ -272,7 +272,7 @@ result = {
     'result': { # Key is metrics
         'overhead': [overhead1, ..., overheadN],
     },
-    'reduce': {
+    'reduce_op': {
         'overhead': None,
     },
 }

superbench/benchmarks/result.py

@@ -31,7 +31,7 @@ class BenchmarkResult():
         self.__end_time = None
         self.__raw_data = dict()
         self.__result = dict()
-        self.__reduce = dict()
+        self.__reduce_op = dict()
 
     def __eq__(self, rhs):
         """Override equal function for deep comparison.
@@ -89,7 +89,7 @@ class BenchmarkResult():
 
         if metric not in self.__result:
             self.__result[metric] = list()
-            self.__reduce[metric] = reduce_type.value if isinstance(reduce_type, Enum) else None
+            self.__reduce_op[metric] = reduce_type.value if isinstance(reduce_type, Enum) else None
         self.__result[metric].append(value)
 
         return True
@@ -177,3 +177,8 @@ class BenchmarkResult():
     def result(self):
         """Decoration function to access __result."""
         return self.__result
+
+    @property
+    def reduce_op(self):
+        """Decoration function to access __reduce_op."""
+        return self.__reduce_op

tests/benchmarks/model_benchmarks/test_model_base.py

@@ -220,7 +220,7 @@ def test_train():
         '"steptime_train_float32": [[2.0, 2.0, 2.0, 2.0, 2.0, 2.0, 2.0, 2.0]], '
         '"throughput_train_float32": [[16000.0, 16000.0, 16000.0, 16000.0, 16000.0, 16000.0, 16000.0, 16000.0]]}, '
         '"result": {"steptime_train_float32": [2.0], "throughput_train_float32": [16000.0]}, '
-        '"reduce": {"steptime_train_float32": "max", "throughput_train_float32": "min"}}'
+        '"reduce_op": {"steptime_train_float32": "max", "throughput_train_float32": "min"}}'
     )
     assert (benchmark._preprocess())
     assert (benchmark._ModelBenchmark__train(Precision.FLOAT32))
@@ -230,7 +230,7 @@ def test_train():
     benchmark = create_benchmark('--num_steps 0')
     expected_result = (
         '{"name": "pytorch-fake-model", "type": "model", "run_count": 1, "return_code": 3, '
-        '"start_time": null, "end_time": null, "raw_data": {}, "result": {}, "reduce": {}}'
+        '"start_time": null, "end_time": null, "raw_data": {}, "result": {}, "reduce_op": {}}'
     )
     assert (benchmark._preprocess())
     assert (benchmark._ModelBenchmark__train(Precision.FLOAT32) is False)
@@ -246,7 +246,7 @@ def test_inference():
         '"steptime_inference_float16": [[4.0, 4.0, 4.0, 4.0, 4.0, 4.0, 4.0, 4.0]], '
         '"throughput_inference_float16": [[8000.0, 8000.0, 8000.0, 8000.0, 8000.0, 8000.0, 8000.0, 8000.0]]}, '
         '"result": {"steptime_inference_float16": [4.0], "throughput_inference_float16": [8000.0]}, '
-        '"reduce": {"steptime_inference_float16": null, "throughput_inference_float16": null}}'
+        '"reduce_op": {"steptime_inference_float16": null, "throughput_inference_float16": null}}'
     )
     assert (benchmark._preprocess())
     assert (benchmark._ModelBenchmark__inference(Precision.FLOAT16))
@@ -256,7 +256,7 @@ def test_inference():
     benchmark = create_benchmark('--num_steps 0')
     expected_result = (
         '{"name": "pytorch-fake-model", "type": "model", "run_count": 1, "return_code": 3, '
-        '"start_time": null, "end_time": null, "raw_data": {}, "result": {}, "reduce": {}}'
+        '"start_time": null, "end_time": null, "raw_data": {}, "result": {}, "reduce_op": {}}'
     )
     assert (benchmark._preprocess())
     assert (benchmark._ModelBenchmark__inference(Precision.FLOAT16) is False)
@@ -296,7 +296,7 @@ def test_benchmark():
         '"throughput_train_float16": [[16000.0, 16000.0, 16000.0, 16000.0, 16000.0, 16000.0, 16000.0, 16000.0]]}, '
         '"result": {"steptime_train_float32": [2.0], "throughput_train_float32": [16000.0], '
         '"steptime_train_float16": [2.0], "throughput_train_float16": [16000.0]}, '
-        '"reduce": {"steptime_train_float32": "max", "throughput_train_float32": "min", '
+        '"reduce_op": {"steptime_train_float32": "max", "throughput_train_float32": "min", '
         '"steptime_train_float16": "max", "throughput_train_float16": "min"}}'
     )
     assert (benchmark.serialized_result == expected_serialized_result)

tests/benchmarks/test_registry.py

@@ -149,7 +149,7 @@ def test_launch_benchmark():
         '"return_code": 0, "start_time": null, "end_time": null, '
         '"raw_data": {"accumulation_result": ["1,3,6,10"]}, '
         '"result": {"accumulation_result": [10]}, '
-        '"reduce": {"accumulation_result": null}}'
+        '"reduce_op": {"accumulation_result": null}}'
     )
     assert (result == expected)
 
@@ -173,7 +173,7 @@ def test_launch_benchmark():
         '"return_code": 0, "start_time": null, "end_time": null, '
         '"raw_data": {"accumulation_result": ["1,3,6"]}, '
         '"result": {"accumulation_result": [6]}, '
-        '"reduce": {"accumulation_result": null}}'
+        '"reduce_op": {"accumulation_result": null}}'
     )
     assert (result == expected)
 

tests/benchmarks/test_result.py

@@ -83,6 +83,6 @@ def test_serialize_deserialize():
         '"start_time": "2021-02-03 16:59:49", "end_time": "2021-02-03 17:00:08", '
         '"raw_data": {"metric1": [[1, 2, 3], [4, 5, 6], [7, 8, 9]]}, '
         '"result": {"metric1": [300, 200], "metric2": [100]}, '
-        '"reduce": {"metric1": "max", "metric2": "avg"}}'
+        '"reduce_op": {"metric1": "max", "metric2": "avg"}}'
     )
     assert (result.to_string() == expected)