Terminology guide

Understanding the language of Pyreal

Here, we offer a short summary of the specific terminology Pyreal used throughout the Pyreal development guides and codebase.

General Concepts

Term	Definition
model	A machine learning predictor, defined as an object with a .predict() function.
x	Data, in the form of a pandas DataFrame, usually of shape (n_instances, n_features).
x_train	Training data for the model and explanation algorithms
explanation	An explanation of a model or model prediction
local explanation	An explanation of a specific model prediction
global explanation	An explanation of a model overall
feature space	A specific format of data (for example, ML models take in data in a specific feature space)

Pyreal Classes

Term	Definition
RealApp	Pyreal objects that hold important information and generate various types of interpretable explanations
Explainer	Pyreal objects that take in data and a model and return a specific type of explanation
Transformer	Pyreal objects that transform data and explanations from one feature space to another
Explanation	Pyreal objects that hold information about a specific generated explanation, as produced by Explainers

Feature Spaces

Term	Definition
x_orig, explanation_orig	Data/explanation using the original feature space (whatever feature space the data starts in).
x_algorithm, explanation_algorithm	Data/explanation using the feature space expected by the explanation algorithm (explanation-ready feature space)
x_model, explanation_model	Data/explanation using the feature space the model expects
x_interpret, explanation_interpret	Data/explanation using the most human-readable feature space available
model transformers	Transformers that transform data between the original feature space to the model-ready feature space. Transformers with model=True flag.
algorithm transformers	Transformers that transform data and explanations between the original feature space to the explanation-algorithm-ready feature spaces. Transformers with algorithm=True flag.
interpret transformers	Transformers that transform data and explanations between the original feature space to the interpretable feature space. Transformers with interpret=True flag.