pyreal.transformers.NarrativeTransformer#

class pyreal.transformers.NarrativeTransformer(llm=None, openai_api_key=None, num_features=5, gpt_model_type='gpt-4o', context_description='', max_tokens=200, training_examples=None, **kwargs)[source]#

Transforms explanations to narrative (natural-language) form.

__init__(llm=None, openai_api_key=None, num_features=5, gpt_model_type='gpt-4o', context_description='', max_tokens=200, training_examples=None, **kwargs)[source]#

Transforms explanations to narrative (natural-language) form. :param llm: Local LLM object or LLM client object to use to generate narratives. One of llm or openai_api_key must be provided. :type llm: LLM model object :param openai_api_key: OpenAI API key to use :type openai_api_key: string :param x_orig: Input to explain :type x_orig: DataFrame of shape (n_instances, n_features :param num_features: Number of features to include in the explanation, when relevant.

If None, all features will be included

Parameters:

  • gpt_model_type (string) – OpenAI model to use to generate the explanation, if passing in an openai api key

  • context_description (string) – Description of the model’s prediction task, in sentence format. This will be passed to the LLM and may help produce more accurate explanations. For example: “The model predicts the price of houses.”

  • max_tokens (int) – Maximum number of tokens to use in the explanation

  • string (training_examples (dictionary of) –

    list of tuples): Few-shot training examples. Keys are explanation type (currently support: “feature_contributions”) Values are lists of tuples, where the first element is the input to the model

    and the second element is the example output.

    Use the RealApp train_llm functions to populate this

Returns:

list of strings of length n_instances

Narrative version of feature contribution explanation, one item per instance

Methods

__init__([llm, openai_api_key, ...])

Transforms explanations to narrative (natural-language) form. :param llm: Local LLM object or LLM client object to use to generate narratives. One of llm or openai_api_key must be provided. :type llm: LLM model object :param openai_api_key: OpenAI API key to use :type openai_api_key: string :param x_orig: Input to explain :type x_orig: DataFrame of shape (n_instances, n_features :param num_features: Number of features to include in the explanation, when relevant. If None, all features will be included :type num_features: int :param gpt_model_type: OpenAI model to use to generate the explanation, if passing in an openai api key :type gpt_model_type: string :param context_description: Description of the model's prediction task, in sentence format. This will be passed to the LLM and may help produce more accurate explanations. For example: "The model predicts the price of houses." :type context_description: string :param max_tokens: Maximum number of tokens to use in the explanation :type max_tokens: int :param training_examples (dictionary of string: list of tuples): Few-shot training examples. Keys are explanation type (currently support: "feature_contributions") Values are lists of tuples, where the first element is the input to the model and the second element is the example output. Use the RealApp train_llm functions to populate this.

data_transform(x)

Transforms data x to a new feature space.

fit(x, **params)

Fit this transformer to data

fit_transform(x, **fit_params)

Fits this transformer to data and then transforms the same data

inverse_data_transform(x_new)

Wrapper for inverse_data_transform.

inverse_transform(x_new)

Transforms data x_new from new feature space back into the original feature space.

inverse_transform_explanation(explanation)

Transforms the explanation from the second feature space handled by this transformer to the first.

inverse_transform_explanation_additive_feature_contribution(...)

Inverse transforms additive feature contribution explanations

inverse_transform_explanation_additive_feature_importance(...)

Inverse transforms additive feature importance explanations

inverse_transform_explanation_decision_tree(...)

Inverse transforms decision-tree explanations

inverse_transform_explanation_example(...)

Inverse transforms example-based explanations

inverse_transform_explanation_feature_based(...)

Inverse transforms feature-based explanations

inverse_transform_explanation_feature_contribution(...)

Inverse transforms feature contribution explanations

inverse_transform_explanation_feature_importance(...)

Inverse transforms feature importance explanations

inverse_transform_explanation_similar_example(...)

Inverse transforms similar-example-based explanations

parse_feature_contribution_explanation_for_llm(...)

set_flags([model, interpret, algorithm])

set_training_examples(explanation_type, ...)

Set examples of narrative explanations for the request explanation type.

transform(x)

Wrapper for data_transform.

transform_explanation(explanation)

Transforms the explanation from the first feature space handled by this transformer to the second.

transform_explanation_additive_feature_contribution(...)

Transforms additive feature contribution explanations

transform_explanation_additive_feature_importance(...)

Transforms additive feature importance explanations

transform_explanation_decision_tree(explanation)

Inverse transforms feature-based explanations

transform_explanation_example(explanation)

Transforms example-based explanations

transform_explanation_feature_based(explanation)

Transforms feature-based explanations

transform_explanation_feature_contribution(...)

Transforms feature contribution explanations

transform_explanation_feature_importance(...)

Transforms feature importance explanations

transform_explanation_similar_example(...)

Transforms example-based explanations