RuntimeError: stack expects each tensor to be equal size, but got [2] at entry 0 and [2, 2] at entry 7

[DJHeisenberg01]

Hi Lambeq support,

I am trying to train a binary classification model using my own dataset, i am using the classic pipeline example in lambeq documentation and when i pass train_dataset to trainer.fit() of PytorchTrainer it throws this error: RuntimeError: stack expects each tensor to be equal size, but got [2] at entry 0 and [2, 2] at entry 7. I am using BobcatParser for sentences to diagram and SpiderAnsatz for circuits.

Thanks a lot for help

[dimkart]

Hi, it seems that not all of your diagrams have the same output wires, you can try to restrict the output of the parser by using the root_cats argument in its constructor.

[DJHeisenberg01]

Hi, thanks for the help, what kind of value can i pass to root_cats? Because if i put as value 'S' 'NP' and 'PP' the parser can't parse some sentence.

[dimkart]

Can you provide the sentence with the problem here? In general, what kind of text (sentences, noun phrases, something else?) your dataset contains?

[DJHeisenberg01]

I have a dataset with a list of security requirements (I need to classify my security requirements by dividing between privacy and non-privacy requirements), so they are sentences that tend to impart rules.
For example if i use as root_cats = [ 'S', 'NP', 'PP'] a sentence that is not being parsed is : "Auto-populates information available in EHR to the patient core data set templates and allows edits and additional entries ."

[dimkart]

I see, these are imperative sentences, the type of which is usually n.r s (i.e. 2 wires in the output) and this is what causes the problem. You have these options:

• Write a rewrite rule (or modifying the diagrams directly if you prefer) so that, when the sentence has two output wires, it adds a box that combines them into one s wire. This should be done at the string diagram level, i.e. before converting to circuits.
• Don't use a syntax-based model; instead, use one of the word-sequence readers, e.g. cups_reader or a stairs_reader, which guarantees that every sentence has a single output wire.

I hope this helps, let us know if not.

[DJHeisenberg01]

Thanks for the help, using cups and stairs it works, if i have to using BobcatParser i'll try to write a rewrite rule, but for now i'll use word-sequence readers.