def single_model_forecasting(data, lag, h_train, h_test, lr, epoch, batch_size,
hidden_num, method, use_cuda):
# normalize time series
scaler = MinMaxScaler(feature_range=(0, 1))
dataset = scaler.fit_transform(data)
trainData, testData = divideTrainTest(dataset)
flag = True # using RNN format or not
trainX, trainY = create_multi_ahead_samples(trainData,
lag,
h_train,
RNN=flag)
testX, testY = create_multi_ahead_samples(testData, lag, h_test, RNN=flag)
trainY = np.squeeze(trainY).reshape(-1, h_train)
testY = np.squeeze(testY).reshape(-1, h_test)
print("train X shape:", trainX.shape)
print("train y shape:", trainY.shape)
print("test X shape:", testX.shape)
print("test y shape:", testY.shape)
net = train(trainX,
trainY,
epoch=epoch,
lr=lr,
batchSize=batch_size,
lag=lag,
method=method,
hidden_num=hidden_num,
use_cuda=use_cuda)
testPred = predict_iteration(net,
testX,
h_test,
use_cuda=use_cuda,
RNN=flag)
# trainPred = predict_iteration(net, trainX, h_train, use_cuda=use_cuda, RNN=flag)
# print("train pred shape:", trainPred.shape)
print("test pred shape:", testPred.shape)
testPred = scaler.inverse_transform(testPred)
testY = scaler.inverse_transform(testY)
# evaluation
MAE = eval.calcMAE(testY, testPred)
print("test MAE", MAE)
MRSE = eval.calcRMSE(testY, testPred)
print("test RMSE", MRSE)
MAPE = eval.calcMAPE(testY, testPred)
print("test MAPE", MAPE)
SMAPE = eval.calcSMAPE(testY, testPred)
print("test SMAPE", SMAPE)
return testPred, testY
def decompose_ML_forecasting(model, lag, freq, h_train, h_test):
# 序列分解
trend, seasonal, residual = ts_decompose(ts, freq)
print("ts decomposition is finished!")
print("trend shape is", trend.shape)
print("season shape is", seasonal.shape)
print("residual shape is", residual.shape)
# forecasting sub-series independently
trend_pred, trend_y = ML_forecasting(trend,
model,
lag=lag,
train_look_ahead=h_train,
test_look_ahead=h_test)
res_pred, res_y = ML_forecasting(residual,
model,
lag=lag,
train_look_ahead=h_train,
test_look_ahead=h_test)
season_pred, season_y = ML_forecasting(seasonal,
model,
lag=lag,
train_look_ahead=h_train,
test_look_ahead=h_test)
trend_pred = trend_pred.reshape(-1, h_test)
trend_y = trend_y.reshape(-1, h_test)
res_pred = res_pred.reshape(-1, h_test)
res_y = res_y.reshape(-1, h_test)
season_pred = season_pred.reshape(-1, h_test)
season_y = season_y.reshape(-1, h_test)
print("trend_pred shape is", trend_pred.shape)
print("res_pred shape is", res_pred.shape)
print("season_pred shape is", season_pred.shape)
print("trend_y shape is", trend_y.shape)
print("res_y shape is", res_y.shape)
print("season_y shape is", season_y.shape)
testPred = trend_pred + res_pred + season_pred
testY = trend_y + res_y + season_y
# evaluation
MAE = eval.calcMAE(testY, testPred)
print("test MAE", MAE)
MRSE = eval.calcRMSE(testY, testPred)
print("test RMSE", MRSE)
MAPE = eval.calcMAPE(testY, testPred)
print("test MAPE", MAPE)
SMAPE = eval.calcSMAPE(testY, testPred)
print("test SMAPE", SMAPE)
return testPred, testY
def decomposition_model_forecasting(ts, dataset, lag, h_train, h_test, freq, epoch, lr, batch_size, hidden_num, use_cuda, method):
# time series decomposition
trend, seasonal, residual = ts_decompose(ts, freq)
print("ts decomposition is finished!")
