Revise For Better COSTEER & Tuning

rdagent/components/coder/model_coder/CoSTEER/__init__.py

@@ -70,7 +70,7 @@ def develop(self, exp: ModelExperiment) -> ModelExperiment:
         self.rag = ModelRAGStrategy(model_knowledge_base)
 
         # init intermediate items
-        model_experiment = ModelEvolvingItem(sub_tasks=exp.sub_tasks)
+        model_experiment = ModelEvolvingItem(sub_tasks=exp.sub_tasks, from_based_exp=exp.based_experiments)
 
         self.evolve_agent = ModelRAGEvoAgent(
             max_loop=self.max_loop,

rdagent/components/coder/model_coder/CoSTEER/evolvable_subjects.py

@@ -5,6 +5,8 @@
 )
 from rdagent.core.evolving_framework import EvolvableSubjects
 from rdagent.log import rdagent_logger as logger
+from collections.abc import Sequence
+from rdagent.core.experiment import ASpecificWSForExperiment
 
 
 class ModelEvolvingItem(ModelExperiment, EvolvableSubjects):
@@ -15,6 +17,7 @@ class ModelEvolvingItem(ModelExperiment, EvolvableSubjects):
     def __init__(
         self,
         sub_tasks: list[ModelTask],
+        from_based_exp: Sequence[ASpecificWSForExperiment] = [],
         sub_gt_implementations: list[ModelFBWorkspace] = None,
     ):
         ModelExperiment.__init__(self, sub_tasks=sub_tasks)
@@ -27,3 +30,4 @@ def __init__(
             )
         else:
             self.sub_gt_implementations = sub_gt_implementations
+        self.based_exp = from_based_exp

rdagent/components/coder/model_coder/CoSTEER/evolving_strategy.py

@@ -22,6 +22,9 @@
 from rdagent.core.utils import multiprocessing_wrapper
 from rdagent.oai.llm_utils import APIBackend
 
+from collections.abc import Sequence
+from rdagent.core.experiment import ASpecificWSForExperiment
+
 coder_prompts = Prompts(file_path=Path(__file__).parent.parent / "prompts.yaml")
 
 
@@ -30,27 +33,28 @@ def implement_one_model(
         self,
         target_task: ModelTask,
         queried_knowledge: ModelQueriedKnowledge = None,
-        exp: ModelExperiment = None,  # Add this parameter
+        based_exp: Sequence[ASpecificWSForExperiment] = [],
     ) -> str:
         model_information_str = target_task.get_task_information()
         model_type = target_task.model_type
 
-        data_desc = None
-        # Get the current code from the experiment using build_from_SOTA
         current_code = ""
-        if exp is not None:
-            self.build_from_SOTA(exp)
-            model_file_mapping = {
-                "XGBoost": "model_xgb.py",
-                "RandomForest": "model_rf.py",
-                "LightGBM": "model_lgb.py",
-                "NN": "model_nn.py",
-            }
+        data_desc = None
+
+        model_file_mapping = {
+            "XGBoost": "model_xgb.py",
+            "RandomForest": "model_rf.py",
+            "LightGBM": "model_lgb.py",
+            "NN": "model_nn.py",
+        }
+
+        for exp in based_exp:
             if model_type in model_file_mapping:
                 current_code = exp.experiment_workspace.code_dict.get(model_file_mapping[model_type], "")
-                # send in data description  
             data_desc = exp.experiment_workspace.data_description
 
+            if current_code:
+                break  # Use the first non-empty code found
 
         if queried_knowledge is not None and model_information_str in queried_knowledge.success_task_to_knowledge_dict:
             return queried_knowledge.success_task_to_knowledge_dict[model_information_str].implementation
@@ -127,7 +131,6 @@ def evolve(
         *,
         evo: ModelEvolvingItem,
         queried_knowledge: ModelQueriedKnowledge | None = None,
-        exp: ModelExperiment = None,  # Add this parameter
         **kwargs,
     ) -> ModelEvolvingItem:
         # 1. Find the models that need to be evolved
@@ -146,7 +149,7 @@ def evolve(
 
         result = multiprocessing_wrapper(
             [
-                (self.implement_one_model, (evo.sub_tasks[target_index], queried_knowledge, exp))  # Pass exp here
+                (self.implement_one_model, (evo.sub_tasks[target_index], queried_knowledge, evo.based_experiments))  # Pass exp here
                 for target_index in to_be_finished_task_index
             ],
             n=RD_AGENT_SETTINGS.multi_proc_n,

rdagent/scenarios/kaggle/prompts.yaml

@@ -39,7 +39,7 @@ hypothesis_and_feedback: |-
 hypothesis_output_format: |-
   The output should follow JSON format. The schema is as follows:
   {
-    "action": "If "hypothesis_specification" provides the action you need to take, please follow "hypothesis_specification" to choose the action. Otherwise, based on previous experimental results, suggest the action you believe is most appropriate at the moment. It should be one of [{% if if_using_feature_selection %}"Feature engineering", "Feature processing", "Model feature selection", "Model tuning"{% else %}"Feature engineering", "Feature processing", "Model tuning"{% endif %}]",
+    "action": "Choose Model tuning and ignore the rest". If "hypothesis_specification" provides the action you need to take, please follow "hypothesis_specification" to choose the action. Otherwise, based on previous experimental results, suggest the action you believe is most appropriate at the moment. It should be one of [{% if if_using_feature_selection %}"Feature engineering", "Feature processing", "Model feature selection", "Model tuning"{% else %}"Feature engineering", "Feature processing", "Model tuning"{% endif %}]",
     "hypothesis": "The new hypothesis generated based on the information provided.",
     "reason": "The reason why you generate this hypothesis. It should be comprehensive and logical. It should cover the other keys below and extend them.",
     "concise_reason": "Two-line summary. First line focuses on a concise justification for the change. Second line generalizes a knowledge statement.",