Configuration Reference

Single-source reference for config.yaml files used by the CLI and Experiment.from_yaml.

Root Keys

Key	Type	Required	Default	Description
name	string	No	Folder name	Human-friendly experiment name shown in the CLI and used for artifact paths.
description	string	No	""	Freeform description surfaced in the CLI panels.
replicates	integer	No	1	Number of replicates to run for each condition.
steps	list	Yes	–	Ordered list of step factory names from steps.py. Entries can be strings or {factory: {param: value}} mappings.
datasource	mapping or list of mappings	Yes	–	Defines input aliases. Each alias maps to a factory in datasources.py or a DAG reference (see below).
outputs	mapping	No	{}	Declares artifacts and optional loaders/writers from outputs.py.
treatments	mapping	No	{}	Treatment name → parameter dict applied to the context.
hypotheses	list	No	[]	Hypothesis objects built from verifiers.py entries.
cli	mapping	No	{}	CLI metadata for grouping, icons, and ordering.

cli Section

Key	Type	Required	Default	Description
group	string	No	"Graphs" for DAGs, "Experiments" otherwise	Sidebar grouping label.
priority	integer	No	999	Sort order within a group (lower = higher).
icon	string (emoji/char)	No	📈 for DAGs, 🧪 otherwise	Icon shown in the CLI tree.
color	string (hex)	No	null	Optional accent color for the CLI label.
hidden	boolean	No	false	If true, the experiment/graph is omitted from discovery.

datasource Section

Each entry maps an alias to one of:

• A factory name from datasources.py: data: load_data.
• A DAG reference to an upstream artifact: summary: "experiment_a#summary".

Aliases defined as a list of single-item mappings are merged, which lets you spread datasource declarations across files when needed.

outputs Section

Declare artifacts produced by steps so downstream nodes and the CLI know how to read/write them.

Field	Type	Required	Default	Description
file_name	string	No	Alias name	File name to write under the step directory.
writer	string	No	null	Function name in outputs.py that receives (data, path) and writes bytes.
loader	string	No	null	Function name in outputs.py that reads the artifact back in.

Example:

outputs:
  summary:
    file_name: summary.json
    writer: dump_json
    loader: load_json

treatments Section

Map treatment names to context parameters. Treatments inherit by name in DAGs; downstream selections activate same-named upstream treatments automatically.

treatments:
  high_lr:
    lr: 0.05
  dropout:
    rate: 0.2

hypotheses Section

Each list item builds a Hypothesis using factories from verifiers.py.

Field	Type	Required	Description
name	string	Recommended	Display name for the hypothesis; defaults to the verifier/ranker name if omitted.
verifier	string	Yes	Factory name in verifiers.py that returns a verifier callable.
metrics	string or list	Yes	Metric key or keys to pass into the verifier (single string or list of groups).

Example:

hypotheses:
  - name: ttest_accuracy
    verifier: two_sample_ttest
    metrics: ["accuracy"]