Error in Dinosaur programming assignment exercise 4

Course Q&A Deep Learning Specialization Sequence Models
,
1

I get an error in the optimization call. I am guessing that either my X or Y is not formed correctly but I am not sure how to fix it. I am forming both by:

In the case of X, I am adding [None] to single_sample_ix

In the case of Y, I am adding single_sample_ix to [ix_newline]

Here is the error:

---------------------------------------------------------------------------
IndexError                                Traceback (most recent call last)
<ipython-input-31-725c093d6b91> in <module>
----> 1 parameters, last_name = model(data.split("\n"), ix_to_char, char_to_ix, 22001, verbose = True)
      2 
      3 assert last_name == 'Trodonosaurus\n', "Wrong expected output"
      4 print("\033[92mAll tests passed!")

<ipython-input-30-61c3d62593d9> in model(data_x, ix_to_char, char_to_ix, num_iterations, n_a, dino_names, vocab_size, verbose)
     61         # Perform one optimization step: Forward-prop -> Backward-prop -> Clip -> Update parameters
     62         # Choose a learning rate of 0.01
---> 63         curr_loss, gradients, a_prev = optimize(X, Y, a_prev, parameters, learning_rate = 0.01)
     64 
     65         ### END CODE HERE ###

<ipython-input-18-229d16b0dfd3> in optimize(X, Y, a_prev, parameters, learning_rate)
     32 
     33     # Forward propagate through time (≈1 line)
---> 34     loss, cache = rnn_forward(X, Y, a_prev, parameters)
     35 
     36     # Backpropagate through time (≈1 line)

~/work/W1A2/utils.py in rnn_forward(X, Y, a0, parameters, vocab_size)
    100 
    101         # Update the loss by substracting the cross-entropy term of this time-step from it.
--> 102         loss -= np.log(y_hat[t][Y[t],0])
    103 
    104     cache = (y_hat, a, x)

IndexError: only integers, slices (`:`), ellipsis (`...`), numpy.newaxis (`None`) and integer or boolean arrays are valid indices
2

Yes, this is frequently a problem for people. I’ve added some print statements in my model() code to show what it should look like:

initial loss 23.070858062030304
len(X) 12
type(Y) = <class 'list'>
Y = [20, 21, 18, 9, 1, 19, 1, 21, 18, 21, 19, 0]
len(Y) 12
 X =  [None, 20, 21, 18, 9, 1, 19, 1, 21, 18, 21, 19] 
 Y =        [20, 21, 18, 9, 1, 19, 1, 21, 18, 21, 19, 0] 

j =  0 idx =  0
single_example = turiasaurus
single_example_chars ['t', 'u', 'r', 'i', 'a', 's', 'a', 'u', 'r', 'u', 's']
single_example_ix [20, 21, 18, 9, 1, 19, 1, 21, 18, 21, 19]
 X =  [None, 20, 21, 18, 9, 1, 19, 1, 21, 18, 21, 19] 
 Y =        [20, 21, 18, 9, 1, 19, 1, 21, 18, 21, 19, 0] 

Iteration: 0, Loss: 23.087336

So you can see that I’ve added the token None as the first element of the list.

3

Note that None is a python token and a totally different thing than "None", which is a python string.

4

Got it! My compositions of X and Y were ok, it was my syntax on single_sample_ix that was wrong. I had never used python composition before. Thanks!

2 Likes
5

That’s great news that you were able to get to a solution based on that approximate hint!

6

Thanks! This final course is definitely more demanding. Three programming assignments this week with trickier programming techniques. But RNNs are very interesting!

2 Likes
7

C5 starts out difficult, and gets tougher from there.

1 Like
8

Indeed, but it’s totally worth it. As you say, the material is really interesting and highly relevant. It culminates with the core tech that underlies all the current LLMs.