🔧Values & Overrides

This page explains how to provide overrides to a configuration function using dotted keys and nested dictionaries, and which one takes precedence when both are provided.

Basics

A configuration function receives an HP object and returns anything (dict, dataclass, model, etc.). Parameters are named explicitly using name=...:

from hypster import HP, instantiate

def child(hp: HP):
    x = hp.int(10, name="x")
    y = hp.int(20, name="y")
    return {"x": x, "y": y}

def parent(hp: HP):
    return hp.nest(child, name="child")

Dotted keys

Use dotted keys to override nested parameters:

instantiate(parent, values={"child.x": 15})
# => {"x": 15, "y": 20}

Nested dicts

Provide a nested dict under the nested scope name:

instantiate(parent, values={"child": {"x": 25}})
# => {"x": 25, "y": 20}

Precedence: nested dict wins

If you provide both dotted keys and a nested dict for the same scope, the nested dict wins.

instantiate(
    parent,
    values={
        "child.x": 100,         # dotted notation
        "child": {"x": 200},  # nested dict
    },
)
# => {"x": 200, "y": 20}

Deeply nested

Dotted keys can target deeper levels when there is no nested dict override at that level. You can mix and match as needed, keeping in mind that a nested dict for a scope overrides dotted keys for that scope.

Unknown and unreachable parameters

• Unknown or unreachable values (values for parameters on a branch that wasn’t taken) are handled by on_unknown in instantiate:

  • on_unknown="warn" (default): issue a warning

  • on_unknown="raise": raise an error

  • on_unknown="ignore": silently ignore

Hypster will include helpful suggestions when a name looks like a typo.