from __future__ import annotations import inspect import logging import warnings from collections import defaultdict from enum import Enum from functools import wraps from itertools import product from typing import ( Any, Callable, DefaultDict, Dict, Iterable, List, Mapping, NamedTuple, Optional, Tuple, TypeVar, Union, ) import pandas as pd from pandas import DataFrame from typing_extensions import TypeAlias from phoenix.evals.evaluators import LLMEvaluator from phoenix.evals.exceptions import PhoenixTemplateMappingError from phoenix.evals.executors import ExecutionStatus, get_executor_on_sync_context from phoenix.evals.models import BaseModel, OpenAIModel, set_verbosity from phoenix.evals.templates import ( ClassificationTemplate, MultimodalPrompt, PromptOptions, PromptTemplate, normalize_classification_template, ) from phoenix.evals.utils import ( NOT_PARSABLE, get_tqdm_progress_bar_formatter, openai_function_call_kwargs, parse_openai_function_call, printif, snap_to_rail, ) logger = logging.getLogger(__name__) ColumnName: TypeAlias = str Label: TypeAlias = str Score: TypeAlias = Optional[float] Explanation: TypeAlias = Optional[str] Record: TypeAlias = Mapping[str, Any] Index: TypeAlias = int # snapped_response, explanation, response ParsedLLMResponse: TypeAlias = Tuple[Optional[str], Optional[str], str, str] class ClassificationStatus(Enum): DID_NOT_RUN = ExecutionStatus.DID_NOT_RUN.value COMPLETED = ExecutionStatus.COMPLETED.value COMPLETED_WITH_RETRIES = ExecutionStatus.COMPLETED_WITH_RETRIES.value FAILED = ExecutionStatus.FAILED.value MISSING_INPUT = "MISSING INPUT" PROCESSOR_TYPE = TypeVar("PROCESSOR_TYPE") def deprecate_dataframe_arg(func: Callable[..., Any]) -> Callable[..., Any]: # Remove this once the `dataframe` arg in `llm_classify` is no longer supported @wraps(func) def wrapper(*args: Any, **kwargs: Any) -> Any: signature = inspect.signature(func) if "dataframe" in kwargs: warnings.warn( "`dataframe` argument is deprecated; use `data` instead", DeprecationWarning, stacklevel=2, ) kwargs["data"] = kwargs.pop("dataframe") bound_args = signature.bind_partial(*args, **kwargs) bound_args.apply_defaults() return func(*bound_args.args, **bound_args.kwargs) return wrapper @deprecate_dataframe_arg def llm_classify( data: Union[pd.DataFrame, List[Any]], model: BaseModel, template: Union[ClassificationTemplate, PromptTemplate, str], rails: List[str], data_processor: Optional[Callable[[PROCESSOR_TYPE], PROCESSOR_TYPE]] = None, system_instruction: Optional[str] = None, verbose: bool = False, use_function_calling_if_available: bool = True, provide_explanation: bool = False, include_prompt: bool = False, include_response: bool = False, include_exceptions: bool = False, max_retries: int = 10, exit_on_error: bool = True, run_sync: bool = False, concurrency: Optional[int] = None, progress_bar_format: Optional[str] = get_tqdm_progress_bar_formatter("llm_classify"), ) -> pd.DataFrame: """ Classifies each input row of the dataframe using an LLM. Returns a pandas.DataFrame where the first column is named `label` and contains the classification labels. An optional column named `explanation` is added when `provide_explanation=True`. Args: data (Union[pd.DataFrame, List[Any]): A collection of data which can contain template variables and other information necessary to generate evaluations. If a passed a DataFrame, there must be column names that match the template variables. If passed a list, the elements of the list will be mapped to the template variables in the order that the template variables are defined. model (BaseEvalModel): An LLM model class. template (Union[ClassificationTemplate, PromptTemplate, str]): The prompt template as either an instance of PromptTemplate, ClassificationTemplate or a string. If a string, the variable names should be surrounded by curly braces so that a call to `.format` can be made to substitute variable values. rails (List[str]): A list of strings representing the possible output classes of the model's predictions. data_processor (Optional[Callable[[T], T]]): An optional callable that is used to process the input data before it is mapped to the template variables. This callable is passed a single element of the input data and can return either a pandas.Series with indices corresponding to the template variables or an iterable of values that will be mapped to the template variables in the order that the template variables are defined. system_instruction (Optional[str], optional): An optional system message. verbose (bool, optional): If True, prints detailed info to stdout such as model invocation parameters and details about retries and snapping to rails. Default False. use_function_calling_if_available (bool, default=True): If True, use function calling (if available) as a means to constrain the LLM outputs. With function calling, the LLM is instructed to provide its response as a structured JSON object, which is easier to parse. provide_explanation (bool, default=False): If True, provides an explanation for each classification label. A column named `explanation` is added to the output dataframe. include_prompt (bool, default=False): If True, includes a column named `prompt` in the output dataframe containing the prompt used for each classification. include_response (bool, default=False): If True, includes a column named `response` in the output dataframe containing the raw response from the LLM. max_retries (int, optional): The maximum number