CODEWITHSUNDEEP
pythonnumpy
handloomweaver
handloomweaver

Reputation: 5011

How to numpy searchsorted bisect a range of 2 values on 1 column and get min value in 2nd column

So I have a 2 column numpy array of integers, say:

tarray = array([[ 368,  322],
       [ 433,  420],
       [ 451,  412],
       [ 480,  440],
       [ 517,  475],
       [ 541,  503],
       [ 578,  537],
       [ 607,  567],
       [ 637,  599],
       [ 666,  628],
       [ 696,  660],
       [ 726,  687],
       [ 756,  717],
       [ 785,  747],
       [ 815,  779],
       [ 845,  807],
       [ 874,  837],
       [ 905,  867],
       [ 934,  898],
       [ 969,  928],
       [ 994,  957],
       [1027,  987],
       [1057, 1017],
       [1086, 1047],
       [1117, 1079],
       [1148, 1109],
       [1177, 1137],
       [1213, 1167],
       [1237, 1197],
       [1273, 1227],
       [1299, 1261],
       [1333, 1287],
       [1357, 1317],
       [1393, 1347],
       [1416, 1377]])

I am using np.searchsorted to bisect lower and upper ranges of values into column 0 i.e can both times e.g 241,361 bisect into the array.

ranges = [array([241, 290, 350, 420, 540, 660, 780, 900]),
 array([ 361,  410,  470,  540,  660,  780,  900, 1020])]

e.g: np.searchsorted(tarray[:,0], ranges)

This then results in:

array([[ 0,  0,  0,  1,  5,  9, 13, 17],
       [ 0,  1,  3,  5,  9, 13, 17, 21]])

where each position in the two resulting arrays is the range of values. What I then want to do is get the position of minimum value in column 1 of the resulting slice. e.g here is what I mean simply in Python via iteration (if result of searchsorted is 2 column array 'f'):

f = array([[ 0,  0,  0,  1,  5,  9, 13, 17],
       [ 0,  1,  3,  5,  9, 13, 17, 21]])

for i,(x,y) in enumerate(zip(*f)):
    if y - x:
        print ranges[1][i], tarray[x:y]

the result is:

410 [[368 322]]
470 [[368 322]
 [433 420]
 [451 412]]
540 [[433 420]
 [451 412]
 [480 440]
 [517 475]]
660 [[541 503]
 [578 537]
 [607 567]
 [637 599]]
780 [[666 628]
 [696 660]
 [726 687]
 [756 717]]
900 [[785 747]
 [815 779]
 [845 807]
 [874 837]]
1020 [[905 867]
 [934 898]
 [969 928]
 [994 957]]

Now to explain what I want: within the sliced ranges I want the row that has the minimum value in column 1.

e.g 540 [[433 420]
 [451 412]
 [480 440]
 [517 475]]

I want the final result to be 412 (as in [451 412])

e.g 

for i,(x,y) in enumerate(zip(*f)):
    if y - x:
        print ranges[1][i], tarray[:,1:2][x:y].min()

410 322
470 322
540 412
660 503
780 628
900 747
1020 867

Basically I want to vectorise this so I can get back one array and not need to iterate as it is non performant for my needs. I want the minimum value in column 1 for a bisected range of values on column 0.

I hope I am being clear!

Upvotes: 3

Views: 649

Answers (1)

Eelco Hoogendoorn
Eelco Hoogendoorn

Reputation: 10759

This appears to achieve your intended goals, using the numpy_indexed package (disclaimer: I am its author):

import numpy_indexed as npi
# to vectorize the concatenation of the slice ranges, we construct all indices implied in the slicing
counts = f[1] - f[0]
idx = np.ones(counts.sum(), dtype=np.int)
idx[np.cumsum(counts)[:-1]] -= counts[:-1]
tidx = np.cumsum(idx) - 1 + np.repeat(f[0], counts)

# combined with a unique label tagging the output of each slice range, this allows us to use grouping to find the minimum in each group
label = np.repeat(np.arange(len(f.T)), counts)
subtarray = tarray[tidx]
ridx, sidx = npi.group_by(label).argmin(subtarray[:, 0])

print(ranges[1][ridx])
print(subtarray[sidx, 1])

Upvotes: 1

Related Questions