Replace values greater/lower than x only in some rows on numpy array-CodePudding

Let's say I have a numpy array:

arr = np.random.rand(10, 3)    
array([[0.51619267, 0.69536865, 0.87577509],
       [0.82922802, 0.19907414, 0.97310683],
       [0.0961705 , 0.16849081, 0.9426728 ],
       [0.84950943, 0.7767404 , 0.38282767],
       [0.1204213 , 0.67595169, 0.46029065],
       [0.98476311, 0.59316099, 0.0877238 ],
       [0.31916443, 0.5374729 , 0.18507312],
       [0.99078367, 0.77783068, 0.05689834],
       [0.03063616, 0.9887299 , 0.41034183],
       [0.43509505, 0.11150762, 0.27512664]])

In this array I want to replace values lower than 0.1 with some other value (let's say 99), but only for those rows that contain the value larger than 0.98. I can find those rows like this:

selected_rows = np.any(arr > 0.98, axis=1)
array([False, False, False, False, False,  True, False,  True,  True,
       False])

But I'm not sure how to combine both conditions. I tried the following:

arr[selected_rows][arr[selected_rows] < 0.1] = 99

but it doesn't have any effect on original arr. Any ideas?

CodePudding user response：

You can use broadcasting with vectorial AND to combine the row selection and the 2D selection of arr<0.1:

arr[(arr<0.1) & selected_rows[:,None]] = 99

output (using NaN instead of 99 for clarity):

array([[0.51619267, 0.69536865, 0.87577509],
       [0.82922802, 0.19907414, 0.97310683],
       [0.0961705 , 0.16849081, 0.9426728 ],  # 0.096 remains as no value > 0.98
       [0.84950943, 0.7767404 , 0.38282767],
       [0.1204213 , 0.67595169, 0.46029065],
       [0.98476311, 0.59316099,        nan],
       [0.31916443, 0.5374729 , 0.18507312],
       [0.99078367, 0.77783068,        nan],
       [       nan, 0.9887299 , 0.41034183],
       [0.43509505, 0.11150762, 0.27512664]])

CodePudding user response：

My answer is not succinct as the one from mozway (which has a nice approach BTW), but works well on your problem. Here's the code:

def replace(a, val):
    if np.any(a > 0.98):
        indices = np.where(a < 0.1)[0]
        if indices.size:
            np.put_along_axis(a, indices, val, axis=None)
    return a


arr = [...]
np.apply_along_axis(replace, 1, arr, np.nan)

The above code makes use of numpy.apply_along_axis to apply given function over specified axis (1 in this case). We are sending np.nan (as opposed to 99 just for clarity in the output) as a value to given function. Consult documentation to know more about sending args and kwargs to a given function.

The replace function is self-explanatory. It uses numpy.put_along_axis which takes an array, the indices to put values against, the value itself, and the axis.

Output:

array([[0.51619267, 0.69536865, 0.87577509],
       [0.82922802, 0.19907414, 0.97310683],
       [0.0961705 , 0.16849081, 0.9426728 ],
       [0.84950943, 0.7767404 , 0.38282767],
       [0.1204213 , 0.67595169, 0.46029065],
       [0.98476311, 0.59316099,        nan],
       [0.31916443, 0.5374729 , 0.18507312],
       [0.99078367, 0.77783068,        nan],
       [       nan, 0.9887299 , 0.41034183],
       [0.43509505, 0.11150762, 0.27512664]])