Unique function not deleting duplicate rows.

Asked by luc on 4 May 2015

Latest activity Answered by Robert on 17 Oct 2018

46 views (last 30 days)

M.mat

attached my matrix "M" and here is my code.

[trash,idx] = unique(M,'rows');
pleb=M(idx,:)
gg=sort(pleb)

When inspecting gg we see that there are still duplicate rows.

I've also tried to do it in different ways, for example;

[~, III, ~] = unique(M,'first','rows'); %removing double points
III = sort(III);
pleb = M(III,:);
gg=sort(pleb);

But they either delete non duplicate data, or delete too few data.

What am I doing wrong?

2 Comments

Stephen Cobeldick on 4 May 2015

"What am I doing wrong": not clicking on both buttons to attach the data: you need to first click Choose file and then Attach file. Please try attaching your data again.

luc on 4 May 2015

I attached it the right way now.

4 Answers

Link

Answer by Stephen Cobeldick on 4 May 2015

Edited by Stephen Cobeldick on 4 May 2015

Accepted Answer

It is likely that the data are floating point and that they are not actually equal, which confuses many beginners and people not used to working with numeric data. Although what is displayed on the command window might look the same, floating point values can differ at the low end of their significand, so testing for equality (like unique does) does not work.

To understand more about this topic read these:

http://matlab.wikia.com/wiki/FAQ#Why_is_0.3_-_0.2_-_0.1_.28or_similar.29_not_equal_to_zero.3F

http://www.mathworks.com/matlabcentral/answers/41536-why-is-2-24-10-22-4-not-equal-to-0

http://www.mathworks.com/matlabcentral/answers/69-why-does-1-2-3-1-3-not-equal-zero

http://www.mathworks.com/matlabcentral/answers/57444-faq-why-is-0-3-0-2-0-1-not-equal-to-zero

Alternatively, if the data are strings, trailing spaces are often overlooked by users...

8 Comments

Show 5 older comments

Stephen Cobeldick on 4 May 2015

@luc: Your comment makes it clear that there is significant confusion about what unique and sort actually do. Reading the documentation would help you.

The function unique (when used with the option rows) removes rows that are identical. Have a look at this simple example:

>> A = [0,1,2; 1,1,1; 0,1,2; 3,3,3]
A =
     0     1     2
     1     1     1
     0     1     2
     3     3     3
>> unique(A,'rows')
ans =
     0     1     2
     1     1     1
     3     3     3

This is the correct and documented behavior of unique: it has removed the repeated instances of the row 0,1,2 because this row occurs twice in the matrix A, but the two rows which consist of repeated values are still in the output. When we run unique(M) using your matrix M we get an output matrix that is the same size as M, because none of the rows are duplicated.

The function sort, when given matrix M, sorts each column of M independently. Like this simple example shows, using the matrix used above:

>> sort(A)
ans =
     0     1     1
     0     1     2
     1     1     2
     3     3     3

Note that the rows are not the same after being sorted: each column has been sorted independently, so there is no guarantee that the output rows are the same as the input rows (and in your case they are not).

Here are a few things we have learned:

All of the rows of the matrix M are already unique, so unique(M) does nothing.
sort(M) produces rows that are not the same as the rows beforehand, so there is not reason why they should be unique either.

You might like to try using sortrows instead of sort...

>> sortrows(A)
ans =
   0     1     2
   0     1     2
   1     1     1
   3     3     3

gives quite a different output to sort, but it is still clear that none of the rows are duplicates.

And it is also easy to show that none of the rows is made up of repeated values, just in case this was what you were hoping to locate:

>> min(max(abs(diff(M,2)),[],2))
ans =
  0.5343

If any single row had only the same value repeated then this result would be zero.

To exactly define the operation you are trying to achieve you need to tell us what this is and describe the output that you need, and even give some examples...

Stephen Cobeldick on 4 May 2015

@John D'Errico: Oh... but then what about:

1 == sqrt((+1)^2) == sqrt((-1)^2) == -1

I don't know that song, but I was humming the terms of Grandi's series in my head... still didn't reach 1/2 though... have to add a few more... will get there soon...

luc on 4 May 2015

Thanks Stephen,

I think the sorting part had me confused.

nice explanation, I learned something.

Stephen Cobeldick on 5 May 2015

@luc: I'm glad to be able to help!

