update for grand prix s4e9

alpha_automl/resource/base_grammar.bnf

@@ -1,5 +1,5 @@
 S -> CLASSIFICATION_TASK | REGRESSION_TASK | CLUSTERING_TASK | TIME_SERIES_FORECAST_TASK | SEMISUPERVISED_TASK | NA_TASK
-CLASSIFICATION_TASK -> IMPUTER ENCODERS FEATURE_GENERATOR FEATURE_SCALER FEATURE_SELECTOR CLASSIFIER CLASSIFICATION_ENSEMBLER
+CLASSIFICATION_TASK -> IMPUTER ENCODERS FEATURE_GENERATOR FEATURE_SCALER FEATURE_SELECTOR CLASSIFIER
 REGRESSION_TASK -> IMPUTER ENCODERS FEATURE_GENERATOR FEATURE_SCALER FEATURE_SELECTOR REGRESSOR REGRESSION_ENSEMBLER
 CLUSTERING_TASK -> IMPUTER ENCODERS FEATURE_GENERATOR FEATURE_SCALER FEATURE_SELECTOR CLUSTERER
 TIME_SERIES_FORECAST_TASK -> IMPUTER TIME_SERIES_FORECASTER | REGRESSION_TASK

alpha_automl/wrapper_primitives/llm_feature_engine.py

@@ -11,14 +11,15 @@
 logger = logging.getLogger(__name__)
 
 class LLMFeatureGenerator(BasePrimitive):
-    def __init__(self, extra_system_prompt=None):
+    def __init__(self, description=None, extra_system_prompt=None):
+        self.description = description
         self.extra_system_prompt = extra_system_prompt
         self.prompt = None
         self.code = None
         pass
 
     def fit(self, X, y=None):
-        self.prompt = build_prompt_from_df(description="", df=X)
+        self.prompt = build_prompt_from_df(description=self.description, df=X)
         self.code = generate_code(self.prompt, self.extra_system_prompt)
         return self
 
@@ -28,7 +29,7 @@ def transform(self, X, y=None):
         access_scope = {"df": X_cp, "pd": pd, "np": np}
         parsed = ast.parse(self.code)
         exec(compile(parsed, filename="<ast>", mode="exec"), access_scope, loc)
-        return np.array(X_cp)
+        return X_cp
 
 def get_prompt(
     df, description, iterative=1, data_description_unparsed=None, samples=None, **kwargs
@@ -55,6 +56,7 @@ def get_prompt(
 This code also drops columns, if these may be redundant and hurt the predictive performance of the downstream classifier (Feature selection). Dropping columns may help as the chance of overfitting is lower, especially if the dataset is small.
 The classifier will be trained on the dataset with the generated columns and evaluated on a holdout set. The evaluation metric is accuracy. The best performing code will be selected.
 Added columns can be used in other codeblocks, dropped columns are not available anymore.
+Remember do not include np.inf, -np.inf in any generated columns.
 
 Code formatting for each added column:
 ```python