last line of predict funtion in <model>_utils.py

[cirdeCyL]

There is the Condition
any(k != border for k in pep_lens[border:])
in the predict function of lstm_utils.py and ae_utils.py.
What is the purpose of this condition?

[kruus]

In ae_utils, I've decided that line is a logic bug after trying to run some applications with batch
size > 1. My guess is there's an attempt to magically "mark" the 'missing' value in a terminal partial
batch (in get_full_batches) by a peptide of length 'missing'.

Unfortunately, the predict step looks at the terminal peptide length for full batches too, which
is incorrect. The code I received tried to handle this nonsense by setting batch size to 1.
This means for batch size > 1, asserting the number of outputs equal to the number of inputs
will sometimes fail.

So after some length debug and quick-tests, I changed my ae_utils.py to remove this broken
partial-batch-creation-and-detection approach. Instead I:

1. mark missing peptides as 4 padding peptides (index 0) or [ list(pep_atox.keys())[0] * 4 ] * missing
2. mark missing tcrs as 'X' terminator tcrs (more generally [ list(tcr_atox.keys())[-1] ] * missing
3. detect the size of the current full/partial batch with bs = (padded_peps[:0] != 0).sum().item(),
   and only extend the python list with this number of entries

No guarantees, but fixing along these lines now allows me to get_full_batches with batch sizes > 1
where a final partial batch needs to be created. :)