print("trend shape is", trend.shape)
print("season shape is", seasonal.shape)
print("residual shape is", residual.shape)
# forecasting sub-series independently
print('Trend')
trend_pred, trend_y = single_model_forecasting(trend, lag=lag, h_train=h_train, h_test=h_test, epoch=epoch, lr=lr,
hidden_num=hidden_num, batch_size=batch_size, method=method, use_cuda=use_cuda)
print('Residual')
res_pred, res_y = single_model_forecasting(residual, lag=lag, h_train=h_train, h_test=h_test, epoch=epoch, lr=lr,
hidden_num=hidden_num, batch_size=batch_size, method=method, use_cuda=use_cuda)
print('Season')
season_pred, season_y = single_model_forecasting(seasonal, lag=lag, h_train=h_train, h_test=h_test, epoch=epoch, lr=lr,
hidden_num=hidden_num, batch_size=batch_size, method=method, use_cuda=use_cuda)
trend_pred = trend_pred.reshape(-1, h_test)
trend_y = trend_y.reshape(-1, h_test)
res_pred = res_pred.reshape(-1, h_test)
res_y = res_y.reshape(-1, h_test)
season_pred = season_pred.reshape(-1, h_test)
season_y = season_y.reshape(-1, h_test)
print("trend_pred shape is", trend_pred.shape)
print("res_pred shape is", res_pred.shape)
print("season_pred shape is", season_pred.shape)
print("trend_y shape is", trend_y.shape)
print("res_y shape is", res_y.shape)
print("season_y shape is", season_y.shape)
testPred = trend_pred + res_pred + season_pred
testY = trend_y + res_y + season_y
# evaluation
MAE = eval.calcMAE(testY, testPred)
print("test MAE", MAE)
MRSE = eval.calcRMSE(testY, testPred)
print("test RMSE", MRSE)
MAPE = eval.calcMAPE(testY, testPred)
print("test MAPE", MAPE)
SMAPE = eval.calcSMAPE(testY, testPred)
print("test SMAPE", SMAPE)
return testPred, testY
def single_model_forecasting(data, lookBack, epoch, lr, batchSize, method,
modelPath, hidden_num, use_cuda):
# 归一化数据
scaler = MinMaxScaler(feature_range=(0, 1))
dataset = scaler.fit_transform(data)
trainData, testData = divideTrainTest(dataset)
flag = True
trainX, trainY = createSamples(trainData, lookBack, RNN=flag)
testX, testY = createSamples(testData, lookBack, RNN=flag)
print("testX shape:", testX.shape)
print("testy shape:", testY.shape)
print("trainX shape:", trainX.shape)
print("trainy shape:", trainY.shape)
net = train(trainX,
trainY,
epoch=epoch,
lr=lr,
batchSize=batchSize,
modelPath=modelPath,
lookBack=lookBack,
method=method,
hidden_num=hidden_num,
use_cuda=use_cuda)
testPred = predict(testX, net, use_cuda=use_cuda)
trainPred = predict(trainX, net, use_cuda=use_cuda)
print("train pred shape:", trainPred.shape)
print("test pred shape:", testPred.shape)
testPred = scaler.inverse_transform(testPred)
testY = scaler.inverse_transform(testY)
MAE = eval.calcMAE(testY, testPred)
print("test MAE", MAE)
MRSE = eval.calcRMSE(testY, testPred)
print("test RMSE", MRSE)
MAPE = eval.calcMAPE(testY, testPred)
print("test MAPE", MAPE)
SMAPE = eval.calcSMAPE(testY, testPred)
print("test SMAPE", SMAPE)
return trainPred, testPred, MAE, MRSE, SMAPE
def decomposition_model_forecasting(ts, dataset, lookBack, freq, epoch, lr,
batchSize, hidden_num, use_cuda):
# 序列分解
trend, seasonal, residual = ts_decompose(ts, freq)
print("fcd decompose is finised!")
