fix extractor/profileisolator bug and update test case

Signed-off-by: Sunyanan Choochotkaew <[email protected]>

Author: sunya-ch

src/train/extractor/extractor.py

@@ -147,7 +147,7 @@ def get_workload_feature_data(self, query_results, features):
 
             if all(col in aggr_query_data.columns for col in container_id_cols):
                 aggr_query_data.rename(columns={query: feature}, inplace=True)
-                aggr_query_data[container_id_colname] = aggr_query_data[container_id_cols].apply(lambda x: '/'.join(x), axis=1)
+                aggr_query_data[container_id_colname] = aggr_query_data[container_id_cols].apply(lambda x: '/'.join([str(xi) for xi in x]), axis=1)
                 # separate for each container_id
                 container_id_list = pd.unique(aggr_query_data[container_id_colname])
 
@@ -230,17 +230,20 @@ def get_power_data(self, query_results, energy_components, source):
                 if usage_ratio_query not in query_results:
                     # sum over mode (idle, dynamic) and unit col
                     df = aggr_query_data.groupby([TIMESTAMP_COL]).sum().reset_index().set_index(TIMESTAMP_COL)
+                    time_diff_values = df.reset_index()[[TIMESTAMP_COL]].diff().dropna().values.mean()
                     df = df.loc[:, df.columns != unit_col]
                     # rename
                     colname = component_to_col(component)
                     df.rename(columns={query: colname}, inplace=True)
                     # find current value from aggregated query
                     df = df.sort_index()[colname].diff().dropna()
+                    df /= time_diff_values
                     df = df.mask(df.lt(0)).ffill().fillna(0).convert_dtypes()
                     power_data_list += [df]
                 else:
                      # sum over mode (idle, dynamic)
                     aggr_query_data = aggr_query_data.groupby([unit_col, TIMESTAMP_COL]).sum().reset_index().set_index(TIMESTAMP_COL)
+                    time_diff_values = aggr_query_data.reset_index()[[TIMESTAMP_COL]].diff().dropna().values.mean()
                     # add per unit_col
                     unit_vals = pd.unique(aggr_query_data[unit_col])
                     for unit_val in unit_vals:
@@ -250,16 +253,19 @@ def get_power_data(self, query_results, energy_components, source):
                         df.rename(columns={query: colname}, inplace=True)
                         # find current value from aggregated query
                         df = df.sort_index()[colname].diff().dropna()
+                        df /= time_diff_values
                         df = df.mask(df.lt(0)).ffill().fillna(0).convert_dtypes()
                         power_data_list += [df]
             else:
                 # sum over mode
                 aggr_query_data = aggr_query_data.groupby([TIMESTAMP_COL]).sum()
+                time_diff_values = aggr_query_data.reset_index()[[TIMESTAMP_COL]].diff().dropna().values.mean()
                 # rename
                 colname = component_to_col(component)
                 aggr_query_data.rename(columns={query: colname}, inplace=True)
                 # find current value from aggregated query
                 df = aggr_query_data.sort_index()[colname].diff().dropna()
+                df /= time_diff_values
                 df = df.mask(df.lt(0)).ffill().fillna(0).convert_dtypes()
                 power_data_list += [df]
         power_data = pd.concat(power_data_list, axis=1).dropna()

src/train/isolator/isolator.py

@@ -128,7 +128,8 @@ def isolate(self, data, label_cols, energy_source):
             data = data.reset_index()
         data = data.set_index([TIMESTAMP_COL])
         target_data, _ = isolate_container(data, self.background_containers, label_cols)
-        isolated_data = target_data.copy()
+        isolated_data = squeeze_data(target_data, label_cols)
+        print(isolated_data)
         try:
             for label_col in label_cols:
                 component = col_to_component(label_col)
@@ -137,7 +138,6 @@ def isolate(self, data, label_cols, energy_source):
                     return None
                 isolated_data[label_col] = isolated_data[label_col] - isolated_data['profile']
                 isolated_data.drop(columns='profile', inplace=True)
-            isolated_data = isolated_data.reset_index()
             if index_list[0] is not None:
                 isolated_data = isolated_data.set_index(index_list)
             return isolated_data

