add inf evaluator to factor costeer and some minor improvement (#435)

rdagent/components/coder/factor_coder/CoSTEER/evaluators.py

@@ -133,6 +133,28 @@ class FactorCodeEvaluator(FactorEvaluator):
         return critic_response, None
 
 
+class FactorInfEvaluator(FactorEvaluator):
+    def evaluate(
+        self,
+        implementation: Workspace,
+        gt_implementation: Workspace,
+    ) -> Tuple[str, object]:
+        _, gen_df = self._get_df(gt_implementation, implementation)
+        if gen_df is None:
+            return (
+                "The source dataframe is None. Please check the implementation.",
+                False,
+            )
+        INF_count = gen_df.isin([float("inf"), -float("inf")]).sum().sum()
+        if INF_count == 0:
+            return "The source dataframe does not have any infinite values.", True
+        else:
+            return (
+                f"The source dataframe has {INF_count} infinite values. Please check the implementation.",
+                False,
+            )
+
+
 class FactorSingleColumnEvaluator(FactorEvaluator):
     def evaluate(
         self,
@@ -417,6 +439,9 @@ class FactorValueEvaluator(FactorEvaluator):
                     "Output dataframe has more columns than input feature which is not acceptable in feature processing tasks. Please check the implementation to avoid generating too many columns. Consider this implementation as a failure."
                 )
 
+        feedback_str, inf_evaluate_res = FactorInfEvaluator(self.scen).evaluate(implementation, gt_implementation)
+        conclusions.append(feedback_str)
+
         # Check if the index of the dataframe is ("datetime", "instrument")
         feedback_str, _ = FactorOutputFormatEvaluator(self.scen).evaluate(implementation, gt_implementation)
         conclusions.append(feedback_str)
@@ -465,6 +490,7 @@ class FactorValueEvaluator(FactorEvaluator):
             and row_result <= 0.99
             or output_format_result is False
             or daily_check_result is False
+            or inf_evaluate_res is False
         ):
             decision_from_value_check = False
         else:

rdagent/scenarios/qlib/experiment/prompts.yaml

@@ -65,7 +65,7 @@ qlib_factor_output_format: |-
   0   your factor name  40914 non-null  float64
   dtypes: float64(1)
   memory usage: <ignore>
-  None
+  Notice: The non-null count is OK to be different to the total number of entries since some instruments may not have the factor value on some days.
   One possible format of `result.h5` may be like following:
   datetime    instrument
   2020-01-02  SZ000001     -0.001796

rdagent/scenarios/qlib/experiment/utils.py

@@ -85,10 +85,21 @@ def get_file_desc(p: Path) -> str:
 
         buffer = io.StringIO()
         df.info(verbose=True, buf=buffer, show_counts=False)
+
+        df_info = buffer.getvalue()
+        if isinstance(df.index, pd.MultiIndex):
+            df_info += f"\nMultiIndex names:, {df.index.names})"
+        if "REPORT_PERIOD" in df.columns:
+            one_instrument = df.index.get_level_values("instrument")[0]
+            df_on_one_instrument = df.loc[pd.IndexSlice[:, one_instrument], ["REPORT_PERIOD"]]
+            df_info += f"""
+A snapshot of one instrument, from which you can tell the distribution of the data:
+{df_on_one_instrument.head(10)}
+"""
         return JJ_TPL.render(
             file_name=p.name,
-            type_desc="generated by `df.info(verbose=True, show_counts=False)`",
-            content=buffer.getvalue(),
+            type_desc="h5, generated by `df.info(verbose=True, show_counts=False)` and appendix info",
+            content=df_info,
         )
     elif p.name.endswith(".md"):
         with open(p) as f: