context_length
Hello,
My understanding is that the algorithm randomly selects past_values and future values from the target values in the training set. Where the past_values start is random. How can the model be enforced to only use the last context_length for the training like the test set? For example, if the target values are:
[1,2,3,4,5,6,7,8,9,10] and the prediction length = 3, context_length+max(lags)=5, the training sample can be one of these (just as examples of many other possible past and future values):
past_values=[0,0,0,1,2] , future_values=[3,4,5]
past_values=[1,2,3,4,5], future_values=[6,7,8]
However, if the same data is used for test set, the respective values would always be:
past_values=[6,7,8,9,10]
Is there a way to enforce the training values to behave like test values, and always use the last context_length, like below:
past_values=[3,4,5,6,7], future values=[8,9,10]
I mean setting last three values as future values, and the 5 previous values as past values.
Thank you