tests/data/prom_output/.DS_Store

@@ -62,18 +62,22 @@ def save_extract_results(instance, feature_group, extracted_data, node_level, sa
     save_csv(save_path, filename, extracted_data)
 
 def get_expected_power_columns(energy_components=test_energy_components, num_of_unit=test_num_of_unit):
-    return [component_to_col(component, "package", unit_val) for component in energy_components for unit_val in range(0,num_of_unit)]
+    # TODO: if ratio applied, 
+    # return [component_to_col(component, "package", unit_val) for component in energy_components for unit_val in range(0,num_of_unit)]
+    return [component_to_col(component) for component in energy_components]
 
 def assert_extract(extracted_data, power_columns, energy_components, num_of_unit, feature_group):
     extracted_data_column_names = extracted_data.columns
     # basic assert
     assert extracted_data is not None, "extracted data is None"
     assert len(power_columns) > 0, "no power label column {}".format(extracted_data_column_names)
     assert node_info_column in extracted_data_column_names, "no {} in column {}".format(node_info_column, extracted_data_column_names)
-    expected_power_column_length = len(energy_components) * num_of_unit
+    # TODO: if ratio applied, expected_power_column_length = len(energy_components) * num_of_unit
+    expected_power_column_length = len(energy_components) 
     # detail assert
     assert len(power_columns) == expected_power_column_length, "unexpected power label columns {}, expected {}".format(power_columns, expected_power_column_length)
-    expected_col_size = expected_power_column_length + len(FeatureGroups[FeatureGroup[feature_group]]) + 1 + num_of_unit # power ratio
+    # TODO: if ratio applied, expected_col_size must + 1 for power_ratio
+    expected_col_size = expected_power_column_length + len(FeatureGroups[FeatureGroup[feature_group]]) + num_of_unit # power ratio
     assert len(extracted_data_column_names) == expected_col_size, "unexpected column length: expected {}, got {}({}) ".format(expected_col_size, extracted_data_column_names, len(extracted_data_column_names))
 
 def process(query_results, feature_group, save_path=extractor_output_path, customize_extractors=test_customize_extractors, energy_source=test_energy_source, num_of_unit=2):

tests/data/prom_output/idle.json

@@ -52,7 +52,7 @@ def save_results(instance, extractor_name, feature_group, isolated_data, save_pa
 def assert_isolate(extractor_result, isolated_data):
     isolated_data_column_names = isolated_data.columns
     assert isolated_data is not None, "isolated data is None"
-    value_df = isolated_data.drop(columns=container_level_index)
+    value_df = isolated_data.reset_index().drop(columns=container_level_index)
     negative_df = value_df[(value_df<0).all(1)]
     assert len(negative_df) == 0, "all data must be non-negative \n {}".format(negative_df) 
     assert len(extractor_result.columns) == len(isolated_data_column_names), "unexpected column length: expected {}, got {}({}) ".format(len(extractor_result.columns), isolated_data_column_names, len(isolated_data_column_names))
@@ -90,10 +90,9 @@ def process(test_isolators=test_isolators, customize_isolators=[], extract_path=
                 save_results(test_instance, extractor_name, feature_group, isolated_data, save_path=save_path)
 
 
-
 if __name__ == '__main__':
     # Add customize isolator here
-    customize_isolators = [TrainIsolator(idle_data=test_idle_data, profiler=DefaultProfiler)]
-    customize_isolators = [TrainIsolator(target_hints=["coremark"])]
-    customize_isolators += [ProfileBackgroundIsolator(test_profiles, test_idle_data)]
+    customize_isolators = [ProfileBackgroundIsolator(test_profiles, test_idle_data)]
+    customize_isolators += [TrainIsolator(idle_data=test_idle_data, profiler=DefaultProfiler)]
+    customize_isolators += [TrainIsolator(target_hints=["coremark"])]
     process(customize_isolators=customize_isolators)