def process_by_ml_name(ml):
from dataset_loader import DataSetLoader
from sklearn import cross_validation
print 'start ', ml
loader = DataSetLoader()
x, y = loader.loadData()[DataSetLoader.dataset_name[0]]
score_lst = []
for ml in ml:
print 'start cross val'
scores = cross_validation.cross_val_score(ml, x, y, cv=5)
print 'end cross val'
score_lst.append(scores.mean())
return score_lst
def load_dataset(self):
loader = DataSetLoader()
lst = loader.loadData()
return lst
def process(self):
ml_lst = self.gen_ml_lst()
dataset_lst = self.load_dataset()
result = {}
ml_value = ml_lst['svm']
self.log_debug.info('*************************************** ' + self.dataset_name)
all_data = []
self.log_debug.info('***** start ' + self.dataset_name)
if self.dataset_name == 'shuttle':
d_loader = DataSetLoader()
data_value = d_loader.svm_shuttle('data/statlog/shuttle.data')['shuttle']
x_data = data_value[0]
y_data = data_value[1]
if is_run_missing:
print 'before************** ', x_data[0], y_data
x_data, y_data = self.svm_uni_remove(x_data, y_data)
print 'after****************', x_data[0], y_data
elif self.dataset_name == 'segment':
d_loader = DataSetLoader()
data_value = d_loader.load_segment()
x_data = data_value[0]
y_data = data_value[1]
else:
data_value = dataset_lst[self.dataset_name]
x_data = data_value[0]
y_data = data_value[1]
if is_run_missing:
print 'before************** ', x_data[0], y_data
x_data, y_data = self.remove_by_chi2_process(x_data, y_data)
print 'after****************', x_data[0], y_data
datasets_data_lst = []
ml = None
for d_size in self.data_size:
self.log_debug.info('***** start size ' + str(d_size))
ran_num = random.randint(1, 100)
x_train, x_test, y_train, y_test = train_test_split(x_data, y_data, test_size=d_size, random_state=ran_num)
print 'x train ', x_train
self.log_debug.info('********* start cross validation')
ml = self.cross_validation(ml_value, x_train, y_train)
self.log_kernel.info('************* kernel : ' + str(ml.kernel) + " | degree : " + str(ml.degree))
self.log_debug.info('************* end cross validation')
acc_lst = []
f1_lst = []
time_pred = []
total_ins = []
precision_lst = []
recall_lst = []
for i in range(0, self.loop):
self.log_debug.info('loop {} size {} data set {} ml {}'.format(i, d_size, self.dataset_name, 'svm'))
ran_num = random.randint(1, 10000)
x_train, x_test, y_train, y_test = train_test_split(x_data, y_data, test_size=d_size, random_state=ran_num)
try:
ml_c = copy.deepcopy(ml)
ml_c.fit(x_train, y_train)
start = time.time()
y_pred = ml_c.predict(x_test)
except Exception as e:
self.log_error.info(str(e))
total_time = time.time() - start
acc = accuracy_score(y_test, y_pred)
print 'y_test ', y_test
print 'y_pred ', y_pred
fsc = f1_score(y_test, y_pred)
acc_lst.append(acc)
f1_lst.append(fsc)
time_pred.append(total_time)
total_ins.append(len(y_test))
pre_score = precision_score(y_test, y_pred)
recall = recall_score(y_test, y_pred)
precision_lst.append(pre_score)
recall_lst.append(recall)
self.log_debug.info('------------- end loop -----')
datasets_data_lst.append(np.mean(acc_lst))
datasets_data_lst.append(float("{:.5f}".format(np.mean(f1_lst))))
datasets_data_lst.append(np.mean(time_pred))
datasets_data_lst.append(np.mean(total_ins))
self.log.info('---------------------------------------------')
self.log.info('data size ' + str(d_size) + ' data set ' + self.dataset_name)
self.log.info(acc_lst)
self.log.info(f1_lst)
self.log.info(time_pred)
self.log.info(total_ins)
self.log.info('---------------------------------------------')
self.log_debug.info('*********** end size')
self.log.info('ml type ' + str(ml.kernel))
all_data.append(datasets_data_lst)
self.log_debug.info('******* end data set')
self.result[self.dataset_name] = all_data
self.log_debug.info('************ end ml')
file_name = '{}_svm_result_{}.obj'.format(self.dataset_name, self.loop)
pickle.dump(self.result, open(file_name, 'wb'))
self.report_all(result)
y_pred = self.predict(X)
average_score = (accuracy_score(y, y_pred) + f1_score(y, y_pred)) / 2.0
return average_score
def predict(self, x):
f_result = open(self.path_test_data, 'w')
self.__write_data_file(f_result, x, [0] * len(x))
f_result.close()
create_predict = libsvm_path + '/svm-predict' + ' {} {} {}'.format(self.path_test_data,
self.path_model_result,
self.path_result)
print create_predict
os.system(create_predict)
return self.__read_result()
def get_params(self, deep=True):
return {"kernel": self.kernel}
def set_params(self, **parameters):
for parameter, value in parameters.items():
setattr(self, parameter, value)
if __name__ == '__main__':
from dataset_loader import DataSetLoader
from sklearn.cross_validation import train_test_split
loader = DataSetLoader()
x, y = loader.loadData()['heart']
x_train, x_test, y_train, y_test = train_test_split(x, y, test_size=0.75, random_state=42)
ml = LibSVMWrapper(kernel=0)
ml.fit(x_train, y_train)
