def get_lstm_diff(KPI_ID_name):
if not os.path.exists(output_path):
os.makedirs(output_path)
# if __name__ == '__main__':
train_data_path = 'resources/train.csv'
test_data_path = 'resources/test.csv'
augment_data_path = 'resources/augment_data/'
test_augment_data_path = 'resources/test_augment_data/'
full_result_path = 'resources/result/prediction.csv'
split_result_path = 'resources/result_split/prediction.csv'
# output_path = 'resources/label_prediction_plot'
test_data_raw = pd.read_csv(test_data_path)
train_data_raw = pd.read_csv(train_data_path)
KPI_LIST, KPI_ID = SplitKPIList(train_data_raw)
KPI_LIST_test, KPI_ID_test = SplitKPIList(test_data_raw)
# KPI_ID_name = '07927a9a18fa19ae' # 07927a9a18fa19ae 76f4550c43334374 a5bf5d65261d859a
index = KPI_ID.index(KPI_ID_name)
train_X_raw = pd.DataFrame(KPI_LIST[index])
X_train_reserve = train_X_raw.copy()
X_train_reserve = pd.DataFrame(X_train_reserve)
# y_reserve = pd.DataFrame()
y_reserve = X_train_reserve['value']
y_reserve = pd.DataFrame(y_reserve)
y_reserve = y_reserve.shift(-1)
y_reserve = pd.DataFrame(y_reserve)
y_reserve.ffill(inplace=True)
y_reserve = pd.DataFrame(y_reserve)
# print("y_reserve:", y_reserve)
y_reserve = y_reserve.values
X_reserve = X_train_reserve['value'].values
test_index = KPI_ID_test.index(KPI_ID_name)
test_X_raw = pd.DataFrame(KPI_LIST_test[test_index])
# X_raw = pd.DataFrame(KPI_LIST[index])
X_test = test_X_raw.copy()
del X_test['KPI ID']
del X_test['timestamp']
label = KPI_LIST[index].copy()['label']
KPI_LIST[index]['label'] = KPI_LIST[index]['label'].map(conv2Noneflag)
KPI_LIST[index] = KPI_LIST[index].dropna(axis=0)
del KPI_LIST[index]['KPI ID']
del KPI_LIST[index]['label']
del KPI_LIST[index]['timestamp']
df = pd.DataFrame(KPI_LIST[index])
# print("df:\n", df)
df_y = df.shift(-1)
df_y.ffill(inplace=True)
# print("df_y:\n", df_y)
y_test = X_test.shift(-1)
y_test.ffill(inplace=True)
test_data = X_test.values
test_label = y_test.values
train_X_scaler, train_y_scaler, train_scaled_X, train_scaled_y = scale_data(
df, df_y)
test_scaled_X = train_X_scaler.transform(test_data)
test_scaled_y = train_y_scaler.transform(test_label)
pre_item = 'value'
model_file = './BasicLSTM_output/' + KPI_ID_name + '_lstm_model.h5'
# 加载LSTM模型
if os.path.exists(model_file):
lstm_model = load_model(model_file)
else:
# 训练LSTM模型
lstm_model = fit_lstm(train_scaled_X, train_scaled_y, model_file)
# print('{}模型文件不存在'.format(model_file))
# exit(0)
test_diff_df = get_diff(test_data, test_scaled_y, test_scaled_X,
lstm_model, train_y_scaler, KPI_ID_name + 'test')
X_reserve = X_reserve.reshape(-1, 1)
train_scaled_X = train_X_scaler.transform(X_reserve)
train_scaled_y = train_y_scaler.transform(y_reserve)
train_diff_df = get_diff(X_train_reserve.values, train_scaled_y,
train_scaled_X, lstm_model, train_y_scaler,
KPI_ID_name + '_train')
return train_diff_df, test_diff_df
is_shift = False
shift = 2
# score_threshold = 0.997
# KPI_ID_name = '76f4550c43334374' 8a20c229e9860d0c
train_data_path = 'resources/train.csv'
test_data_path = 'resources/test.csv'
augment_data_path = 'resources/augment_data/'
test_augment_data_path = 'resources/test_augment_data/'
full_result_path = 'resources/result/prediction.csv'
rand_result_path = 'resources/result/rand_expand_prediction.csv'
output_path = 'resources/label_prediction_plot'
manual_features_path = 'resources/manual_feature'
test_data_raw = pd.read_csv(test_data_path)
train_data_raw = pd.read_csv(train_data_path)
KPI_LIST, KPI_ID = SplitKPIList(train_data_raw)
KPI_LIST_test, KPI_ID_test = SplitKPIList(test_data_raw)
# print(KPI_LIST)
# print('KPI_ID:', KPI_ID)
# KPI_ID:
# ['71595dd7171f4540',
# '88cf3a776ba00e7c',
# 'affb01ca2b4f0b45',
# '769894baefea4e9e',
# '02e99bd4f6cfb33f',
# '54e8a140f6237526',
# 'b3b2e6d1a791d63a',
# '07927a9a18fa19ae',
# 'a40b1df87e3f1c87',
# 'c58bfcbacb2822d1',