parameters-and-config

Parameters and Configuration

The only config loader

Since Kedro 0.19.0 the only configuration loader is OmegaConfigLoader. The older ConfigLoader and TemplatedConfigLoader were removed.

# Correct
from kedro.config import OmegaConfigLoader

# WRONG — removed in 0.19.0, do NOT generate these imports
# from kedro.config import ConfigLoader
# from kedro.config import TemplatedConfigLoader

Parameter file structure

Parameters live in files whose names start with parameters inside conf/:

conf/base/parameters.yml          # main parameters
conf/base/parameters_model.yml    # additional — any name starting with "parameters"
conf/local/parameters.yml         # environment override (local takes precedence)

Kedro merges all matching files. Within the same environment, defining the same full key path in two files (for example model_options.learning_rate) raises ValueError; different sub-keys under the same top-level key are fine. Across environments, local overrides base (see "Configuration environments" below).

Keys starting with _ are not hidden in parameters. Unlike the catalog, they stay in the merged result, so nodes can read them (for example with params:_defaults). Top-level _ keys are exempt from duplicate-key errors.

Using parameters in pipelines — the params: prefix

To pass parameters to nodes, use the params: prefix in inputs. Without it Kedro looks for a dataset, not a parameter — this fails silently or raises a missing-dataset error.

# Single parameter — pass by dotted key
Node(func=train, inputs=["data", "params:learning_rate"], outputs="model")

# Nested group — the node receives the whole sub-dict
Node(func=train, inputs=["data", "params:model_options"], outputs="model")

# Entire parameters dict — use bare "parameters"
Node(func=train, inputs=["data", "parameters"], outputs="model")

Do NOT omit the prefix:

# WRONG — Kedro looks for a dataset called "learning_rate"
Node(func=train, inputs=["data", "learning_rate"], outputs="model")