Skip to content

on_boarding.md

Latest commit

 

History

History
96 lines (60 loc) · 6.61 KB

on_boarding.md

File metadata and controls

96 lines (60 loc) · 6.61 KB
graph LR
    Experiment_Orchestrator["Experiment Orchestrator"]
    NAS_HPO_Algorithms["NAS/HPO Algorithms"]
    Model_Search_Space["Model & Search Space"]
    Data_Management["Data Management"]
    Training_Evaluation_Engine["Training & Evaluation Engine"]
    Benchmarking_Analysis["Benchmarking & Analysis"]
    Experiment_Orchestrator -- "orchestrates execution" --> NAS_HPO_Algorithms
    Experiment_Orchestrator -- "triggers training" --> Training_Evaluation_Engine
    Experiment_Orchestrator -- "submits results" --> Benchmarking_Analysis
    NAS_HPO_Algorithms -- "samples architectures" --> Model_Search_Space
    NAS_HPO_Algorithms -- "requests evaluation" --> Training_Evaluation_Engine
    Model_Search_Space -- "provides model definition" --> Training_Evaluation_Engine
    Data_Management -- "supplies data" --> Training_Evaluation_Engine
    Data_Management -- "supplies benchmark data" --> Benchmarking_Analysis
    Training_Evaluation_Engine -- "outputs metrics" --> Benchmarking_Analysis
    click Experiment_Orchestrator href "https://github.com/CodeBoarding/GeneratedOnBoardings/blob/main/AutoDL-Projects/Experiment_Orchestrator.md" "Details"
    click NAS_HPO_Algorithms href "https://github.com/CodeBoarding/GeneratedOnBoardings/blob/main/AutoDL-Projects/NAS_HPO_Algorithms.md" "Details"
    click Model_Search_Space href "https://github.com/CodeBoarding/GeneratedOnBoardings/blob/main/AutoDL-Projects/Model_Search_Space.md" "Details"
    click Data_Management href "https://github.com/CodeBoarding/GeneratedOnBoardings/blob/main/AutoDL-Projects/Data_Management.md" "Details"
    click Training_Evaluation_Engine href "https://github.com/CodeBoarding/GeneratedOnBoardings/blob/main/AutoDL-Projects/Training_Evaluation_Engine.md" "Details"
    click Benchmarking_Analysis href "https://github.com/CodeBoarding/GeneratedOnBoardings/blob/main/AutoDL-Projects/Benchmarking_Analysis.md" "Details"
Loading

CodeBoardingDemoContact

Details

The AutoDL-Projects framework is designed as a modular research toolkit for Automated Machine Learning (AutoML), specifically focusing on Neural Architecture Search (NAS) and Hyperparameter Optimization (HPO).

Experiment Orchestrator [Expand]

Manages the overall lifecycle of AutoML/NAS experiments, from setup to execution and result collection, driven by configurations. It serves as the central control plane.

Related Classes/Methods:

NAS/HPO Algorithms [Expand]

Encapsulates the core logic for various Neural Architecture Search and Hyperparameter Optimization algorithms, implementing the specific search strategies to find optimal architectures or hyperparameters.

Related Classes/Methods:

Model & Search Space [Expand]

Defines neural network models, including specialized structures for NAS (e.g., supernets, cells), and specifies the possible architectural configurations or hyperparameter ranges for AutoML/NAS. This component is crucial for defining what can be searched.

Related Classes/Methods:

Data Management [Expand]

Handles the loading, preprocessing, and augmentation of datasets for training, validation, and evaluation across various experiments and benchmarks. It ensures data is prepared and accessible for the entire pipeline.

Related Classes/Methods:

Training & Evaluation Engine [Expand]

Implements fundamental routines for model training, validation, and performance metric calculation. This includes optimizers, schedulers, and the core training loop, forming the backbone of model learning.

Related Classes/Methods:

Benchmarking & Analysis [Expand]

Provides tools for interacting with established NAS benchmarks (e.g., NAS-Bench-201, NATS-Bench), collecting statistics, and visualizing experimental results. This component is vital for comparing and understanding algorithm performance.

Related Classes/Methods:

FAQ