graph LR
ExperimentManager["ExperimentManager"]
Recorder["Recorder"]
Exp["Exp"]
TaskGenerator["TaskGenerator"]
TaskManager["TaskManager"]
Operator["Operator"]
OnlineManager["OnlineManager"]
OnlineUpdater["OnlineUpdater"]
Exp -- "delegates to" --> ExperimentManager
Exp -- "delegates to" --> Recorder
ExperimentManager -- "utilizes" --> Recorder
TaskGenerator -- "creates tasks for" --> TaskManager
OnlineManager -- "utilizes" --> Operator
OnlineManager -- "coordinates with" --> OnlineUpdater
The qlib workflow subsystem is designed to manage the lifecycle of quantitative research experiments and online trading operations. The Exp component acts as a high-level entry point, abstracting interactions with the core ExperimentManager and Recorder components. ExperimentManager is responsible for orchestrating experiment runs, while Recorder handles the persistence of experiment artifacts. For task-driven workflows, TaskGenerator creates individual tasks that are then managed and executed by TaskManager. In the context of online operations, OnlineManager coordinates the entire process, leveraging Operator for execution and OnlineUpdater for continuous model and prediction updates. This modular design ensures clear separation of concerns and facilitates efficient management of complex quantitative workflows.
The central orchestrator for managing the lifecycle of experiments. It handles the creation, initiation, termination, searching, and listing of experiment runs, ensuring reproducibility and tracking in quantitative research.
Related Classes/Methods:
Manages the persistence and retrieval of experiment-related artifacts, parameters, and metrics. It ensures that all outputs and configurations of an experiment run are properly saved and can be reloaded for analysis or further use.
Related Classes/Methods:
Serves as a high-level, user-facing facade for experiment management. It simplifies interactions with ExperimentManager and Recorder, providing a more abstract interface for users to define and run experiments.
Related Classes/Methods:
Responsible for generating individual tasks that constitute a larger research or deployment workflow. This often involves segmenting data or defining specific execution steps based on criteria like time.
Related Classes/Methods:
Manages the state and execution of the tasks generated by TaskGenerator. It provides functionalities for querying, inserting, replacing, fetching, and running these tasks, ensuring orderly workflow progression.
Related Classes/Methods:
Provides the fundamental execution capabilities for online serving and real-time execution of trading strategies. This includes generating, executing, updating, and simulating online operations.
Related Classes/Methods:
Orchestrates the entire online workflow, from initial model training and routine updates to preparing signals for live serving. It acts as the central coordinator for continuous online operations.
Related Classes/Methods:
Manages the process of updating online predictions or models, including preparing and retrieving necessary data for continuous updates to ensure the system operates with the most current information.
Related Classes/Methods: