v2

Author: jingyuanlm

rdagent/scenarios/data_science/proposal/exp_gen/prompts_v2.yaml

@@ -365,12 +365,14 @@ hypothesis_rewrite:
 
       - Runtime Assessment
         The current maximum runtime of the code is {time_max} hours.
-         Please assess whether this is a time-intensive competition—specifically, if the runtime of a single hypothesis falls within 50% to 90% of the total allowed time {full_time} hours.
-         In such cases, you must remove any hypotheses that are likely to exceed the time limit, especially those involving ensemble models or an excessive number of cross-validation folds.
-         If the competition is not time-intensive, ensure that the expected runtime of each retained hypothesis does not exceed 1.5 × {time max} hours, unless you have strong confidence that the hypothesis—despite its high computational cost—will lead to significantly better performance or results.
+        Please assess whether this is a time-intensive competition—specifically, if the runtime of a single hypothesis falls within 50% to 100% of the total allowed time {full_time} hours. In such cases, you must remove any hypotheses that are likely to exceed the time limit, especially those involving ensemble models or an excessive number of cross-validation folds.
 
-     - Retention Condition
-       If the above conditions are met, you do not need to delete any hypotheses.
+      - Retention Condition 
+        If the competition is not time-intensive, ensure that the expected runtime of each retained hypothesis does not exceed 1.5 × {time_max} hours, unless you have strong confidence that the hypothesis—despite its high computational cost—will lead to significantly better performance or results. You can consider keeping this hypotheses.
+      
+      - Critical Challenge Addressing
+        If a hypothesis addresses a critical challenge that has caused prior experiment failures, it should be removed.
+  
 
     {% if rewrite_output_format is not none %}
     ## Output Format

rdagent/scenarios/data_science/proposal/exp_gen/proposal.py

@@ -731,6 +731,7 @@ def hypothesis_rewrite(
         scenario_desc: str,
         sota_exp_desc: str,
         exp_feedback_list_desc: str,
+        trace: DSTrace,
     ) -> Dict:
         """
         Generate improved hypotheses based on critique feedback for each original hypothesis.
@@ -746,10 +747,11 @@ def hypothesis_rewrite(
             hypothesis_critique_pairs += f"**Component:** {hypothesis_data.get('component', 'Unknown')}\n"
             hypothesis_critique_pairs += f"**Reasoning:** {hypothesis_data.get('reason', 'Not provided')}\n"
             hypothesis_critique_pairs += f"**Critique:** {critique_data.get('critique', 'No critique available')}\n\n"
-
+        
+        time_list = [tr[0].running_info.running_time for tr in trace.hist[:-1]]
         time_status = None
-        time_max = 2*3600#trace.hist.runing_info.get("time_max", None)
-        full_time = DS_RD_SETTING.full_timeout
+        time_max = max(time_list)/ 3600 #2*3600
+        full_time = DS_RD_SETTING.full_timeout / 3600
 
         if DS_RD_SETTING.enable_scale_check and RD_Agent_TIMER_wrapper.timer.started:
             remain_time = RD_Agent_TIMER_wrapper.timer.remain_time()
@@ -1141,6 +1143,7 @@ def gen(
                     scenario_desc=scenario_desc,
                     sota_exp_desc=sota_exp_desc,
                     exp_feedback_list_desc=exp_feedback_list_desc,
+                    trace = trace
                 )
                 logger.info(f"Successfully completed hypothesis critique and rewrite process")
             except Exception as e: