Efficient way for mapping array values

Question

I have numpy array, and I do want to map the ranges of the array values into specific values:

import numpy as np

array = [
[328, 124, 146, 117, 147, 109, 139, 100, 140, 98, 128, 101, 121, 95, 111, 89, 114, 82, 106, 77, 97, 71, 273], 
[132, 121, 121, 117, 114, 110, 99, 163, 89, 81, 83, 81, 91, 91, 89, 144, 89, 81, 81, 75, 69, 68, 68], 
[164, 114, 117, 110, 105, 105, 97, 91, 142, 80, 131, 77, 122, 81, 108, 74, 77, 75, 69, 71, 68, 63, 87], 
[121, 116, 110, 103, 104, 99, 90, 84, 77, 137, 77, 125, 99, 128, 79, 73, 69, 71, 68, 70, 60, 53, 64], 
[151, 100, 103, 99, 379, 88, 264, 78, 135, 77, 121, 97, 113, 75, 100, 62, 232, 65, 350, 54, 56, 50, 75], 
[101, 93, 121, 90, 84, 92, 86, 78, 74, 113, 66, 115, 72, 100, 63, 67, 60, 57, 57, 47, 47, 44, 43], 
[146, 89, 116, 86, 87, 85, 79, 73, 101, 59, 98, 63, 107, 64, 62, 56, 56, 52, 48, 50, 46, 38, 68], 
[89, 113, 108, 83, 82, 78, 74, 125, 69, 67, 57, 65, 67, 56, 58, 105, 49, 52, 69, 59, 63, 34, 43], 
[127, 87, 78, 79, 105, 74, 163, 64, 61, 61, 56, 235, 61, 48, 51, 43, 134, 40, 63, 28, 31, 26, 55], 
[76, 79, 75, 70, 72, 69, 62, 58, 53, 52, 49, 50, 53, 43, 39, 37, 41, 36, 31, 29, 26, 18, 17], 
[274, 73, 96, 71, 92, 64, 87, 57, 82, 53, 81, 30, 73, 41, 66, 37, 52, 31, 49, 22, 50, 15, 211]]

array = np.array(array)

def mapVal(val, table = [140, 195, 250, 350, 450]):
    """
    Function: mapVal, to quantify depth values manually.
    ---
    Parameters:
    @param: val, nd-array.
    @param: table, list, mapping table.

    ---
    @return: None.
    """
    
    val[val <= table[0]]  = 0
    val[val >  table[0] & val <=  table[1]]  = 1
    val[val >  table[1] & val <=  table[2]]  = 2
    val[val >  table[2] & val <=  table[3]]  = 3
    val[val >  table[3] & val <=  table[4]]  = 4
    val[val >  table[4]]  = 5

But I'm getting this Error:

ValueError: The truth value of an array with more than one element is ambiguous. Use a.any() or a.all()

I'm awar that I can use inefficient loops, but I do like to ask if there is a better solution, thanks in advance.

You need to add parenthesis - e.g. val[(val > table[0]) & (val <= table[1])], since the & is a bitwise operation and takes priority by default. This should address your ValueError — rcshon
– rcshon, Commented Apr 20, 2022 at 12:16
Does this fully answer your question? or are you still in search for a "more efficient" way? — rcshon
– rcshon, Commented Apr 20, 2022 at 12:18
You method is already decent. Perhaps use a for-loop for conciseness. If you really must find a more efficient way, you can use np.digitize(array, table) in your case. — rcshon
– rcshon, Commented Apr 20, 2022 at 12:37

rcshon · Accepted Answer · 2022-04-20 12:40:36Z

3

As mentioned in the comments, first solve your ValueError by using parenthesis around your conditional checks to avoid having the bit-wise and operator & be evaluated first, which is the default behaviour.

For example:

val[(val > table[0]) & (val <= table[1])]

Next, for your search for a more efficient method, you can use the built-in np.digitize function to help you do what you're doing:

np.digitize(array, table)

answered Apr 20, 2022 at 12:40

rcshon

9271 gold badge7 silver badges12 bronze badges

Sign up to request clarification or add additional context in comments.

4 Comments

Bilal Over a year ago

np.digitize(array, table) doesn't yield the same output of my manual function mapVal.

rcshon Over a year ago

they should be the same. perhaps ensure that you are passing a new instance of the array with the original values for verification since your function modifies the input array.

Jérôme Richard Over a year ago

For np.digitize to work correctly, the table should contain all the values and not just some of them. Otherwise some values are not replaced. Thus, the map should be corrected first. On solution is to use np.searchsorted to generate it. Boundaries also still need to be checked. This approach should be pretty fast.

rcshon Over a year ago

You are right. that's oversight on my part

Jérôme Richard · Accepted Answer · 2022-04-20 13:16:49Z

2

There is a much simpler and much faster way to do what you want:

table = np.array([140, 195, 250, 350, 450])
result = np.searchsorted(table, array)

This assume that the table is sorted though which is the case here. If it is not, then you can use np.argsort to reorder the table and np.digitize to remap the values.

answered Apr 20, 2022 at 13:16

Jérôme Richard

53.3k6 gold badges48 silver badges77 bronze badges

3 Comments

Bilal Over a year ago

thanks for your kind suggestion, I have compared the output of of my manual function mapVal, and it looks different from the output of your solution, the objective of mapVal function is to generate a discrete output with values [1, 2, 3, 4, 5] that corresponds to pre-defined ranges.

Jérôme Richard Over a year ago

This is surprising. I checked the results and it was fine on my machine with your example. Note that your function mutate the input array so you certainly need to reload the values between the use of your function and the proposed code. Alternatively, you can .copy() the input array so to be safe.

Bilal Over a year ago

you are right, thanks for your kind explanation

Collectives™ on Stack Overflow

Efficient way for mapping array values

2 Answers 2

4 Comments

3 Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

2 Answers 2

4 Comments

3 Comments

Your Answer

Sign up or log in

Post as a guest

Related