feat: integrate azure deepseek r1 (#591)

* fix several task & integrate deepseek R1

* fix CI

---------

Co-authored-by: Xu Yang <xuyang1@microsoft.com>

rdagent/components/coder/data_science/ensemble/eval.py

@@ -50,7 +50,7 @@ class EnsembleCoSTEEREvaluator(CoSTEEREvaluator):
         }
         de = DockerEnv(conf=ds_docker_conf)
 
-        fname = "ensemble_test.txt"
+        fname = "test/ensemble_test.txt"
         test_code = (DIRNAME / "eval_tests" / "ensemble_test.txt").read_text()
         test_code = (
             Environment(undefined=StrictUndefined)

rdagent/components/coder/data_science/ensemble/exp.py

@@ -7,4 +7,7 @@ from typing import Dict, Optional
 from rdagent.components.coder.CoSTEER.task import CoSTEERTask
 from rdagent.core.utils import cache_with_pickle
 
-EnsembleTask = CoSTEERTask
+
+# Because we use isinstance to distinguish between different types of tasks, we need to use sub classes to represent different types of tasks
+class EnsembleTask(CoSTEERTask):
+    pass

rdagent/components/coder/data_science/feature/eval.py

@@ -52,7 +52,7 @@ class FeatureCoSTEEREvaluator(CoSTEEREvaluator):
         de = DockerEnv(conf=ds_docker_conf)
 
         # TODO: do we need to clean the generated temporary content?
-        fname = "feature_test.py"
+        fname = "test/feature_test.py"
         test_code = (DIRNAME / "eval_tests" / "feature_test.txt").read_text()
         implementation.inject_files(**{fname: test_code})
 

rdagent/components/coder/data_science/feature/exp.py

@@ -7,4 +7,7 @@ from typing import Dict, Optional
 from rdagent.components.coder.CoSTEER.task import CoSTEERTask
 from rdagent.core.utils import cache_with_pickle
 
-FeatureTask = CoSTEERTask
+
+# Because we use isinstance to distinguish between different types of tasks, we need to use sub classes to represent different types of tasks
+class FeatureTask(CoSTEERTask):
+    pass

rdagent/components/coder/data_science/model/eval.py

@@ -62,7 +62,7 @@ class ModelGeneralCaseSpecEvaluator(CoSTEEREvaluator):
         }
         de = DockerEnv(conf=ds_docker_conf)
 
-        fname = "model_test.py"
+        fname = "test/model_test.py"
         test_code = (
             (DIRNAME / "eval_tests" / "model_test.txt").read_text().replace("model01", target_task.name)
         )  # only check the model changed this time

rdagent/components/coder/data_science/model/exp.py

@@ -11,6 +11,7 @@ from rdagent.oai.llm_utils import md5_hash
 from rdagent.utils.env import DockerEnv, DSDockerConf
 
 
+# Because we use isinstance to distinguish between different types of tasks, we need to use sub classes to represent different types of tasks
 class ModelTask(CoSTEERTask):
     def __init__(
         self,

rdagent/components/coder/data_science/raw_data_loader/eval.py

@@ -54,7 +54,7 @@ class DataLoaderCoSTEEREvaluator(CoSTEEREvaluator):
         de = DockerEnv(conf=ds_docker_conf)
 
         # TODO: do we need to clean the generated temporary content?
-        fname = "data_loader_test.py"
+        fname = "test/data_loader_test.py"
         test_code = (DIRNAME / "eval_tests" / "data_loader_test.txt").read_text()
         implementation.inject_files(**{fname: test_code})
         stdout = implementation.execute(env=de, entry=f"python {fname}")

rdagent/components/coder/data_science/raw_data_loader/exp.py

@@ -12,3 +12,8 @@ from rdagent.utils.agent.tpl import T
 from rdagent.utils.env import DockerEnv, DSDockerConf
 
 DataLoaderTask = CoSTEERTask
+
+
+# Because we use isinstance to distinguish between different types of tasks, we need to use sub classes to represent different types of tasks
+class DataLoaderTask(CoSTEERTask):
+    pass

rdagent/components/coder/data_science/workflow/eval.py

@@ -98,9 +98,9 @@ class WorkflowGeneralCaseSpecEvaluator(CoSTEEREvaluator):
             stdout += "\nSubmission file (submission.csv) is not generated."
         else:
             base_check_code = (DIRNAME / "eval_tests" / "submission_format_test.txt").read_text()
-            implementation.inject_files(**{"submission_format_test.py": base_check_code})
+            implementation.inject_files(**{"test/submission_format_test.py": base_check_code})
             # stdout += "----Submission Check 1-----\n"
-            stdout += implementation.execute(env=de, entry="python submission_format_test.py")
+            stdout += implementation.execute(env=de, entry="python test/submission_format_test.py")
 
             # MLEBench Check
             # !!! Since we are running on a sampled dataset, mlebench check is not required.
@@ -109,9 +109,9 @@ class WorkflowGeneralCaseSpecEvaluator(CoSTEEREvaluator):
             #     .read_text()
             #     .replace("<competition_id>", self.scen.competition)
             # )
-            # implementation.inject_files(**{"mle_submission_format_test.py": mle_check_code})
+            # implementation.inject_files(**{"test/mle_submission_format_test.py": mle_check_code})
             # stdout += "----Submission Check 2-----\n"
-            # stdout += implementation.execute(env=mde, entry=f"python mle_submission_format_test.py")
+            # stdout += implementation.execute(env=mde, entry=f"python test/mle_submission_format_test.py")
 
         system_prompt = T(".prompts:workflow_eval.system").r(
             scenario=self.scen.get_scenario_all_desc(),

rdagent/components/coder/data_science/workflow/exp.py

@@ -7,4 +7,7 @@ from typing import Dict, Optional
 from rdagent.components.coder.CoSTEER.task import CoSTEERTask
 from rdagent.core.utils import cache_with_pickle
 
-WorkflowTask = CoSTEERTask
+
+# Because we use isinstance to distinguish between different types of tasks, we need to use sub classes to represent different types of tasks
+class WorkflowTask(CoSTEERTask):
+    pass