How to modify numpy arrays using multiple masks-CodePudding

Let's say I have an array a:

a = np.array([1,2,3,4,5])

I can make a mask:

mask1 = a > 2

and then view the masked array:

[In]: a[mask1]

[Out]: array([3, 4, 5])

or modify it:

a[mask1] = [8, 9, 10]

[In]: a
[Out]: array([ 1,  2,  8,  9, 10])

Now I want to mask the masked array again, based on some other criteria, e.g.:

mask2 = a[mask1] > 8

and I can view it:

[In]: a[mask1][mask2]
[Out]: array([ 9, 10])

Now here is the problem; when I try to modify the doubly-masked array, it doesn't work anymore:

a[mask1][mask2] = [20, 30]

[In]: a
[Out]: array([ 1,  2,  8,  9, 10])

I know it has to do with numpy returning views of arrays and so, but why does it work with one mask, but not with multiple masks, and how can I make it work with multiple masks?

CodePudding user response：

Update array with chained index should generally be avoided.

You can convert the masks into range index and update, something like this:

idx = np.arange(len(a))[mask1][mask2]
a[idx] = [20, 30]

Output:

array([ 1,  2,  8, 20, 30])

CodePudding user response：

Ok I found 'a' solution, but don't know if it's the best solution:

I create a current_mask array with the shape of my array, and initialize it to all True:

a = np.array([1,2,3,4,5])
current_mask = np.ones(a.shape, dtype=np.bool_)

Then, whenever I create a new mask, I update my current_mask with it:

mask1 = a > 2
current_mask[current_mask] &= mask1

Like before, I can view and modify the elements using current_mask:

[In]: a[current_mask]
[Out]: array([3, 4, 5])

[In]: a[current_mask] = [8,9,10]
[In]: a
[Out]: array([ 1,  2,  8,  9, 10])

Now, I can create a new mask using the current mask, and update the current mask with it:

mask2 = a[current_mask] > 8
current_mask[current_mask] &= mask2

Now, I can both view AND modify the elements:

[In]: a[current_mask]
[Out]: array([ 9, 10])

[In]: a[current_mask] = [20, 30]
[In]: a
[Out]: array([ 1,  2,  8, 20, 30])