Matrix D workings

Course Q&A Natural Language Processing NLP with Probabilistic Models
1

I understand how to get C matrix and how to do the backward process. I can’t get the D Matrix. Can you provide a numerical example please?

My confusion comes from the example you have in the Backward Pass. I don’t understand why the max probability from the previous step would be different (hence a different index) among the cells in the next column.

Screenshot 2022-08-26 at 12.38.46
Screenshot 2022-08-26 at 12.38.46539×584 50.7 KB

For example, in last column of C the max prob is in the first cell. So we go to the first cell in the last column of D and see that the max prob in the previous column is in row 3. and so on.

But in row 3 of the last column of D we see that the max prob came from row 4 of the previous column. I would expect the index to be the same as the previous example, since the max prob in the previous column is in row 3.

Hope this helps

2

Hi @gkouro

This is just an example, and the idea is that not all w_5 column values should be 3, because there could be different paths that lead to higher probability to get t_3 (not necessarily of max w_4 but the whole combination of w_i).

1 Like