def do_grid_search(self):
self.get_dataset()
n_folds = mdata.get_grid_search_fold()
skf = StratifiedKFold(n_splits=n_folds)
skf.get_n_splits(self.X_train, self.y_train)
clf, pipeline = self.build_clf_pipeline()
log.debug('get_params: %s' % (pipeline.get_params().keys()))
params = mdata.get_hyper_param_pairs(self.class_name)
if len(params) > 0:
temp_params = {}
for (param, range) in params.items():
if mdata.is_param_float_type(param) is True:
temp_range = [float(value_str) for value_str in range]
elif mdata.is_param_int_type(param) is True:
temp_range = [int(value_str) for value_str in range]
else:
temp_range = range
temp_name = "clf__" + param
temp_params[temp_name] = temp_range
gs = GridSearchCV(estimator=pipeline,
param_grid=temp_params,
n_jobs=-1,
cv=skf,
verbose=2)
gs.fit(self.X_train, self.y_train)
log.debug('GridSearch Params:%s' % gs.param_grid)
log.debug('Best CV accuracy: %g\nBest params: %s\n' %
(gs.best_score_, gs.best_params_))
mdata.set_grid_search_param_values(gs.best_params_)
def save_hyper_param(self):
value_str = str(self.range_combo.currentText())
log.debug('start - index: %s' % value_str)
if value_str != config.txt_default:
if mdata.is_param_float_type(self.param_name) is True:
param_value = float(value_str)
elif mdata.is_param_int_type(self.param_name) is True:
param_value = int(value_str)
mdata.set_hyper_param_value(self.param_name, param_value)
def set_param_value(self, index):
log.debug('start - index: %d' % index)
value_str = str(self.range_combo.currentText())
param_value = None
if value_str != config.txt_default:
if mdata.is_param_float_type(self.param_name) is True:
param_value = float(value_str)
elif mdata.is_param_int_type(self.param_name) is True:
param_value = int(value_str)
mdata.set_hyper_param_value(self.param_name, param_value)