C3W1 Assignment - perplexity

[dsong99]

I am having trouble in C3W1 Assignment - perplexity calculation. I passed unit test first 8 examples, but fail 9-th example. I see the preds dimension is 4 while tartget is 3:

preds.shape (2, 1, 7, 3)
target.shape (2, 7)

I tried tf.squeeze(preds)
preds shape after squeeze: (2, 7, 3)

Still get the error:

AssertionError: Fail in Example 9. Batch of 2. Expected -0.31333333427707355 but got -0.27916666759798925

[Deepti_Prasad]

Does this helps ?

NLP C3W1 Assignment about log perplexity

@mejiroalbon Yes the axis=-1 is already mentioned but this test report is trying to point out that when you created the log probability to the prediction list, it is asking you to mention it’s index positioning for prediction list. There is 2-point hint given which mentions To convert the target into the same dimension as the predictions tensor use tf.one_hot with target and preds.shape[-1]. So if you have not mentioned preds.shape[-1], the generalization issue is occurring as axis is alread…

[dsong99]

No. That’s for example 1. I have passed 8 examples. The 9th preds dimension is different.

I read #C3W1 Assignment - perplexity score ,

ValueError: operands could not be broadcast together with shapes (2,7) (2,)

Hi @RyeToast The problem most probably lies in your Exercise 01 -line_to_tensor(). This error says that you tried to slice a scalar (for example, 4[:3], which is not possible). Exercise 01 is a very simple exercise which you should’ve implemented using two lines of code (that are identical to the code couple of cells above). This function should’ve returned ids variable which you could slice (for example, ids[:3], should be possible) because ids should’ve been a tensor. Cheers

[Deepti_Prasad]

did you resolve issue?

[dsong99]

Yes. I nailed it. I guess TensorFlow version is changed in the lab, I’d say this is a bug in the lab. I worked around it, and all tests passed.

Took me two days!!!