Training_Orchestrator.md

graph LR
    Training_Orchestrator["Training Orchestrator"]
    Model["Model"]
    Loss_Function["Loss Function"]
    Optimizer["Optimizer"]
    Dataset_DataLoader["Dataset/DataLoader"]
    Hooks["Hooks"]
    Meters["Meters"]
    Distributed_Utilities["Distributed Utilities"]
    Training_Orchestrator -- "calls forward pass on" --> Model
    Training_Orchestrator -- "computes loss using" --> Loss_Function
    Training_Orchestrator -- "triggers optimization on" --> Optimizer
    Training_Orchestrator -- "manages data flow from" --> Dataset_DataLoader
    Training_Orchestrator -- "invokes" --> Hooks
    Training_Orchestrator -- "updates" --> Meters
    Training_Orchestrator -- "utilizes" --> Distributed_Utilities
    click Training_Orchestrator href "https://github.com/CodeBoarding/GeneratedOnBoardings/blob/main/ClassyVision/Training_Orchestrator.md" "Details"
    click Hooks href "https://github.com/CodeBoarding/GeneratedOnBoardings/blob/main/ClassyVision/Hooks.md" "Details"

Loading

Details

The Training Orchestrator subsystem in ClassyVision is centered around the ClassificationTask component, which orchestrates the entire machine learning training and evaluation lifecycle.

Training Orchestrator [Expand]

The core orchestrator of the training and evaluation lifecycle for classification tasks. It manages the entire flow, from data loading and model execution to loss computation, optimization, and distributed training coordination. It acts as the central control unit, defining the training and evaluation steps.

Related Classes/Methods:

• classy_vision.tasks.classification_task.ClassificationTask:106-1358

Model

Represents the neural network model used for predictions.

Related Classes/Methods:

• classy_vision.models.classy_model.ClassyModel:130-420

Loss Function

Calculates the discrepancy between model predictions and ground truth.

Related Classes/Methods:

• classy_vision.losses.classy_loss.ClassyLoss:12-71

Optimizer

Updates model parameters based on the computed loss.

Related Classes/Methods:

• classy_vision.optim.classy_optimizer.ClassyOptimizer:45-277

Dataset/DataLoader

Manages and provides data batches for training and evaluation.

Related Classes/Methods:

• classy_vision.dataset.Dataset

Hooks [Expand]

Provides extension points for custom logic at various stages of the training lifecycle.

Related Classes/Methods:

• classy_vision.hooks.classy_hook.ClassyHook:29-121

Meters

Tracks and aggregates performance metrics during training and evaluation.

Related Classes/Methods:

• classy_vision.meters.Meter

Distributed Utilities

Provides functionalities for distributed training operations.

Related Classes/Methods:

• classy_vision.generic.distributed_util.DistributedUtil

FAQ