def _convert_hyperparams_to_vizier_params()

in src/python/tensorflow_cloud/tuner/utils.py [0:0]

• 56 lines of code
• 15 McCabe index (conditional complexity)

def _convert_hyperparams_to_vizier_params(
    hyperparams: hp_module.HyperParameters) -> List[Any]:
    """Converts HyperParameters to a list of ParameterSpec in study_config."""
    param_type = []
    for hp in hyperparams.space:
        param = {}
        param["parameter"] = hp.name
        if isinstance(hp, hp_module.Choice):
            values = hp.values
            if isinstance(values[0], str):
                param["type"] = _CATEGORICAL
                param["categorical_value_spec"] = {"values": values}
            else:
                param["type"] = _DISCRETE
                param["discrete_value_spec"] = {"values": values}
        elif isinstance(hp, hp_module.Int):
            if hp.step is None or hp.step == 1:
                param["type"] = _INTEGER
                param["integer_value_spec"] = {
                    "min_value": hp.min_value,
                    "max_value": hp.max_value,
                }
                if hp.sampling is not None:
                    param.update(_get_scale_type(hp.sampling))
            else:
                # Note: hp.max_value is inclusive, while the end index of
                # range() is exclusive, hence the +1
                values = list(range(hp.min_value, hp.max_value + 1, hp.step))
                param["type"] = _DISCRETE
                param["discrete_value_spec"] = {"values": values}
        elif isinstance(hp, hp_module.Float):
            if hp.step is None:
                param["type"] = _DOUBLE
                param["double_value_spec"] = {
                    "min_value": hp.min_value,
                    "max_value": hp.max_value,
                }
                if hp.sampling is not None:
                    param.update(_get_scale_type(hp.sampling))
            else:
                # Match how KerasTuner generates the range
                values = np.arange(
                    hp.min_value, hp.max_value + 1e-7, step=hp.step).tolist()
                param["type"] = _DISCRETE
                param["discrete_value_spec"] = {"values": values}
        elif isinstance(hp, hp_module.Boolean):
            param["type"] = _CATEGORICAL
            param["categorical_value_spec"] = {"values": ["True", "False"]}
        elif isinstance(hp, hp_module.Fixed):
            if isinstance(hp.value, (str, bool)):
                param["type"] = _CATEGORICAL
                param["categorical_value_spec"] = {"values": [str(hp.value)]}
            else:
                param["type"] = _DISCRETE
                param["discrete_value_spec"] = {"values": [float(hp.value)]}
        else:
            raise ValueError(
                "`HyperParameter` type not recognized: {}".format(hp))

        param_type.append(param)

    return param_type