feat: grading using mlebench

rdagent/app/kaggle/conf.py

@@ -72,6 +72,9 @@ class Config:
     auto_submit: bool = False
     """Automatically upload and submit each experiment result to Kaggle platform"""
 
+    mle_submit: bool = False
+    """Automatically upload and submit each experiment result to mlebench"""
+
     mini_case: bool = False
     """Enable mini-case study for experiments"""
 

rdagent/app/kaggle/loop.py

@@ -113,6 +113,28 @@ def running(self, prev_out: dict[str, Any]):
                 except Exception as e:
                     logger.error(f"Other exception when use kaggle api:\n{e}")
 
+            if KAGGLE_IMPLEMENT_SETTING.mle_submit:
+                csv_path = exp.experiment_workspace.workspace_path / "submission.csv"
+                try:
+                    result = subprocess.run(
+                        [
+                            "mlebench",
+                            "grade-sample",
+                            str(csv_path.absolute()),
+                            KAGGLE_IMPLEMENT_SETTING.competition,
+                        ],
+                        check=True,
+                        capture_output=True,
+                        text=True,
+                    )
+                    with open(exp.experiment_workspace.workspace_path / "mle_submission_report.txt", "w") as f:
+                        f.write(result.stdout)
+                        f.write(result.stderr)
+                except subprocess.CalledProcessError as e:
+                    logger.error(f"Auto submission failed: \n{e}")
+                except Exception as e:
+                    logger.error(f"Other exception when use mle api:\n{e}")
+
         return exp
 
     skip_loop_error = (ModelEmptyError, FactorEmptyError)

rdagent/scenarios/kaggle/developer/runner.py

@@ -71,6 +71,35 @@ def develop(self, exp: KGModelExperiment) -> KGModelExperiment:
 
         return exp
 
+class MLEModelRunner(KGCachedRunner[KGModelExperiment]):
+    @cache_with_pickle(KGCachedRunner.get_cache_key, KGCachedRunner.assign_cached_result)
+    def develop(self, exp: KGModelExperiment) -> KGModelExperiment:
+        if exp.based_experiments and exp.based_experiments[-1].result is None:
+            exp.based_experiments[-1] = self.init_develop(exp.based_experiments[-1])
+
+        sub_ws = exp.sub_workspace_list[0]
+        if sub_ws is not None:
+            # TODO: There's a possibility of generating a hybrid model (lightgbm + xgboost), which results in having two items in the model_type list.
+            model_type = sub_ws.target_task.model_type
+
+            if sub_ws.code_dict == {}:
+                raise ModelEmptyError("No model is implemented.")
+            else:
+                model_file_name = f"model/model_{model_type.lower()}.py"
+                exp.experiment_workspace.inject_code(**{model_file_name: sub_ws.code_dict["model.py"]})
+        env_to_use = {"PYTHONPATH": "./"}
+
+        result = exp.experiment_workspace.execute(run_env=env_to_use)
+
+        if result is None:
+            raise CoderError("No result is returned from the experiment workspace")
+
+        report_path = exp.experiment_workspace.workspace_path / "mle_submission_report.txt"
+        with open(report_path, "r") as f:
+            exp.result = f.read()
+
+        return exp
+
 
 class KGFactorRunner(KGCachedRunner[KGFactorExperiment]):
     @cache_with_pickle(KGCachedRunner.get_cache_key, KGCachedRunner.assign_cached_result)