graph LR
Model_Criticism_Module["Model Criticism Module"]
Model_Hub_Integration_Module["Model Hub Integration Module"]
Criticism_Report_Generator["Criticism Report Generator"]
Posterior_Predictive_Check_PPC_Handler["Posterior Predictive Check (PPC) Handler"]
Hub_Metadata_Manager["Hub Metadata Manager"]
Hub_Model_Interface["Hub Model Interface"]
Base_Model_Class["Base Model Class"]
AnnData_Manager["AnnData Manager"]
Utilities["Utilities"]
Model_Utilities["Model Utilities"]
Model_Criticism_Module -- "uses" --> Base_Model_Class
Model_Criticism_Module -- "uses" --> AnnData_Manager
Model_Criticism_Module -- "uses" --> Utilities
Model_Hub_Integration_Module -- "uses" --> Base_Model_Class
Model_Hub_Integration_Module -- "uses" --> AnnData_Manager
Model_Hub_Integration_Module -- "uses" --> Utilities
Criticism_Report_Generator -- "interacts with" --> Posterior_Predictive_Check_PPC_Handler
Criticism_Report_Generator -- "interacts with" --> Base_Model_Class
Criticism_Report_Generator -- "interacts with" --> Model_Utilities
Posterior_Predictive_Check_PPC_Handler -- "interacts with" --> Base_Model_Class
Hub_Model_Interface -- "uses" --> Hub_Metadata_Manager
Hub_Model_Interface -- "interacts with" --> AnnData_Manager
click Utilities href "https://github.com/CodeBoarding/GeneratedOnBoardings/blob/main/scvi-tools/Utilities.md" "Details"
This component provides essential tools for post-training model assessment and management within scvi-tools. It encompasses functionalities for generating comprehensive criticism reports to evaluate model performance and biological insights, as well as facilitating the seamless sharing, downloading, and management of pre-trained models through integration with external model hubs. This aligns with the project's emphasis on modularity, data-centric design, and robust model evaluation.
This module provides a suite of tools for post-training model assessment and diagnostics. It enables users to generate comprehensive criticism reports and perform statistical checks on model predictions to ensure their validity and biological relevance.
Related Classes/Methods:
scvi.criticism
This module facilitates the seamless sharing, downloading, and management of pre-trained models. It integrates with external model hubs (e.g., Hugging Face Hub, S3) to enable collaborative model development and dissemination.
Related Classes/Methods:
scvi.hub
Responsible for orchestrating the generation of comprehensive criticism reports. It aggregates various evaluation metrics and formats them into a readable output, providing a holistic view of model performance.
Related Classes/Methods:
Performs statistical checks on the posterior predictive distribution of models. It calculates key metrics such as coefficient of variation, zero fraction, and calibration error to assess the model's fit and predictive accuracy against observed data.
Related Classes/Methods:
Manages the metadata associated with models intended for or retrieved from the scvi-tools model hub. It handles the loading, parsing, and representation of essential model information, ensuring proper versioning and discoverability.
Related Classes/Methods:
Provides a unified interface for interacting with models stored in the scvi-tools model hub. It supports core operations such as pushing models to and pulling models from external platforms and provides convenient access to the loaded model and its associated AnnData object.
Related Classes/Methods:
Provides the foundational structure and common functionalities for all probabilistic models within scvi-tools. All specific models inherit from this class, ensuring a consistent interface for training, inference, and data handling.
Related Classes/Methods:
Manages the AnnData objects, which are the primary data structure for single-cell omics data in scvi-tools. It handles data registration, validation, and access, ensuring data integrity and consistency across different model operations.
Related Classes/Methods:
Utilities [Expand]
Provides common helper functions and general utilities used across the scvi-tools project. These include various data manipulation functions, statistical helpers, and general-purpose tools that support the core functionalities.
Related Classes/Methods:
scvi.utils
Contains internal utility functions specifically designed to support operations related to scvi-tools models. These functions often assist in model-specific data transformations or calculations.
Related Classes/Methods: