Choose Key Parameters
Use all parameters (the default)
Omit key= to use every parameter as part of the key:
Use this when every parameter genuinely changes the result.
Specify key parameters explicitly
Use key= to exclude parameters that don't define the experiment, for example
computational settings like n_iter, batch_size, or verbose:
Now only alpha and seed determine whether a run is skipped, regardless of n_iter.
Exclude parameters with default values from the key
If a parameter has a default value and you call the function both with and without it, the two calls will be treated as different runs unless you exclude it from the key:
# without key=, these are different runs:
run_experiment(alpha=0.1, seed=1) # n_iter=1000 (default)
run_experiment(alpha=0.1, seed=1, n_iter=500) # n_iter=500
# with key=['alpha', 'seed'], both are the same run:
run_experiment(alpha=0.1, seed=1)
run_experiment(alpha=0.1, seed=1, n_iter=500) # skipped
See How stet Works for a fuller discussion of how parameter identity is determined.