def from_file(self, config_file):
"""Fill the contents of settings by reading a config file.
Arguments
---------
config_file: str
Source configuration file.
"""
if config_file is None or not os.path.isfile(config_file):
if config_file is not None:
print(f"Didn't find configuration file: {config_file}")
return
config = ConfigParser()
config.optionxform = str
config.read(config_file)
# Read the each of the sections.
for sect in config:
if sect == "global_settings":
for key, value in config.items(sect):
try:
setattr(self, key, SETTINGS_TYPE_DICT[key](value))
except (KeyError, TypeError):
print(f"Warning: value with key '{key}' is ignored "
"(spelling mistake, wrong type?).")
elif sect in [
"model_param", "query_param", "balance_param",
"feature_param"
]:
setattr(self, sect, dict(config.items(sect)))
elif sect != "DEFAULT":
print(f"Warning: section [{sect}] is ignored in "
f"config file {config_file}")
model = get_classifier(self.model)
_convert_types(model.default_param, self.model_param)
balance_model = get_balance_model(self.balance_strategy)
_convert_types(balance_model.default_param, self.balance_param)
query_model = get_query_model(self.query_strategy)
_convert_types(query_model.default_param, self.query_param)
feature_model = get_feature_model(self.feature_extraction)
_convert_types(feature_model.default_param, self.feature_param)
def test_balance(balance_strategy,
n_partition=100,
n_feature=200,
n_sample=100):
model = get_balance_model(balance_strategy)
assert isinstance(model.param, dict)
assert model.name == balance_strategy
X, y = generate_data(n_feature=n_feature, n_sample=n_sample)
for _ in range(n_partition):
n_train = np.random.randint(10, n_sample)
while True:
train_idx = np.random.choice(np.arange(len(y)),
n_train,
replace=False)
num_zero = np.count_nonzero(y[train_idx] == 0)
num_one = np.count_nonzero(y[train_idx] == 1)
if num_zero > 0 and num_one > 0:
break
shared = {"query_src": {}, "current_queries": {}}
X_train, y_train = model.sample(X, y, train_idx, shared)
check_partition(X, y, X_train, y_train, train_idx)