Align factor coder into new framework (#47)

* use CoSTEER as component name * rename factorimplementation to avoid confusion * rename modelimplementation * align benchmark and evolving evaluators * add scenario to evaluator init function * rename all factorimplementationknowledge in CoSTEER * remove all scenario related information in component * remove useless code --------- Co-authored-by: xuyang1 <xuyang1@microsoft.com>
2026-08-09 13:00:56 +00:00 · 2024-07-05 17:42:00 +08:00
parent f61453fbb1
commit 1d9b4cd2ec
57 changed files with 974 additions and 1093 deletions
@@ -0,0 +1,99 @@
+from abc import abstractmethod
+from pathlib import Path
+from typing import Tuple
+
+from jinja2 import Environment, StrictUndefined
+
+from rdagent.components.coder.factor_coder.factor import FactorExperiment
+from rdagent.core.prompts import Prompts
+from rdagent.core.proposal import (
+    Hypothesis,
+    Hypothesis2Experiment,
+    HypothesisGen,
+    HypothesisSet,
+    Scenario,
+    Trace,
+)
+from rdagent.oai.llm_utils import APIBackend
+
+prompt_dict = Prompts(file_path=Path(__file__).parent / "prompts.yaml")
+
+
+FactorHypothesis = Hypothesis
+
+
+class FactorHypothesisGen(HypothesisGen):
+    def __init__(self, scen: Scenario):
+        super().__init__(scen)
+
+    # The following methods are scenario related so they should be implemented in the subclass
+    @abstractmethod
+    def prepare_context(self, trace: Trace) -> Tuple[dict, bool]:
+        ...
+
+    @abstractmethod
+    def convert_response(self, response: str) -> FactorHypothesis:
+        ...
+
+    def gen(self, trace: Trace) -> FactorHypothesis:
+        context_dict, json_flag = self.prepare_context(trace)
+
+        system_prompt = (
+            Environment(undefined=StrictUndefined)
+            .from_string(prompt_dict["factor_hypothesis_gen"]["system_prompt"])
+            .render(
+                scenario=self.scen.get_scenario_all_desc(),
+                hypothesis_output_format=context_dict["hypothesis_output_format"],
+            )
+        )
+        user_prompt = (
+            Environment(undefined=StrictUndefined)
+            .from_string(prompt_dict["factor_hypothesis_gen"]["user_prompt"])
+            .render(
+                hypothesis_and_feedback=context_dict["hypothesis_and_feedback"],
+                RAG=context_dict["RAG"],
+            )
+        )
+
+        resp = APIBackend().build_messages_and_create_chat_completion(user_prompt, system_prompt, json_mode=json_flag)
+
+        hypothesis = self.convert_response(resp)
+
+        return hypothesis
+
+
+class FactorHypothesis2Experiment(Hypothesis2Experiment[FactorExperiment]):
+    def __init__(self) -> None:
+        super().__init__()
+
+    @abstractmethod
+    def prepare_context(self, hs: HypothesisSet) -> Tuple[dict, bool]:
+        ...
+
+    @abstractmethod
+    def convert_response(self, response: str) -> FactorExperiment:
+        ...
+
+    def convert(self, hs: HypothesisSet) -> FactorExperiment:
+        context, json_flag = self.prepare_context(hs)
+        system_prompt = (
+            Environment(undefined=StrictUndefined)
+            .from_string(prompt_dict["factor_hypothesis2experiment"]["system_prompt"])
+            .render(
+                scenario=hs.trace.scen.get_scenario_all_desc(),
+                experiment_output_format=context["experiment_output_format"],
+            )
+        )
+        user_prompt = (
+            Environment(undefined=StrictUndefined)
+            .from_string(prompt_dict["factor_hypothesis2experiment"]["user_prompt"])
+            .render(
+                hypothesis_and_feedback=context["hypothesis_and_feedback"],
+                factor_list=context["factor_list"],
+                RAG=context["RAG"],
+            )
+        )
+
+        resp = APIBackend().build_messages_and_create_chat_completion(user_prompt, system_prompt, json_mode=json_flag)
+
+        return self.convert_response(resp)