Intent

• A repo to expose some examples of hyperparameter tuning in Azure Machine Learning.
• While hyperparameter tuning is fairly well-documented, when you start to leverage 'R' or more specifically, the 'command' argument in the ScriptRunConfig, there are some additional lines of code to work through to ensure parameters cycle through multiple combinations.
• The 'command' argument is mutually exclusive to the use of 'script/arguments' argument.
• This repo has three examples of hyperparameter tuning, using Python and 'R':
  • iris-standard This is the cleanest, best-documented example with the iris dataset, using Python
  • iris-command-argument This uses the above dataset with Python, but with the command argument vs. the script, arguments parameters.
  • wine-quality-R-command-argument This uses the command argument with 'R' as the base for training
• If on a Mac, the following packages are helpful to install in your conda environment:
  • python-dotenv
  • azureml-defaults
  • azureml-train

File Structure

• LICENSE.TXT
• README.md
• requirements.txt
• scripts
  • "Setup scripts" (not a file structure)
    • create-workspace-sprbac.sh shell script embedded in 'workflow.sh' to create workspace/infra
    • clusters.py creates a cluster; part of 'workflow.sh' process
    • Authentication and environment variables
      • authentication.py Used to authenticate the workspace with a service principal
      • config.json gets created during workflow
      • sub.env subscription info: needs to be in place prior to execution
        • Manually create a file called sub.env with one line: SUB_ID="<enter subscription id>"
      • variables.env gets created during workflow
  • iris-standard
    • target.py entry script to initiate hyperparameter training
    • train.py training script in Python
  • iris-command-argument
    • target.py entry script to initiate hyperparameter training
    • train.py training script in Python
  • wine-quality-R-command-argument
    • Dockerfile Docker file specifications for R environment
    • conda-specs.yaml conda specifications for R environment
    • create_environment.py create custom environment to run 'R' scripts
      • Run this file before running the target.py script
    • target.py entry script to initiate hyperparameter training
    • train.R training script in R
    • wine_quality.csv input data
  • name-generator
    • adjectives.txt used as input into random_name.py
    • nouns.txt used as input into random_name.py
    • random_name.py uses adjectives.txt. and nouns.txt to create a random name