[Chapter 15] Behaviour of Variable Length Sequences in RNN

In the SketchRNN Exercise, The behaviour of variable length sequences is defined very succintly.
In the solution :

This is what is used to process after batching ☝. All sequences are of different length. So I investigated a bit further into how parsing was done and how length of different sequences are handled, here parsing is done on each batch since map is applied after batching.


Here I have 3 Examples of different lengths. I have parsed them together using parse_example. The first output is the initial shape of these 3 Examples. Does parse_example internally take consider the largest example here` 53 by 3` and then construct a sparse tensor of size `3 by 53 by 3` which is same as `3 by 159` and fill the missing values by zero?