minimalist hyperparam configuration files stored as YAML
from dataclasses import dataclass
from mlparams import from_yaml, to_yaml
@dataclass(frozen=True)
class SomeOtherConfiguration:
dtype: str
shape: tuple[int, ...]
@dataclass(frozen=True)
class SomeConfiguration:
foo: str
bar: dict[str, int]
subconfig: SomeOtherConfiguration
config = SomeConfiguration(
foo="foo",
bar={
"asdf": 5
},
subconfig=SomeOtherConfiguration(
dtype="float32",
shape=(2, 3, 4)
)
)
# store to YAML file
with open("myconfig.yaml", "w") as f:
to_yaml(config, f)
# load from YAML file with correct type info
with open("myconfig.yaml", "r") as f:
config = from_yaml(SomeConfiguration, f)Will create / parse the following YAML config file
foo: "foo"
bar:
asdf: 5
subconfig:
dtype: "float32"
shape: [2, 3, 4]