of times to retry on exceptions. Defaults to 10. exit_on_error (bool, default=True): If True, stops processing evals after all retries are exhausted on a single eval attempt. If False, all evals are attempted before returning, even if some fail. run_sync (bool, default=False): If True, forces synchronous request submission. Otherwise evaluations will be run asynchronously if possible. concurrency (Optional[int], default=None): The number of concurrent evals if async submission is possible. If not provided, a recommended default concurrency is set on a per-model basis. progress_bar_format(Optional[str]): An optional format for progress bar shown. If not specified, defaults to: llm_classify |{bar}| {n_fmt}/{total_fmt} ({percentage:3.1f}%) " "| ⏳ {elapsed}<{remaining} | {rate_fmt}{postfix}". If 'None' is passed in specifically, the progress_bar log will be disabled. Returns: pandas.DataFrame: A dataframe where the `label` column (at column position 0) contains the classification labels. If provide_explanation=True, then an additional column named `explanation` is added to contain the explanation for each label. The dataframe has the same length and index as the input dataframe. The classification label values are from the entries in the rails argument or "NOT_PARSABLE" if the model's output could not be parsed. The output dataframe also includes three additional columns in the output dataframe: `exceptions`, `execution_status`, and `execution_seconds` containing details about execution errors that may have occurred during the classification as well as the total runtime of each classification (in seconds). """ concurrency = concurrency or model.default_concurrency # clients need to be reloaded to ensure that async evals work properly model.reload_client() use_openai_function_call = ( use_function_calling_if_available and isinstance(model, OpenAIModel) and model.supports_function_calling ) model_kwargs = ( openai_function_call_kwargs(rails, provide_explanation) if use_openai_function_call else {} ) eval_template = normalize_classification_template(rails=rails, template=template) prompt_options = PromptOptions(provide_explanation=provide_explanation) labels: Iterable[Optional[str]] = [None] * len(data) explanations: Iterable[Optional[str]] = [None] * len(data) printif(verbose, f"Using prompt:\n\n{eval_template.prompt(prompt_options)}") if generation_info := model.verbose_generation_info(): printif(verbose, generation_info) def _map_template(data: pd.Series[Any]) -> MultimodalPrompt: try: variables = {var: data[var] for var in eval_template.variables} empty_keys = [k for k, v in variables.items() if v is None] if empty_keys: raise PhoenixTemplateMappingError( f"Missing template variables: {', '.join(empty_keys)}" ) return eval_template.format( variable_values=variables, options=prompt_options, ) except KeyError as exc: raise PhoenixTemplateMappingError(f"Missing template variable: {exc}") def _process_response(response: str) -> Tuple[str, Optional[str]]: if not use_openai_function_call: if provide_explanation: unrailed_label, explanation = ( eval_template.extract_label_from_explanation(response), response, ) printif( verbose and unrailed_label == NOT_PARSABLE, f"- Could not parse {repr(response)}", ) else: unrailed_label = response explanation = None else: unrailed_label, explanation = parse_openai_function_call(response) return snap_to_rail(unrailed_label, rails, verbose=verbose), explanation def _normalize_to_series( data: PROCESSOR_TYPE, ) -> pd.Series[Any]: if isinstance(data, pd.Series): return data variable_count = len(eval_template.variables) if variable_count == 1: return pd.Series({eval_template.variables[0]: data}) elif variable_count > 1: if isinstance(data, str): raise ValueError("The data cannot be mapped to the template variables") elif isinstance(data, Iterable): return pd.Series( { template_var: input_val for template_var, input_val in zip(eval_template.variables, data) } ) return pd.Series() async def _run_llm_classification_async( input_data: PROCESSOR_TYPE, ) -> ParsedLLMResponse: with set_verbosity(model, verbose) as verbose_model: if data_processor: maybe_awaitable_data = data_processor(input_data) if inspect.isawaitable(maybe_awaitable_data): processed_data = await maybe_awaitable_data else: processed_data = maybe_awaitable_data else: processed_data = input_data prompt = _map_template(_normalize_to_series(processed_data)) response = await verbose_model._async_generate( prompt, instruction=system_instruction, **model_kwargs ) inference, explanation = _process_response(response) return inference, explanation, response, str(prompt) def _run_llm_classification_sync( input_data: PROCESSOR_TYPE, ) -> ParsedLLMResponse: with set_verbosity(model, verbose) as verbose_model: if data_processor: processed_data = data_processor(input_data) if inspect.isawaitable(processed_data): raise ValueError("Cannot run the data processor asynchronously.") else: processed_data = input_data prompt = _map_template(_normalize_to_series(processed_data)) response = verbose_model._generate( prompt, instruction=system_instruction, **model_kwargs ) inference, explanation = _process_response(response) return inference, explanation, response, str(prompt) fallback_return_value: ParsedLLMResponse = (None, None, "", "") executor = get_executor_on_sync_context( _run_llm_classification_sync, _run_llm_classification_async, run_sync=run_sync, concurrency=concurrency, tqdm_bar_format=progress_bar_format, max_retries=max_retries, exit_on_error=exit_on_error, fallback_return_value=fallback_return_value, timeout=model._timeout, ) list_of_inputs: