Skip to content

on_boarding.md

Latest commit

 

History

History
99 lines (63 loc) · 8.54 KB

on_boarding.md

File metadata and controls

99 lines (63 loc) · 8.54 KB
graph LR
    DiCE_Core_API["DiCE Core API"]
    Data_Abstraction_Layer["Data Abstraction Layer"]
    Model_Abstraction_Layer["Model Abstraction Layer"]
    Counterfactual_Generation_Engine["Counterfactual Generation Engine"]
    Feasible_Counterfactuals_Module["Feasible Counterfactuals Module"]
    Explanation_Output_Serialization["Explanation Output & Serialization"]
    DiCE_Core_API -- "Initializes & Configures" --> Data_Abstraction_Layer
    DiCE_Core_API -- "Initializes & Configures" --> Model_Abstraction_Layer
    DiCE_Core_API -- "Selects & Orchestrates" --> Counterfactual_Generation_Engine
    Counterfactual_Generation_Engine -- "Requests Predictions & Gradients" --> Model_Abstraction_Layer
    Counterfactual_Generation_Engine -- "Requests Data Transformations" --> Data_Abstraction_Layer
    Counterfactual_Generation_Engine -- "Utilizes Feasibility Constraints" --> Feasible_Counterfactuals_Module
    Feasible_Counterfactuals_Module -- "Accesses Data Distribution" --> Data_Abstraction_Layer
    Counterfactual_Generation_Engine -- "Provides Generated Explanations" --> Explanation_Output_Serialization
    click DiCE_Core_API href "https://github.com/CodeBoarding/GeneratedOnBoardings/blob/main/DiCE/DiCE_Core_API.md" "Details"
    click Data_Abstraction_Layer href "https://github.com/CodeBoarding/GeneratedOnBoardings/blob/main/DiCE/Data_Abstraction_Layer.md" "Details"
    click Model_Abstraction_Layer href "https://github.com/CodeBoarding/GeneratedOnBoardings/blob/main/DiCE/Model_Abstraction_Layer.md" "Details"
    click Counterfactual_Generation_Engine href "https://github.com/CodeBoarding/GeneratedOnBoardings/blob/main/DiCE/Counterfactual_Generation_Engine.md" "Details"
    click Feasible_Counterfactuals_Module href "https://github.com/CodeBoarding/GeneratedOnBoardings/blob/main/DiCE/Feasible_Counterfactuals_Module.md" "Details"
    click Explanation_Output_Serialization href "https://github.com/CodeBoarding/GeneratedOnBoardings/blob/main/DiCE/Explanation_Output_Serialization.md" "Details"
Loading

CodeBoardingDemoContact

Details

The DiCE project implements a highly modular and extensible architecture designed for generating diverse counterfactual explanations. The process begins with the DiCE Core API, which serves as the central orchestrator, initializing and coordinating the Data Abstraction Layer for all data handling and the Model Abstraction Layer for seamless, model-agnostic interaction with various machine learning models. The core of the explanation process is encapsulated within the Counterfactual Generation Engine, which employs different algorithmic strategies. This engine dynamically interacts with both the Model Abstraction Layer to obtain necessary predictions and gradients, and the Data Abstraction Layer for data transformations. For specific requirements like generating realistic counterfactuals, the Feasible Counterfactuals Module provides specialized support. Finally, the generated explanations are structured and prepared for consumption by the Explanation Output & Serialization component, completing the explanation pipeline. This clear separation of concerns facilitates easy extension and maintenance, making DiCE adaptable to new data types, models, and explanation algorithms.

DiCE Core API [Expand]

The primary entry point for users, orchestrating the overall counterfactual explanation generation process. It initializes and coordinates interactions between the Data Abstraction Layer, Model Abstraction Layer, and the chosen Explanation Strategy.

Related Classes/Methods:

Data Abstraction Layer [Expand]

Manages all aspects of data handling, including loading, validation, preprocessing (e.g., one-hot encoding, normalization), and transformation. It provides a unified interface, abstracting away complexities of different data types and formats.

Related Classes/Methods:

Model Abstraction Layer [Expand]

Provides a standardized, unified interface for interacting with various machine learning models (e.g., Keras/TensorFlow, PyTorch, XGBoost). It abstracts framework-specific prediction calls and gradient computations, enabling model agnosticism.

Related Classes/Methods:

Counterfactual Generation Engine [Expand]

Implements the core logic for generating diverse counterfactual explanations. This component encompasses various algorithmic strategies (e.g., KD-tree, Genetic, Gradient-based) that extend a base explainer interface.

Related Classes/Methods:

Feasible Counterfactuals Module [Expand]

A specialized module focused on generating feasible counterfactuals, often leveraging Variational Autoencoders (VAEs) to ensure generated explanations adhere to data distribution constraints. It provides specific methods or models for this purpose, used by certain generation strategies.

Related Classes/Methods:

Explanation Output & Serialization [Expand]

Responsible for structuring, serializing, and managing the generated counterfactual explanations. It prepares the explanations for display, storage, or further analysis, often adhering to defined schemas.

Related Classes/Methods:

FAQ