Masking for different dates in different examples (using LSTM)

DL4J
1

I have multiple training sets that look like this (feature values don’t matter, the “Date” field is a timestamp):
Example #1:

   Date  |  Feature1  |  Feature2
   123      4                  56
   124      7                  44
   126      3                  60

Example #2:

    Date  |  Feature1  |  Feature2
    123      3                  52
    125      5                  54
    126      8                  38

My Labels for #1:

    1
    2
    3

My Labels for #2:

    4
    5
    6

So in first example data for for “Date=125” is missing and in second example - for “Date=124”. I know that there is AlignmentMode parameter for SequenceRecordReaderDataSetIterator but as far I as I understand it can only fill up the data if something is missing in the beginning or the end (for example, if my example #1 didn’t have a row with “Date=124”). Is it possible to provide training data like I have it in the examples above or do I need to fill with zeros the features/labels row for “Date=125” in #1 and for “Date=124” in #2 before creating SequenceRecordReaderDataSetIterator?