Link

Answer by Titus Edelhofer on 4 May 2015

Hi Luc,

I don't see duplicate data, but the data change sign ...? Take last 4 rows of pleb and it's

4558   -4.1355   -2.0906
4558   -4.1355    2.0906
4558    4.1355   -2.0906
4558    4.1355    2.0906

Look similar but all 4 are completely different - as long as -2.0906 is different from 2.0906 ;-).

Similar for the other "4-row-blocks".

When you take the abs then the story is different,

Titus

3 Comments

luc on 4 May 2015

Hey Titus,

I noticed that too, but that is because the points are coordinates spanning a sphere so some "mirror points" are to be expected.

If you use this code:

    [~, III, ~] = unique(M,'first','rows'); %removing double points
    III = sort(III);
    M = M(III,:);
    FF=sort(M)
    FF(4,:)==FF(3,:)

U will see that even the rows that are equal are not removed.

Stephen Cobeldick on 4 May 2015

@luc: There is no reason why those rows would be removed, as

all rows of M are already unique
sort(M) sorts each column independently, so there is no reason why these rows should be unique (or removed) either.

You need to actually describe what you are trying to achieve.

Titus Edelhofer on 4 May 2015

Indeed. As I wrote as comment, if you would sort keeping rows as rows, i.e., using

sortrows(M)

then you would see, that there are no duplicate rows.

Link

Answer by John D'Errico on 4 May 2015

Edited by John D'Errico on 4 May 2015

There are NO equal rows. I checked. They are different in sign. There are no rows that are even that close to each other, although the nearest neighbor is not uniformly close.

The check that I made was to find the point for each row that was closest in distance. I.e., the nearest neighbor. There ARE no essentially zero distances.

The overall closest pair of points are 1.7291 units apart.

