Improving Deep Neural Networks: Hyperparameter Tuning, Regularization - Week3 assignment

I am doing the one hot encoding in this assignment. The instruction says it must return a column vector

But when I look into the test it is trying to compare the result against a row vector

So even though my function is returning a column vector it keeps failing at the below assert.

Not sure what I am missing, can someone help

assert np.allclose(result, [0., 1. ,0., 0.] ), “Wrong output. Use tf.one_hot”

1 Like

If that’s the assertion that is failing, notice that the array they are comparing against is actually a 1D array meaning that it has no orientation. It’s not a “row” or “column”, it’s just a vector. But if that compare fails, are you sure that there’s not something else wrong? E.g. the values or the number of values?

1 Like

Oh, sorry, I just tried this and that assertion does fail if you produce a 2D array. The problem is that the instructions are wrong. Eeeeek! You want a 1D output for this function.

I will file a bug right away to get the instructions fixed. Sorry that no-one had noticed this issue until now! I’m kind of shocked actually now that I see it.

2 Likes

Ok, the bug is filed. One other thing I noticed in the process of doing that is that they actually gave you the code in the comments to produce the correct 1D shape:

    # (approx. 1 line)
    # one_hot = None(None(None, None, None), shape=[C, ]) 
    # YOUR CODE STARTS HERE

But it’s totally legitimate to point out that the instructions are just wrong and everything should be consistent here.

Thank you. I did see the instruction, but then I thought that the instruction had a typo :slight_smile:

1 Like

Fair enough! :grinning: Actually they said it incorrectly in two places: both the instruction text before the function cell and in the “docstring” comment in the function body they said it should be a column vector. So two out of three is a majority in any case.

The bug is filed. Since it’s only text stuff, they usually fix that sort of thing pretty quickly.

Thank you