问题
I have a very simple function, as shown below
def new_price(A, B, x):
return np.linalg.inv(A @ B) @ x
These are the inputs I give it
A = np.array([
[2, 0, 1, 0],
[1, 1, 1, 1],
[0, 0, 0, 10]
])
B = np.array([
[3, 3, 3],
[2, 0, 8],
[0, 5, 3],
[0, 0, 10]
])
x = np.array([ 84, 149, 500])
This returns the array [ 1. 3. 5.]
. But, when I make the following equality check, it returns False
v1 = new_price(A, B, x)
v2 = np.array([1.0, 3.0, 5.0])
np.array_equal(new_price(A, B, [ 84, 149, 500]), np.array([1.0, 3.0, 5.0]))
I checked and the shapes and the types of both the arrays are same. What am I missing here?
回答1:
The are not exactly equal:
>>> new_price(A, B, [ 84, 149, 500]) - np.array([1, 3, 5])
array([ 2.84217094e-14, -1.42108547e-14, 0.00000000e+00])
Better use np.allclose()
:
>>> np.allclose(new_price(A, B, [ 84, 149, 500]), np.array([1.0, 3.0, 5.0]))
True
Returns True if two arrays are element-wise equal within a tolerance.
You can adjust the relative and absolute tolerance.
Still true for really small values:
>>> np.allclose(new_price(A, B, [ 84, 149, 500]), np.array([1.0, 3.0, 5.0]),
atol=1e-13, rtol=1e-14)
True
Found the limit:
>>> np.allclose(new_price(A, B, [ 84, 149, 500]), np.array([1.0, 3.0, 5.0]),
atol=1e-14, rtol=1e-14)
False
来源:https://stackoverflow.com/questions/48527910/numpy-array-equal-returns-false-even-though-arrays-have-the-same-shape-and-valu