Learning AutoTorch: A 30 Minutes Course

Goals of this tutorial:

• Understand the Hyper-parameter Optimization (HPO) workflow at a high level.

• Understand how to construct search spaces and HPO algorithms for your own python scripts.

• Search Space and Decorator
  • Search Space
  • Decorators for Searchbale Object and Customized Training Scripts
  • Create Searcher and Run with a Configuration
• Experiment Scheduler using HPO Algorithms
  • AutoTorch System Implementatin Logic
  • FIFO Scheduling vs. Early Stopping
• Random Search vs. Bayesian Optimization
  • Create a Reward Function for Toy Experiments
  • Create Training Function
  • Random Search
  • Bayesian Optimization
  • Compare the performance