Skip to content

Training_Prediction_Orchestration.md

Latest commit

 

History

History
72 lines (41 loc) · 4.62 KB

Training_Prediction_Orchestration.md

File metadata and controls

72 lines (41 loc) · 4.62 KB
graph LR
    Training_Orchestrator["Training Orchestrator"]
    Prediction_Orchestrator["Prediction Orchestrator"]
    Model_Graph_Manager["Model Graph Manager"]
    Training_Loop_Executor["Training Loop Executor"]
    Callback_Handler["Callback Handler"]
    Training_Orchestrator -- "invokes" --> Model_Graph_Manager
    Training_Orchestrator -- "invokes" --> Training_Loop_Executor
    Prediction_Orchestrator -- "invokes" --> Model_Graph_Manager
    Training_Loop_Executor -- "calls" --> Callback_Handler
Loading

CodeBoardingDemoContact

Details

The Training & Prediction Orchestration subsystem encompasses components responsible for initiating and overseeing the training processes of text classification models and handling inference (making predictions on new data). It acts as the operational core for the lifecycle of these models.

Training Orchestrator

Manages the overall training workflow for various text classification models. This includes preparing data, configuring training parameters, and initiating the training process. It abstracts the model-specific training logic into a unified interface.

Related Classes/Methods:

Prediction Orchestrator

Oversees the inference process for trained text classification models. This involves handling input data, loading the appropriate model, and generating predictions. It provides a consistent interface for making predictions across different models.

Related Classes/Methods:

Model Graph Manager

Centralized component responsible for creating, compiling, and initializing the Keras model graph. It provides the foundational structure and configuration (e.g., optimizer, loss function) required for both training and prediction.

Related Classes/Methods:

Training Loop Executor

Executes the core training iterations, fitting the model to provided data. It handles both direct data input (fit) and data provided via generators (fit_generator), managing the progression of the training process.

Related Classes/Methods:

Callback Handler

Manages and dispatches callback functions during the model training process. This enables custom actions such as logging, model saving, early stopping, or learning rate scheduling without modifying the core training loop.

Related Classes/Methods:

FAQ