1010
1111from kepler_model .util .train_types import get_valid_feature_groups , ModelOutputType , FeatureGroups , FeatureGroup , PowerSourceMap , weight_support_trainers
1212from kepler_model .util .config import getConfig , model_toppath , ERROR_KEY , MODEL_SERVER_MODEL_REQ_PATH , MODEL_SERVER_MODEL_LIST_PATH , initial_pipeline_urls , download_path
13- from kepler_model .util .loader import parse_filters , is_valid_model , load_json , load_weight , get_model_group_path , get_archived_file , METADATA_FILENAME , CHECKPOINT_FOLDERNAME , get_pipeline_path , any_node_type , is_matched_type , get_largest_candidates , default_pipelines , get_node_type_from_name
13+ from kepler_model .util .loader import parse_filters , is_valid_model , load_json , load_weight , get_model_group_path , get_archived_file , METADATA_FILENAME , CHECKPOINT_FOLDERNAME , get_pipeline_path , any_node_type , is_matched_type , get_largest_candidates
1414from kepler_model .util .saver import WEIGHT_FILENAME
1515from kepler_model .train import NodeTypeSpec , NodeTypeIndexCollection
1616
@@ -65,26 +65,10 @@ def __init__(self, metrics, output_type, source="rapl-sysfs", node_type=-1, weig
6565"""
6666
6767
68- def select_best_model (spec , valid_grouppath , filters , energy_source , pipeline_name = "" , trainer_name = "" , node_type = any_node_type , weight = False , loose_node_type = True ):
69- # Find initial model list filtered by trainer
70- initial_model_names = [f for f in os .listdir (valid_grouppath ) if f != CHECKPOINT_FOLDERNAME and not os .path .isfile (os .path .join (valid_grouppath , f )) and (trainer_name == "" or trainer_name in f )]
71- if pipeline_name == "" and energy_source in default_pipelines :
72- pipeline_name = default_pipelines [energy_source ]
73-
74- if node_type != any_node_type :
75- model_names = [name for name in initial_model_names if "_{}" .format (node_type ) in name ]
76- if len (model_names ) == 0 :
77- if not loose_node_type :
78- return None , None
79- logger .warning (f"{ valid_grouppath } { node_type } )
80- model_names = initial_model_names
81- else :
82- model_names = initial_model_names
83-
84- # Filter weight models
68+ def select_best_model (spec , valid_groupath , filters , energy_source , pipeline_name = "" , trainer_name = "" , node_type = any_node_type , weight = False ):
69+ model_names = [f for f in os .listdir (valid_groupath ) if f != CHECKPOINT_FOLDERNAME and not os .path .isfile (os .path .join (valid_groupath , f )) and (trainer_name == "" or trainer_name in f )]
8570 if weight :
86- candidates = [name for name in model_names if name .split ("_" )[0 ] in weight_support_trainers ]
87-
71+ model_names = [name for name in model_names if name .split ("_" )[0 ] in weight_support_trainers ]
8872 # Load metadata of trainers
8973 best_cadidate = None
9074 best_response = None
@@ -101,7 +85,7 @@ def select_best_model(spec, valid_grouppath, filters, energy_source, pipeline_na
10185 logger .warn ("no large candidates, select from all availables" )
10286 candidates = model_names
10387 for model_name in candidates :
104- model_savepath = os .path .join (valid_grouppath , model_name )
88+ model_savepath = os .path .join (valid_groupath , model_name )
10589 metadata = load_json (model_savepath , METADATA_FILENAME )
10690 if metadata is None or not is_valid_model (metadata , filters ) or ERROR_KEY not in metadata :
10791 # invalid metadata
@@ -114,7 +98,7 @@ def select_best_model(spec, valid_grouppath, filters, energy_source, pipeline_na
11498 logger .warn ("weight failed: %s" , model_savepath )
11599 continue
116100 else :
117- response = get_archived_file (valid_grouppath , model_name )
101+ response = get_archived_file (valid_groupath , model_name )
118102 if not os .path .exists (response ):
119103 # archived model file does not exists
120104 logger .warn ("archive failed: %s" , response )
@@ -146,36 +130,20 @@ def get_model():
146130 output_type = ModelOutputType [req .output_type ]
147131 best_model = None
148132 best_response = None
149- best_uncertainty = None
150- best_looseness = None
151133 # find best model comparing best candidate from each valid feature group complied with filtering conditions
152134 for fg in valid_fgs :
153- pipeline_name = pipelineName [energy_source ]
154- valid_groupath = get_model_group_path (model_toppath , output_type , fg , energy_source , pipeline_name = pipeline_name )
155- node_type = req .node_type
156- if req .node_type == any_node_type and req .spec is not None and not req .spec .is_none () and pipeline_name in nodeCollection :
157- node_type , uncertainty , looseness = nodeCollection [pipeline_name ].get_node_type (req .spec , loose_search = True )
158- else :
159- uncertainty = 0
160- looseness = 0
135+ valid_groupath = get_model_group_path (model_toppath , output_type , fg , energy_source , pipeline_name = pipelineName [energy_source ])
161136 if os .path .exists (valid_groupath ):
162- best_candidate , response = select_best_model (req .spec , valid_groupath , filters , energy_source , req .pipeline_name , req .trainer_name , node_type , req .weight )
137+ best_candidate , response = select_best_model (req .spec , valid_groupath , filters , energy_source , req .pipeline_name , req .trainer_name , req . node_type , req .weight )
163138 if best_candidate is None :
164139 continue
165- if node_type != any_node_type and best_model is not None and get_node_type_from_name (best_model ['model_name' ]) == node_type :
166- if get_node_type_from_name (best_candidate ['model_name' ]) != node_type :
167- continue
168140 if best_model is None or best_model [ERROR_KEY ] > best_candidate [ERROR_KEY ]:
169141 best_model = best_candidate
170142 best_response = response
171- best_uncertainty = uncertainty
172- best_looseness = looseness
173- logger .info (f"response: model { best_model ['model_name' ]} { best_model ['features' ]} { ERROR_KEY } { best_model [ERROR_KEY ]} { best_uncertainty } { best_looseness } )
174143 if best_model is None :
175144 return make_response ("cannot find model for {} at the moment" .format (model_request ), 400 )
176145 if req .weight :
177146 try :
178- best_response ["model_name" ] = best_model ['model_name' ]
179147 response = app .response_class (response = json .dumps (best_response ), status = 200 , mimetype = "application/json" )
180148 return response
181149 except ValueError as err :
@@ -186,13 +154,13 @@ def get_model():
186154 except ValueError as err :
187155 return make_response ("send archived model error: {}" .format (err ), 400 )
188156
157+
189158# get_available_models: return name list of best-candidate pipelines
190159@app .route (MODEL_SERVER_MODEL_LIST_PATH , methods = ["GET" ])
191160def get_available_models ():
192161 fg = request .args .get ("fg" )
193162 ot = request .args .get ("ot" )
194163 energy_source = request .args .get ("source" )
195- node_type = request .args .get ("type" )
196164 filter = request .args .get ("filter" )
197165
198166 try :
@@ -213,27 +181,21 @@ def get_available_models():
213181 filters = dict ()
214182 else :
215183 filters = parse_filters (filter )
216- if node_type is None :
217- node_type = - 1
218- else :
219- node_type = int (node_type )
220184
221185 model_names = dict ()
222186 for output_type in output_types :
223- logger .debug (f"Searching output type { output_type } )
224187 model_names [output_type .name ] = dict ()
225188 for fg in valid_fgs :
226- logger .debug (f"Searching feature group { fg } )
227189 valid_groupath = get_model_group_path (model_toppath , output_type , fg , energy_source , pipeline_name = pipelineName [energy_source ])
228190 if os .path .exists (valid_groupath ):
229- best_candidate , _ = select_best_model (None , valid_groupath , filters , energy_source , node_type = node_type , loose_node_type = False )
191+ best_candidate , _ = select_best_model (None , valid_groupath , filters , energy_source )
230192 if best_candidate is None :
231193 continue
232194 model_names [output_type .name ][fg .name ] = best_candidate ["model_name" ]
233195 response = app .response_class (response = json .dumps (model_names ), status = 200 , mimetype = "application/json" )
234196 return response
235197 except (ValueError , Exception ) as err :
236- return make_response (f "failed to get best model list: { err } " 400 )
198+ return make_response ("failed to get best model list: {}" . format ( err ) , 400 )
237199
238200
239201# upack_zip_files: unpack all model.zip files to model folder and copy model.json to model/weight.zip
0 commit comments