时间序列数据集训练测试拆分 ML

Question

On machinelearningmastery there is a post about how to create a supervised learning regression type dataset from one time series variable.

For example this:

time, measure
1, 100
2, 110
3, 108
4, 115
5, 120

Can be turned into this below after passing the data through a function series_to_supervised

X, y
?, 100
100, 110
110, 108
108, 115
115, 120
120, ?

In the Multi-Step or Sequence Forecasting section of the machinelearningmastery post, the series_to_supervised can output this below:

   var1(t-2)  var1(t-1)  var1(t)  var1(t+1)
2        0.0        1.0        2        3.0
3        1.0        2.0        3        4.0
4        2.0        3.0        4        5.0
5        3.0        4.0        5        6.0
6        4.0        5.0        6        7.0
7        5.0        6.0        7        8.0
8        6.0        7.0        8        9.0

My question is how would I define the X & y train test split? I am assuming the var1(t) would be the defined as y, right? For example would this be correct below for trainX & trainy? I am experimenting with

#function for time series X,y breakdown
train = series_to_supervised(need_to_train,11,14)

#split data sets
trainX = np.array(train.drop(['var1(t)'],1))
trainy = np.array(train['var1(t)'])

model = XGBRegressor(objective='reg:squarederror', n_estimators=100)

Answer 1

No, var1(t+1) would be the target and taken as y . The whole point is to predict the next step in the future from the current (and past) data.