Setting a new value to an array gives different results than it should

Question

I have this array:

L = np.array([[5,4],[3,2]])

And I want to change the second bracket.

If I do:

print(L[1] + (-0.6)*L[0])

it will give me this array:

[0,-0.4]

But, if I change it like this:

L[1] = L[1] + (-0.6)*L[0]
print(L)

it prints:

# [[5 4] [0 0]]

Why are the outputs different?

• m0nhawk: What part of L you want to overwrite? Second row? What is the desired output?

Answer 1

The first example:

L[1] + (-0.6)*L[0]

returns the result of the specified operation, but it isn't assigned so you just have the result.

In your second example

L[1] = L[1] + (-0.6)*L[0]

the result is written to L[1], however the result is cast to the dtype of L. Which is an integer. So the intermediate result is simply truncated.

This "truncation" is the reason why the result isn't "correct":

>>> (L[1] + (-0.6)*L[0]).astype(int) # simulating the truncation
array([0, 0])

• wim: It is interesting that the augmented assignment statement L[1] += -0.6*L[0] will fail because of a casting rule, in this case, but a regular assignment will not fail.
• MSeifert: @wim Maybe they just forgot to add it to = as well or they just feel like that's explicit "casting" and therefore allowed while [*]= is implicit and therefore not allowed. Or it's because = doesn't use the ufunc-machinery (which allows to specifiy casting=) and therefore was simply "left" as is. But that's just guessing, maybe worth to ask about it on numpys issue tracker.
• wim: It's actually documented behaviour. See the warning mentioned here. Nonetheless, I find the discrepancy surprising (and questionable design choice).