numpy.where() clause error

Question

Briefly... here is the problem:

import numpy as np
a = np.array([ 0, 1, 2, 3, 4, 5, 6, 100, 8, 9])
np.where(a==100, -1, a[a])

What I expect to get is: 0, 1, 2, 3, 4, 5, 6, -1, 8, 9 Instead I'm getting: index 100 out of bounds 0<=index<10

I admit that the index is invalid but is shouldn't eval a[100] but -1 instead... as far as I understand numpy.where() command structure.

What I'm doing wrong in this example?

Just to clarify what I actually trying to do here is more detailed code: It is a lookup table array remapping procedure:

import numpy as np

# gamma-ed look-up table array
lut = np.power(np.linspace(0, 1, 32), 1/2.44)*255.0

def gamma(x):
    ln = (len(lut)-1)
    idx = np.uint8(x*ln)
    frac = x*ln - idx
    return np.where( frac == 0.0,
                    lut[idx],
                    lut[idx]+(lut[idx+1]-lut[idx])*frac)

# some linear values array to remap
lin = np.linspace(0, 1, 64)

# final look-up remap
gamma_lin = gamma(lin)

Answer 1

Expressions that you put as function arguments are evaluated before they are passed to the function (Documentation link). Thus you are getting an index error from the expression a[a] even before np.where is called.

Answer 2

Use the following:

np.where(a==100, -1, a)

As stated by the documentation:

numpy.where(condition[, x, y])

Return elements, either from x or y, depending on condition.
If only condition is given, return condition.nonzero().

Here, a==100 is your condition, -1 the value that should be taken when the condition is met (True), a the values to fall back on.

---

The reason why you're getting an IndexError is your a[a]: you're indexing the array a by itself, which is then equivalent to a[[0,1,2,3,4,5,6,100,8,9]]: that fails because a has less than 100 elements...

---

Another approach is:

a_copy = a.copy()
a_copy[a==100] = -1

(replace a_copy by a if you want to change it in place)

Answer 3

When you write a[a] you try to take index 0,1,2...100... from a which is why you get the index out of bounds error. You should instead write np.where(a==100, -1, a) - I think that will produce the result you are looking for.