rllm.llm.FeatLLMEngineer¶

class rllm.llm.FeatLLMEngineer(file_path: str, metadata_path: str, task_info_path: str, llm: BaseChatModel | BaseLLM | None = None, *, query_num: int = 5, shots: int = 4, test_size: float | int = 0.2, target_column: str | None = None, seed: int = 0)[source]¶

Bases: object

A feature engineering class that leverages a language model to generate feature extraction rules and functions based on input data, metadata, and task descriptions.

df¶

The input dataset loaded from a CSV file.

Type:: pd.DataFrame

data_name¶

The name of the dataset (derived from the file name).

Type:: str

metadata¶

Metadata information loaded from a JSON file.

Type:: dict

task_info¶

Task description loaded from a TXT file.

Type:: str

llm¶

The language model used for querying.

Type:: Union[LC.BaseChatModel, LC.BaseLLM]

query_num¶

Number of queries to generate for the language model.

Type:: int

shots¶

Number of examples to use for few-shot learning.

Type:: int

test_size¶

Proportion or number of test samples.

Type:: Union[float, int]

target_column¶

The target column in the dataset.

Type:: str

seed¶

Random seed for reproducibility.

Type:: int