[meta]

General information about the run.

[meta]
project_name = "medpipe_classifier_demo"
run_mode = "audit"
verbose = 1

Key	Type	Default	Description
project_name	string	required	Name of the project. It cannot be empty. It is used to name the saved pipeline file, <project_name>_fitted.joblib.
run_mode	string	"audit"	Which stages of the pipeline are run. See run modes.
verbose	string, integer or boolean	"compact"	How much is printed to the console. See verbosity.

Run modes

The run mode controls whether the models are tuned with cross-validation and whether figures are produced. Models are always fitted on the training set, evaluated on the test set, and saved.

run_mode	Cross-validation	Figures	Typical use
"fast"	No	No	Quick checks while setting up a configuration.
"cv"	Yes	No	Tuning and comparing models without producing figures.
"eval"	No	Yes	Producing figures for models whose settings are already chosen.
"audit"	Yes	Yes	The full run, for reporting.

The run mode also decides which sections must be present:

• [workflow.validation.cross_validation] is required in "cv" and "audit", the modes that run cross-validation.
• [display] is required in "audit" and "eval".

Note

A list-valued hyperparameter (for example learning_rate = [0.1, 0.05]) is a grid to search over when grid_search = true in a mode that runs cross-validation. In "fast" and "eval" there is no search, and the first value of the list is used.

Every run is written to a new, numbered folder (artifacts/v1, artifacts/v2, and so on). The folder always contains the models, the evaluation results, the resolved configuration, the environment description and the execution log. Cross-validation results (CV/) are added in "cv" and "audit", and figures (plots/) in "eval" and "audit".

Verbosity

verbose only changes what is printed to the console. The execution log saved in the run folder always records everything, down to the debug level.

Level	Accepted values	Console output
Quiet	"quiet", "warning", 0, false	Warnings and errors only.
Compact	"compact", "progress", 1	The main steps of the run (data loading, fitting, evaluation).
Info	"info", "detailed", 2, true	Every informational message.
Debug	"debug", 3	Everything, including internal details.

To change the verbosity for a single run without editing the file, pass verbose_override when creating the pipeline:

pipe = MedpipeClassifier("config.toml", verbose_override="debug")