fix: adjust ds_trace lookup and add stderr redirect to mlebench command (microsoft#853)

* fix: adjust ds_trace lookup and add stderr redirect to mlebench command

* style: reformat SOTA experiment lookup in ds_trace.py

* feat: add DS_RD_SETTING pipeline to MergeExpGen success message

Author: you-n-g

rdagent/log/ui/ds_trace.py

@@ -397,7 +397,11 @@ def main_win(loop_id, llm_data=None):
             loop_data["running"],
             loop_data.get("mle_score", "no submission to score"),
             llm_data=llm_data["running"] if llm_data else None,
-            sota_exp=state.data[loop_id - 1].get("record", {}).get("SOTA experiment", None) if loop_id >= 1 else None,
+            sota_exp=(
+                state.data[loop_id - 1].get("record", {}).get("SOTA experiment", None)
+                if (loop_id - 1) in state.data
+                else None
+            ),
         )
     if "feedback" in loop_data:
         feedback_win(loop_data["feedback"], llm_data.get("feedback", None) if llm_data else None)

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

@@ -1,5 +1,6 @@
 """Merge the version in different traces"""
 
+from rdagent.app.data_science.conf import DS_RD_SETTING
 from rdagent.components.coder.data_science.pipeline.exp import PipelineTask
 from rdagent.core.proposal import ExpGen
 from rdagent.scenarios.data_science.experiment.experiment import DSExperiment
@@ -46,6 +47,7 @@ def gen(self, trace: DSTrace, selection: tuple[int, ...] = (-1,)) -> DSExperimen
                 type="success",
                 heading="Successful iterations:",
                 success_trial_desc="These trials are the steps or changes that led to the success of the solution to be merged",
+                pipeline=DS_RD_SETTING.coder_on_whole_pipeline,
             )
         else:
             exp_to_merge_fb_desc = T("scenarios.data_science.share:describe.feedback").r(

rdagent/scenarios/data_science/test_eval.py

@@ -103,7 +103,8 @@ def __init__(self) -> None:
     def eval(self, competition: str, workspace: FBWorkspace) -> str:
         workspace.execute(
             env=self.env,
-            entry=f"mlebench grade-sample submission.csv {competition} --data-dir /mle/data | tee mle_score.txt",
+            entry=f"mlebench grade-sample submission.csv {competition} --data-dir /mle/data 2>&1 | tee mle_score.txt",
+            # NOTE: mlebench does not give output to stdout. so 2>&1 is very necessary !!!!!!
         )
         workspace.execute(env=self.env, entry="chmod 777 mle_score.txt")
         return (workspace.workspace_path / "mle_score.txt").read_text()