TypeError: unsupported operand type(s) for -: 'int' and '_dynet.Expression'

[sfedia]

Hi! I've encountered an issue related to scalar-matrix arithmetic operations.
Whenever a multiplication, addition, subtraction, or division operation occurs between a float/int and an instance of dynet.Expression, this exception is raised if the float/int precedes a dynet.Expression instance.
If, on the contrary, an instance of dynet.Expression is placed before float/int, no exception occurs. I suppose that this happens due to the strict order of arguments in methods for scalar-matrix operators (matrix goes first, then scalar goes): https://dynet.readthedocs.io/en/latest/operations.html

Traceback (most recent call last):
  File "postcorrection/multisource_wrapper.py", line 63, in <module>
    pretrainer = PretrainHandler(
  File "/Users/fyodorsizov/Documents/git/ocr-post-correction/postcorrection/pretrain_handler.py", line 81, in __init__
    self.pretrain_model(pretrain_src1, pretrain_src2, pretrain_tgt, epochs)
  File "/Users/fyodorsizov/Documents/git/ocr-post-correction/postcorrection/pretrain_handler.py", line 88, in pretrain_model
    self.seq2seq_trainer.train(
  File "/Users/fyodorsizov/Documents/git/ocr-post-correction/postcorrection/seq2seq_trainer.py", line 53, in train
    losses.append(self.model.get_loss(src1, src2, tgt))
  File "/Users/fyodorsizov/Documents/git/ocr-post-correction/postcorrection/multisource_model.py", line 295, in get_loss
    return self.decode_loss(src1, src2, tgt)
  File "/Users/fyodorsizov/Documents/git/ocr-post-correction/postcorrection/multisource_model.py", line 278, in decode_loss
    probs, _ = self.get_pointergen_probs(
  File "/Users/fyodorsizov/Documents/git/ocr-post-correction/postcorrection/multisource_model.py", line 204, in get_pointergen_probs
    copy_probs = a_t * (1 - p_gen)
TypeError: unsupported operand type(s) for -: 'int' and '_dynet.Expression'

Have you seen something like this before? Any idea how to fix this other than changing the order of terms inside the code or overriding __mul__, __add__, etc methods?