Suppose I encode the sequence to sequence dataset as a vector of integer $\{0,1\cdots N\}$ where $N$ is the maximun number of unique words in the dataset.

For example, I am planning to build a chatbot with inputs maximum of 8 words:

Q:Where can I buy these apples ? $$ [8,19,23,5,76,10,2,0] $$ A:At the market downtown $$ [44,7,12,90,0,0,0,0] $$

Can I use multivariate Poisson Regression to fit the data such that when I inputted an unseen sentence

New Input: Where to buy apples ? $$ [8,65,32,10,2,0,0,0] $$ Output: At the market downtown $$ [44,7,12,90,0,0,0,0] $$

Is there any violations in the conditions for using the MVP regression?

  • $\begingroup$ The words have no inherent ordering (maybe alphabetical, but you don’t even seem to be using that), so I am skeptical that this makes sense. For instance, why is it that $market=12$ and $downtown=90?$ $\endgroup$
    – Dave
    Commented Oct 3, 2022 at 14:24
  • $\begingroup$ @Dave I am using the predictors as an identification token through an integer which could be arbitrary (no ordering) but is consistent. $\endgroup$
    – wd violet
    Commented Oct 3, 2022 at 14:48


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.