def __init__(self, hyperparameters, default_parameters=None):
"""
:param hyperparameters: A dictionary of the form
```
{
'hp_1': [value1, value2, value3],
'hp_2': [value1, value2, value3],
...
}
```
This format is like the param_grid in SciKit-Learn:
http://scikit-learn.org/stable/modules/grid_search.html#exhaustive-grid-search
:param default_parameters: Default key-value pairs to add to the
dictionary.
"""
self._hyperparameters = hyperparameters
self._default_kwargs = default_parameters or {}
named_hyperparameters = []
for name, values in self._hyperparameters.items():
named_hyperparameters.append([(name, v) for v in values])
self._hyperparameters_dicts = [
ppp.dot_map_dict_to_nested_dict(dict(tuple_list))
for tuple_list in itertools.product(*named_hyperparameters)
]
def generate_random_hyperparameters(self):
hyperparameters = {}
for hp in self._hyperparameters:
hyperparameters[hp.name] = hp.generate()
hyperparameters = ppp.dot_map_dict_to_nested_dict(hyperparameters)
return ppp.merge_recursive_dicts(
hyperparameters,
copy.deepcopy(self._default_kwargs),
ignore_duplicate_keys_in_second_dict=True,
)
def wrapped_function(params):
start_time = time.time()
if flatten_choice_dictionary:
params = flatten_hyperopt_choice_dict(params)
if dotmap_to_nested_dictionary:
params = dot_map_dict_to_nested_dict(params)
loss = function(
merge_recursive_dicts(
params,
extra_function_kwargs,
ignore_duplicate_keys_in_second_dict=True,
))
if maximize:
loss = -loss
if np.isnan(loss):
loss = max_magnitude
return {
'loss': loss,
'status': STATUS_OK,
'params': params,
# -- store other results like this
'eval_time': time.time() - start_time,
}
def test_dotmap_dict_to_nested_dict(self):
a = {'foo': 1, 'bar.baz': 3}
result = ppp.dot_map_dict_to_nested_dict(a)
expected = {'foo': 1, 'bar': {'baz': 3}}
self.assertDictEqual(result, expected)