Open the chapter list

1. Chapter 1. Start Here
2. Chapter 2. Set Up Your Workspace
3. Chapter 3. Prepare Accelerometer Files
4. Chapter 4. Record Observations
5. Chapter 5. Build Your First Model
6. Chapter 6. Predict with a Model
7. Chapter 7. Understand Your Results
8. Chapter 8. Optimise for Deployment
9. Chapter 9. Scripted Reruns
10. Chapter 10. Troubleshooting

Optimise a Model for Deployment

This is Chapter 8 of 10 in the beginner path.

This chapter is optional. You do not need optimisation to build a useful first model. Optimisation becomes helpful when you want to balance model performance against model size, memory use, or embedded deployment constraints.

Why optimise at all?

A model with the best possible accuracy is not always the most practical model. On an embedded device, for example, you may need a smaller model even if that means giving up a little performance.

So optimisation is really about choosing a trade-off. In moover, that usually means comparing candidates that differ in things like:

• feature subset size
• number of trees
• minimum node size
• rolling features on or off

The idea of a candidate model

When moover optimises, it does not commit to one model immediately. Instead, it trains and evaluates multiple candidate models and writes tables that help you compare them.

In other words, optimisation answers a practical question: Which model is good enough, and small enough, for the job I care about?

Running optimisation

You can turn optimisation on in the training wizard, or you can do it explicitly in code.

library(moover)

spec <- create_spec(
  workspace = list(root = "my_moover_workspace"),
  labels = list(
    tech_file = "tech.csv",
    path = "observations.csv"
  ),
  optimise = list(
    enabled = TRUE
  )
)

# Run only the optimisation stage.
optimise_model(spec)

How to read the candidate tables

The optimisation output usually includes candidate tables that list, for each model:

• accuracy
• macro F1
• model size proxy
• key model settings

A good way to read those tables is to start by deciding what you care about most. For example:

• Do you want the best possible accuracy?
• Do you want the smallest model above a threshold?
• Do you want a balanced compromise?

Once that decision is clear, the candidate tables become much easier to interpret.

A beginner-friendly way to think about the trade-off

If you are new to optimisation, this is a sensible order to follow:

1. build a working model first
2. decide what size or speed constraint matters
3. compare candidates against that constraint
4. choose the smallest model that still performs well enough

That order keeps optimisation grounded in a practical goal rather than turning it into a search for abstract “best” settings.

Move through the tutorial
Previous chapter: Chapter 7. Understand Your Results and Export Folder
Next chapter: Chapter 9. Scripted Reruns and Saved Specs