Mu = unique(M,'rows');
D = ipdm(Mu,'subset','smallestfew','limit',1)
D =
  (87,95)         1.7291
D = ipdm(Mu,'subset','nearest')
D =
   (2,1)          4.1811
   (1,2)          4.1811
  (13,3)          4.1811
  (14,4)          4.1811
   (6,5)          4.1811
   (5,6)          4.1811
   (8,7)          4.1811
   (7,8)          4.1811
  (15,9)          4.1811
  (16,10)         4.1811
  (17,11)         4.1811
  (18,12)         4.1811
   (3,13)         4.1811
   (4,14)         4.1811
   (9,15)         4.1811
  (10,16)         4.1811
  (11,17)         4.1811
  (12,18)         4.1811
  (26,25)         4.1811
  (25,26)         4.1811
  (28,27)         4.1811
  (27,28)         4.1811
  (35,29)         4.1811
  (36,30)         4.1811
  (37,31)         4.1811
  (38,32)         4.1811
  (19,33)         4.1811
  (20,34)         4.1811
  (29,35)         4.1811
  (30,36)         4.1811
  (31,37)         4.1811
  (32,38)         4.1811
  (21,39)         4.1811
  (22,40)         4.1811
  (23,41)         4.1811
  (24,42)         4.1811
  (33,47)         4.1811
  (53,47)         3.3826
  (55,47)         3.3826
  (34,48)         4.1811
  (54,48)         3.3826
  (56,48)         3.3826
  (43,49)         4.1811
  (44,50)         4.1811
  (45,51)         4.1811
  (46,52)         4.1811
  (39,53)         4.1811
  (47,53)         3.3826
  (40,54)         4.1811
  (48,54)         3.3826
  (41,55)         4.1811
  (42,56)         4.1811
  (58,57)         4.1811
  (57,58)         4.1811
  (60,59)         4.1811
  (59,60)         4.1811
  (69,61)         4.1811
  (70,62)         4.1811
  (71,63)         4.1811
  (72,64)         4.1811
  (49,65)         4.1811
  (91,65)         3.3826
  (50,66)         4.1811
  (92,66)         3.3826
  (51,67)         4.1811
  (93,67)         3.3826
  (52,68)         4.1811
  (94,68)         3.3826
  (61,69)         4.1811
  (62,70)         4.1811
  (63,71)         4.1811
  (64,72)         4.1811
  (79,77)         1.7291
  (81,77)         1.7291
  (80,78)         1.7291
  (82,78)         1.7291
  (77,79)         1.7291
  (78,80)         1.7291
  (73,83)         4.1811
  (74,84)         4.1811
  (75,85)         4.1811
  (76,86)         4.1811
  (95,87)         1.7291
  (96,88)         1.7291
  (97,89)         1.7291
  (98,90)         1.7291
  (65,91)         3.3826
  (83,91)         4.1811
 (105,91)         3.3826
  (66,92)         3.3826
  (84,92)         4.1811
 (106,92)         3.3826
  (67,93)         3.3826
  (85,93)         4.1811
 (107,93)         3.3826
  (68,94)         3.3826
  (86,94)         4.1811
 (108,94)         3.3826
  (87,95)         1.7291
 (101,95)         1.7291
  (88,96)         1.7291
 (102,96)         1.7291
  (89,97)         1.7291
 (103,97)         1.7291
  (90,98)         1.7291
 (104,98)         1.7291
  (99,101)        4.1811
 (100,102)        4.1811
 (109,105)        4.1811
 (110,106)        4.1811
 (111,107)        4.1811
 (112,108)        4.1811
 (113,109)        4.1811
 (114,110)        4.1811
 (115,111)        4.1811
 (116,112)        4.1811
 (121,117)        4.1811
 (122,118)        4.1811
 (123,119)        4.1811
 (124,120)        4.1811
 (117,121)        4.1811
 (118,122)        4.1811
 (119,123)        4.1811
 (120,124)        4.1811
 (126,125)        4.1811
 (125,126)        4.1811
 (133,127)        1.7291
 (134,128)        1.7291
 (135,129)        1.7291
 (136,130)        1.7291
 (132,131)        4.1811
 (131,132)        4.1811
 (127,133)        1.7291
 (128,134)        1.7291
 (129,135)        1.7291
 (130,136)        1.7291
 (139,137)        1.7291
 (140,138)        1.7291
 (137,139)        1.7291
 (138,140)        1.7291
 (145,141)        4.1811
 (149,141)        3.3826
 (146,142)        4.1811
 (150,142)        3.3826
 (147,143)        4.1811
 (151,143)        3.3826
 (148,144)        4.1811
 (152,144)        3.3826
 (153,145)        4.1811
 (154,146)        4.1811
 (155,147)        4.1811
 (156,148)        4.1811
 (141,149)        3.3826
 (165,149)        4.1811
 (142,150)        3.3826
 (166,150)        4.1811
 (143,151)        3.3826
 (167,151)        4.1811
 (144,152)        3.3826
 (168,152)        4.1811
 (159,157)        3.3826
 (177,157)        4.1811
 (160,158)        3.3826
 (178,158)        4.1811
 (157,159)        3.3826
 (179,159)        4.1811
 (158,160)        3.3826
 (180,160)        4.1811
 (169,161)        4.1811
 (170,162)        4.1811
 (171,163)        4.1811
 (172,164)        4.1811
 (181,165)        4.1811
 (182,166)        4.1811
 (183,167)        4.1811
 (184,168)        4.1811
 (161,169)        4.1811
 (162,170)        4.1811
 (163,171)        4.1811
 (164,172)        4.1811
 (174,173)        4.1811
 (173,174)        4.1811
 (176,175)        4.1811
 (175,176)        4.1811
 (189,177)        4.1811
 (190,178)        4.1811
 (191,179)        4.1811
 (192,180)        4.1811
 (193,185)        4.1811
 (194,186)        4.1811
 (195,187)        4.1811
 (196,188)        4.1811
 (185,193)        4.1811
 (186,194)        4.1811
 (187,195)        4.1811
 (188,196)        4.1811
 (198,197)        4.1811
 (197,198)        4.1811
 (200,199)        4.1811
 (199,200)        4.1811
 (205,201)        4.1811
 (206,202)        4.1811
 (207,203)        4.1811
 (208,204)        4.1811
 (201,205)        4.1811
 (202,206)        4.1811
 (203,207)        4.1811
 (204,208)        4.1811
 (210,209)        4.1811
 (209,210)        4.1811
 (212,211)        4.1811
 (211,212)        4.1811

6 Comments

Show 3 older comments

luc on 4 May 2015

Thanks for the effort guys, but why am I seeing these double vallues then in my data?

Here's a screenshot as "proof".

http://imgur.com/hD3EcPR

Sean de Wolski on 4 May 2015

First, your screenshot is too small to see.

Second, here's a good exercise to explain the small differences in floating point: Run this:

>> format hex

Then rerun the command. See! They're different, even if just by a little.

luc on 4 May 2015

Hey Sean,

U can click on the screenshot to enlarge it.

But I think Stephen solved my problem. The sort functions grabs each colums independant, and not as a whole.

Thanks guys!

Link

Answer by Robert on 17 Oct 2018

If anyone encounters truly duplicate rows in the output of unique like I did, this may be caused by NaN in your data being treated as distinct values. See this question for more info.

You are now following this question