Sequence Models Week 1

Course Q&A Deep Learning Specialization Sequence Models
1

Hi, I’m really stuck on week 1 exercise 2 part 2.2 Sampling. The instructions are much less clear than previous exercises. It’s not clear what shape the zero vectors should be for step 1, for step 2 I end up with a shape of 27,100 and the probabilities don’t add up to 100 - is that correct? For step 3 the function says a and p must be the same size, but I have a vector of 27 charactors and a y matrix of 27,100 - how to get that to match (ravel is no use) and which vector do I use for the range?
There is no guidance on the shapes of the different variables so this is very tough. Please help!

2

Here are some hints:

  1. x should be one-hot representation of a character. There are vocab_size number of characters in total.
  2. Looking at the forward propagation equations, a_prev should be compatible with W_{aa} to perform dot product.
  3. y holds the probability of the next character. So, use this to sample from the characters that make up the vocabulary.
3

The problem is that is not the correct shape of y. You need to carefully check your computations against the given formulas. Here’s a thread which shows the shapes you should be getting.