Union[Tuple[Any], List[Any]] if isinstance(data, pd.DataFrame): list_of_inputs = [row_tuple[1] for row_tuple in data.iterrows()] dataframe_index = data.index elif isinstance(data, (list, tuple)): list_of_inputs = data dataframe_index = pd.Index(range(len(data))) else: raise ValueError("Invalid 'data' input type.") results, execution_details = executor.run(list_of_inputs) labels, explanations, responses, prompts = zip(*results) all_exceptions = [details.exceptions for details in execution_details] execution_statuses = [details.status for details in execution_details] execution_times = [details.execution_seconds for details in execution_details] classification_statuses = [] for exceptions, status in zip(all_exceptions, execution_statuses): if exceptions and isinstance(exceptions[-1], PhoenixTemplateMappingError): classification_statuses.append(ClassificationStatus.MISSING_INPUT) else: classification_statuses.append(ClassificationStatus(status.value)) return pd.DataFrame( data={ "label": labels, **({"explanation": explanations} if provide_explanation else {}), **({"prompt": prompts} if include_prompt else {}), **({"response": responses} if include_response else {}), **({"exceptions": [[repr(exc) for exc in excs] for excs in all_exceptions]}), **({"execution_status": [status.value for status in classification_statuses]}), **({"execution_seconds": [runtime for runtime in execution_times]}), }, index=dataframe_index, ) class RunEvalsPayload(NamedTuple): evaluator: LLMEvaluator record: Record def run_evals( dataframe: DataFrame, evaluators: List[LLMEvaluator], provide_explanation: bool = False, use_function_calling_if_available: bool = True, verbose: bool = False, concurrency: Optional[int] = None, ) -> List[DataFrame]: """ Applies a list of evaluators to a dataframe. Outputs a list of dataframes in which each dataframe contains the outputs of the corresponding evaluator applied to the input dataframe. Args: dataframe (DataFrame): A pandas dataframe in which each row represents a record to be evaluated. All template variable names must appear as column names in the dataframe (extra columns unrelated to the template are permitted). evaluators (List[LLMEvaluator]): A list of evaluators. provide_explanation (bool, optional): If True, provides an explanation for each evaluation. A column named "explanation" is added to each output dataframe. use_function_calling_if_available (bool, optional): If True, use function calling (if available) as a means to constrain the LLM outputs. With function calling, the LLM is instructed to provide its response as a structured JSON object, which is easier to parse. verbose (bool, optional): If True, prints detailed info to stdout such as model invocation parameters and details about retries and snapping to rails. concurrency (Optional[int], default=None): The number of concurrent evals if async submission is possible. If not provided, a recommended default concurrency is set on a per-model basis. Returns: List[DataFrame]: A list of dataframes, one for each evaluator, all of which have the same number of rows as the input dataframe. """ # use the minimum default concurrency of all the models if concurrency is None: if len(evaluators) == 0: concurrency = 1 else: concurrency = min(evaluator.default_concurrency for evaluator in evaluators) # use the maximum timeout of all the evaluators timeout = max( ( evaluator._model._timeout for evaluator in evaluators if evaluator._model._timeout is not None ), default=None, ) # clients need to be reloaded to ensure that async evals work properly for evaluator in evaluators: evaluator.reload_client() async def _arun_eval( payload: RunEvalsPayload, ) -> Tuple[Label, Score, Explanation]: return await payload.evaluator.aevaluate( payload.record, provide_explanation=provide_explanation, use_function_calling_if_available=use_function_calling_if_available, verbose=verbose, ) def _run_eval( payload: RunEvalsPayload, ) -> Tuple[Label, Score, Explanation]: return payload.evaluator.evaluate( payload.record, provide_explanation=provide_explanation, use_function_calling_if_available=use_function_calling_if_available, verbose=verbose, ) executor = get_executor_on_sync_context( _run_eval, _arun_eval, concurrency=concurrency, tqdm_bar_format=get_tqdm_progress_bar_formatter("run_evals"), exit_on_error=True, fallback_return_value=(None, None, None), timeout=timeout, ) total_records = len(dataframe) payloads = [ RunEvalsPayload(evaluator=evaluator, record=row) for evaluator, (_, row) in product(evaluators, dataframe.iterrows()) ] eval_results: List[DefaultDict[Index, Dict[ColumnName, Union[Label, Explanation]]]] = [ defaultdict(dict) for _ in range(len(evaluators)) ] results, _ = executor.run(payloads) for index, (label, score, explanation) in enumerate(results): evaluator_index = index // total_records row_index = index % total_records eval_results[evaluator_index][row_index]["label"] = label eval_results[evaluator_index][row_index]["score"] = score if provide_explanation: eval_results[evaluator_index][row_index]["explanation"] = explanation eval_dataframes: List[DataFrame] = [] for eval_result in eval_results: eval_data = [eval_result[row_index] for row_index in range(len(eval_result))] eval_dataframes.append(DataFrame(eval_data, index=dataframe.index)) return eval_dataframes