Another little fix for file /edit/release/W3A1/public_tests.py of Week 3

Course Q&A Deep Learning Specialization Neural Networks and Deep Learning
1

The function

nn_model_test()

in

/edit/release/W3A1/public_tests.py

builds an Y containing booleans, which may subvert expectations (Python having no kind of contracts anywhere, forcibly nonstated expectations :sweat_smile:).

Let’s apply the first part of “be conservative of what you send, liberal in what you accept” :sunglasses:

def nn_model_test(target):
    np.random.seed(1)
    X = np.random.randn(2, 4)
    Y = (np.random.randn(1, 4) > 0)  # return array of numeric, not of boolean
    n_h = 5
    ...
1 Like
2

Hello, David!

That is a good point, but, again, unless it is critical, which I think you agreed it is “little” instead, it might not be addressed any time soon. I am sure you understand this :wink:

Indeed, Python is a bit flexible in terms of data type. It brings, in some sense, convenience but it also requires, e.g. library developers, to handle such ambitguity or it just throws an error.

Have you heard of “Type Hint”? It’s becoming popular to specify an expected type for each variable in Python, though the expectations are not enforced - it is like part of the documentation effort. Below is an example:

image
image774×535 34.8 KB

Besides the three inputs - x, y, z, the type of the return is also specified (as None | str).

Having said it is getting popular, it is not like universal. For example, Pandas is adopting it, but Numpy, to my knowledge, is not quite.

You might consider this for your own project :wink:

Cheers,
Raymond

1 Like
3

Thanks Raymond!

I definitely am using the type hints now. They are free checks for the write-time type checker (which sometimes can’t do much, as soon as type Any is encountered, all bets are off) and free documentation for the next person who reads the code.

I deeply mistrust code that doesn’t have them, it’s like driving a car with the windshield covered. Missing type enforcement / missing compile-time checks / dynamic typing was acceptable when programs were small, theory and practice of type checking was nonexistent or esoteric and computers were weak (can’t run the prover). No longer!

If I have to get up at 02:00 to take a look at a runtime error, it better be interesting, not due to an “unexpected type”.

In fact, the next level appears:

https://www.idris-lang.org/

1 Like
4

But type hints are only getting popular, so I am sorry to tell you that you may not see them in many deep learning repositories that will be useful for your future adventure. :rofl: :rofl: :rofl:

1 Like