Configuration Records

A configuration record stores settings for a computation.

In ML code, configuration values might include:

learning rate;
number of epochs;
batch size;
random seed;
input path;
output path.

Putting those values in one record makes the function call clearer.

Pass configuration as one object

Instead of this:

run_experiment(0.1, 3, 42)

prefer this when the settings need names:

The second version is longer, but it is less mysterious.

Pass a configuration record

Ready to run.

Keep configuration boring

A good configuration record is plain and explicit:

Avoid hiding computation in configuration objects. Their job is to name the settings a computation needs.

Milestone shape

By the end of this chapter, you should be comfortable reading code shaped like this:

The record names the settings. The function does the work.

Exercise: Why use a configuration record?

Which answer best describes the reason to pass a configuration record?

Choose the main reason

To give related settings clear names.To make the computation automatically faster.To make Python validate every value at runtime.

Select one choice, then check.

HintFocus on readability

Compare a call containing three unexplained positional values with a record whose fields name each setting.

SolutionConfiguration records name settings

The main benefit here is clarity: related settings travel together with visible field names. A dataclass does not automatically speed up the computation or enforce every type hint at runtime.