Sorting all rows in numpy matrix by target-column

Question

This seems been asked many times, however the answer I found not work now. Let's be simple, here I have a numpy matrix

data = np.matrix([[9, 8],
             [7, 6],
             [5, 7],
             [3, 2],
             [1, 0]])

Then sort by second column as below

[[1, 0],
[3, 2],
[7, 6],
[5, 7],
[9, 8]])

I tried a lot examples like Python Matrix sorting via one column but none of them worked.

I wondering maybe because the answers were posted years ago which do not work for newest Python? My Python is 3.5.1.

Example of my failed trial:

data = np.matrix([[9, 8],
             [7, 6],
             [5, 7],
             [3, 2],
             [1, 0]])
temp = data.view(np.ndarray)
np.lexsort((temp[:, 1], ))

print(temp)
print(data)

What does "not work" mean, which code exactly was tried and you know that np.matrix != np.array? (I really really really don't think, that these other answers are outdated if they are using basic numpy stuff; this would kill numpy for a lot of reasons!) — sascha
– sascha, Commented May 29, 2016 at 22:14
While taking another glance at your question: your title Sorting numpy matrix by column is misleading. The working example in your first link does that, but according to your example you only want to sort the column within the matrix (without rearranging the corresponding other columns too)! — sascha
– sascha, Commented May 29, 2016 at 22:20
@sascha You are right, so I have remove my second link. "Don't work " means the approved answer does not sort my matrix. i have put an example there. — Windtalker
– Windtalker, Commented May 29, 2016 at 22:24
Specify again what you really want: you want to sort all columns by one target-column (as the title says; permuting complete rows) or you want to sort only one column and don't touch the others ones, as your example shows. (the former has a working solution in your link) — sascha
– sascha, Commented May 29, 2016 at 22:38

hpaulj · Accepted Answer · 2016-05-29 22:58:32Z

4

You are a moving target.

Sort each column independently:

In [151]: np.sort(data,axis=0)
Out[151]: 
matrix([[1, 0],
        [3, 2],
        [5, 6],
        [7, 7],
        [9, 8]])

Sort on the values of the second column

In [160]: ind=np.argsort(data[:,1],axis=0)

In [161]: ind
Out[161]: 
matrix([[4],
        [3],
        [1],
        [2],
        [0]], dtype=int32)

In [162]: data[ind.ravel(),:]  # ravel needed because of matrix
Out[162]: 
matrix([[[1, 0],
         [3, 2],
         [7, 6],
         [5, 7],
         [9, 8]]])

Another way to get a valid ind array:

In [163]: ind=np.argsort(data.A[:,1],axis=0)

In [164]: ind
Out[164]: array([4, 3, 1, 2, 0], dtype=int32)

In [165]: data[ind,:]

To use lexsort you need something like

In [175]: np.lexsort([data.A[:,0],data.A[:,1]])
Out[175]: array([4, 3, 1, 2, 0], dtype=int32)

or your 'failed' case - which isn't a fail

In [178]: np.lexsort((data.A[:,1],))
Out[178]: array([4, 3, 1, 2, 0], dtype=int32)

here data[:,1] is the primary key. data[:,0] is the tie breaker (not applicable in your example). I'm just working from the docs.

edited May 29, 2016 at 22:58

answered May 29, 2016 at 22:47

hpaulj

233k14 gold badges260 silver badges392 bronze badges

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

1 Comment

Windtalker Over a year ago

Went through all your hints, thanks so much for the help!

sascha · Accepted Answer · 2016-05-29 22:51:28Z

1

The approach in your link is working:

import numpy as np

data = np.matrix([[9, 8],
         [7, 6],
         [5, 7],
         [3, 2],
         [1, 0]])

print(data[np.argsort(data.A[:, 1])])

[[1 0]
 [3 2]
 [7 6]
 [5 7]
 [9 8]]

And now an example where it's better to see:

data = np.matrix([[1, 9],
         [2, 8],
         [3, 7],
         [4, 6],
         [0, 5]])
[[0 5]
 [4 6]
 [3 7]
 [2 8]
 [1 9]]

edited May 29, 2016 at 22:51

answered May 29, 2016 at 22:45

sascha

33.7k6 gold badges80 silver badges117 bronze badges

3 Comments

Windtalker Over a year ago

Sascha, I do appreciate your help and patience. I made a lot mistakes in my initial post. Biggest mistake was I confused by matrix and ndarray. Anyway, thanks again. Solved my problem

sascha Over a year ago

No worries. Have fun with python and numpy! You also should accept @hpaulj's answer because it is really detailed and solves your problem! Have a look at StackOverflows help to see why you won't need any "SOLVED" in title. You just mark one answer as accepted.

Windtalker Over a year ago

Just flagged. Well if possible I would like to flag you both

Collectives™ on Stack Overflow

Sorting all rows in numpy matrix by target-column

2 Answers 2

1 Comment

3 Comments

Your Answer

Linked

Hot Network Questions

Collectives™ on Stack Overflow

2 Answers 2

1 Comment

3 Comments

Your Answer

Sign up or log in

Post as a guest

Linked

Related