Chapter 13 Grid search

Learning objectives:

• Use the {dials} package to create tuning grids.
  • Compare and contrast regular and non-regular grids.
  • Use dials::parameters() to examine tuning parameters.
  • Use dials::grid_regular() to create a regular tuning grid.
  • Use dials::grid_*random*() functions to create irregular tuning grids.
• Use tune::tune_grid() to conduct a grid search.
  • Use the grid parameter to specify tuning grids.
• Finalize a tuned model.
  • Use tune::select_*() functions to choose a tuned parameter set.
  • Manually specify a parameter set.
• Improve grid search efficiency.
  • Recognize how {parsnip} uses submodel optimization to make tuning more efficient.
  • Specify {tune} parallel-processing rules using the parallel_over parameter.
  • Use finetune::tune_race_anova() to make tuning more efficient via racing methods.

The call to action:

Tuning machine learning models can be time consuming and computationally expensive.

Thoughtful choices in the experimental design of searches can make them easier to deal with.

---

• Last week: how to tag arguments using tune().

• This week: how to optimize the parameters, a priori.

• Next week: iterative methods.