print("trend shape is", trend.shape)
print("season shape is", seasonal.shape)
print("residual shape is", residual.shape)
# 分别预测
MODEL_PATH = "../models/ResRNN_model.pkl"
trTrain, trTest, MAE1, MRSE1, SMAPE1 = single_model_forecasting(
trend,
lookBack=lag,
epoch=epoch,
lr=lr,
use_cuda=use_cuda,
batchSize=batchSize,
method=METHOD,
modelPath=MODEL_PATH,
hidden_num=hidden_num)
resTrain, resTest, MAE2, MRSE2, SMAPE2 = single_model_forecasting(
residual,
lookBack=lag,
epoch=epoch,
lr=lr,
use_cuda=use_cuda,
batchSize=batchSize,
method=METHOD,
modelPath=MODEL_PATH,
hidden_num=hidden_num)
trTrain = trTrain.reshape(-1, 1)
trTest = trTest.reshape(-1, 1)
resTrain = resTrain.reshape(-1, 1)
resTest = resTest.reshape(-1, 1)
print("trTrain shape is", trTrain.shape)
print("resTrain shape is", resTrain.shape)
trendPred, resPred = align(trTrain, trTest, lookBack, resTrain, resTest,
lookBack)
print("trendPred shape is", trendPred.shape)
print("resPred shape is", resPred.shape)
# 获取最终预测结果
finalPred = trendPred + seasonal + resPred
trainPred = trTrain + seasonal[lookBack:lookBack +
trTrain.shape[0]] + resTrain
testPred = trTest + seasonal[2 * lookBack + resTrain.shape[0]:] + resTest
# 获得ground-truth数据
data = dataset[freq // 2:-(freq // 2)]
trainY = data[lookBack:lookBack + trTrain.shape[0]]
testY = data[2 * lookBack + resTrain.shape[0]:]
print(trainY.shape)
print(testY.shape)
print(trainPred.shape)
print(testPred.shape)
# 评估指标
MAE = eval.calcMAE(trainY, trainPred)
print("train MAE", MAE)
MRSE = eval.calcRMSE(trainY, trainPred)
print("train MRSE", MRSE)
MAPE = eval.calcMAPE(trainY, trainPred)
print("train MAPE", MAPE)
MAE = eval.calcMAE(testY, testPred)
print("test MAE", MAE)
MRSE = eval.calcRMSE(testY, testPred)
print("test RMSE", MRSE)
MAPE = eval.calcMAPE(testY, testPred)
print("test MAPE", MAPE)
SMAPE = eval.calcSMAPE(testY, testPred)
print("test SMAPE", SMAPE)
return trainPred, testPred, MAE, MRSE, SMAPE
print("trainX shape:", trainX.shape)
print("trainy shape:", trainY.shape)
groud_truth = []
prediction = []
for i in range(len(testX)):
data = testX[i, :].transpose().flatten()
model = ExponentialSmoothing(data,
trend='add',
seasonal='add',
seasonal_periods=12)
model = model.fit(smoothing_level=None)
pred = model.predict(start=0, end=0 + h_test - 1)
real_y = testY[i].tolist()
groud_truth.extend(pred)
prediction.extend(real_y)
print("data:", i)
print(pred)
print(real_y)
groud_truth = np.array(groud_truth).reshape(-1, 1)
prediction = np.array(prediction).reshape(-1, 1)
MAE = eval.calcMAE(groud_truth, prediction)
RMSE = eval.calcRMSE(groud_truth, prediction)
SMAPE = eval.calcSMAPE(groud_truth, prediction)
MAPE = eval.calcMAPE(groud_truth, prediction)
print('Test MAE: %.8f' % MAE)
print('Test RMSE: %.8f' % RMSE)
print("test MAPE", MAPE)
print('Test SMAPE: %.8f' % SMAPE)