'name': 'real'
},
'dragonfly_config': {
'acq_str': 'ucb-ei',
'n_init_rs': 0
}
}
tu = Tuun(config)
data = {
'x': [
[0.0, 0.0],
[0.0, 2.0],
[2.0, 0.0],
[2.0, 2.0],
[1.0, 1.0],
[1.1, 1.1],
],
'y': [15.0, 6.0, 9.0, 12.0, 1.0, 1.5],
}
suggestion = tu.suggest_to_minimize(data)
# maximization
data['y'] = [-1 * v for v in data['y']]
suggestion_max = tu.suggest_to_maximize(data)
print(type(suggestion))
# <class 'numpy.ndarray'>
assert list(suggestion) == list(suggestion_max)
class TuunTuner(Tuner):
"""
Tuun as a custom Tuner for NNI.
"""
def __init__(self,
tuun_config,
optimize_mode="maximize",
initial_data=None):
"""
Parameters
----------
tuun_config : dict
Config to specify Tuun options.
initial_data : dict
Dictionary with keys x (list) and y (1D numpy ndarray).
"""
self._set_tuun(tuun_config)
self._set_data(initial_data)
assert isinstance(tuun_config, dict)
assert optimize_mode in ['minimize', 'maximize']
self._optimize_mode = OptimizeMode(optimize_mode)
def _set_tuun(self, tuun_config):
"""Configure and instantiate self.tuun."""
self.tuun = Tuun(tuun_config)
def _set_data(self, initial_data):
"""Set self.data."""
if initial_data is None:
self.data = Namespace(x=[], y=[])
else:
initial_data = copy.deepcopy(initial_data)
if isinstance(initial_data, dict):
initial_data = Namespace(**initial_data)
self.data = initial_data
elif isinstance(initial_data, Namespace):
self.data = initial_data
else:
raise TypeError(
'initial_data must be either a dict, Namespace, or None')
def update_search_space(self, search_space):
"""
Update search space. Input search_space contains information that the user
pre-defines. The search space is set before generating first hyper-parameters.
Parameters
----------
search_space : dict
Information to define a search space.
"""
self.parameter_keys = list(search_space.keys())
dom_config_list = json2space(search_space)
self.tuun.set_config_from_list(dom_config_list)
def generate_parameters(self, parameter_id, **kwargs):
"""
Return a dict of trial (hyper-)parameters, as a serializable object.
Parameters
----------
parameter_id : int
ID number for given (hyper-)parameters.
Returns
-------
dict
A set of (hyper-)parameters suggested by Tuun.
"""
if self._optimize_mode == OptimizeMode.Minimize:
suggestion = self.tuun.suggest_to_minimize(self.data)
else: # self._optimize_mode is guaranteed as 'min' or 'max'.
suggestion = self.tuun.suggest_to_maximize(self.data)
parsed_dict = self._parse_suggestion_into_dict(suggestion)
return parsed_dict
def receive_trial_result(self, parameter_id, parameters, value, **kwargs):
"""
Record the result from a trial.
Parameters
----------
parameter_id : int
ID number for given (hyper-)parameters.
parameters : dict
A set of (hyper-)parameters, same format returned by generate_parameters().
value : dict_or_float
Final evaluation/objective metric of the trial. If value is dict, it should
have "default" key.
"""
# Define y
if isinstance(value, dict):
y = float(value['default'])
elif isinstance(value, numbers.Number):
y = float(value)
else:
raise TypeError(
'value must be a Number or dict with "default" key')
# Define x
x = list(parameters.values())
# Update self.data
self.data.x.append(x)
self.data.y.append(y)
def _parse_suggestion_into_dict(self, suggestion):
"""Parse suggestion from Tuun into dict for NNI."""
# Keep things simple for now
parsed_dict = dict(zip(self.parameter_keys, suggestion))
return parsed_dict