Using Interp1 - monotonic increasing values

Hi

I am trying to interpolate physiological values for a study I have run - values for blood pressure, HR and saturations.

I have created a text file and trying to use a script I have used before to run the interpolation.

Text file is:

0  101  51  98  71
5  0  0  97  71
10  103  50  97  73
15  0  0  98  72
20  93  51  98  73
25  0  0  98  73
30  95  52  98  73
35  0  0  98  72
40  94  50  97  73
45  0  0  97  73
50  97  55  97  69
55  0  0  98  60
60  87  51  98  61
65  0  0  98  61
70  89  50  98  60
75  0  0  98  61
80  92  54  98  61
85  0  0  97  61
90  91  51  98  62
95  96  56  98  64
100  0  0  98  64
105  94  54  98  61
110  0  0  98  60
115  95  51  97  60
120  0  0  98  61
125  99  54  98  61
130  0  0  98  61
135  93  52  97  61
140  0  0  98  61
145  92  50  98  61
150  0  0  98  62
155  92  57  97  65
160  0  0  98  65
165  94  50  97  67
170  0  0  98  65
175  95  57  98  69
180  0  0  98  64
185  94  50  98  59
190  0  0  98  53
195  95  57  99  52
200  0  0  98  63
205  98  59  98  58
210  0  0  97  59
215  97  59  97  64
220  0  0  98  64
225  98  50  98  64
230  0  0  99  64
235  98  59  97  65
240  99  66  98  64

Script is:

close all
clear all

A = load(uigetfile()); 

time = A(:,1);
SBP = A(:,2);         
DBP = A(:,3);
sats = A(:,4);
hr = A(:,5);

x = SBP;
y = time;

start=min(time);
finish=max(time);

xi=(start:0.001:finish);
yi=interp1(x,y,xi,'linear');

But getting error:

Error using griddedInterpolant
The grid vectors are not strictly monotonic increasing.

Error in interp1 (line 191)
        F = griddedInterpolant(X,V,method);

Error in interpolate_lab_values (line 34)
yi=interp1(x,y,xi,'linear');

Have tried everything I can think of and find on the net but no luck

Thanks loads!

[Monty]

Sorry - text file hasnt come out - have attached

[nl2605]

Did you try changing the method?

[Monty]

Yup - same with spline/nearest

Answer 1

[John D'Errico]

I think you misunderstand interpolation.

Given your data, you tried to do this:

time = A(:,1); SBP = A(:,2); DBP = A(:,3); sats = A(:,4); hr = A(:,5);
x = SBP; y = time;
start=min(time); finish=max(time);

xi=(start:0.001:finish); yi=interp1(x,y,xi,'linear');

I assume that time is what it says, the time you took a sample at, and that SBP is a blood pressure at that time.

[time,SBP]
ans =
     0   101
     5     0
    10   103
    15     0
    20    93
    25     0
    30    95
    35     0
    40    94
    45     0
    50    97
    55     0
    60    87
    65     0
    70    89
    75     0
    80    92
    85     0
    90    91
    95    96
   100     0
   105    94
   110     0
   115    95
   120     0
   125    99
   130     0
   135    93
   140     0
   145    92
   150     0
   155    92
   160     0
   165    94
   170     0
   175    95
   180     0
   185    94
   190     0
   195    95
   200     0
   205    98
   210     0
   215    97
   220     0
   225    98
   230     0
   235    98
   240    99

Interpolation allows us to take a function of some variable x, and predict a variable y(x), given a set of points (x,y). The function MUST be single valued at any location x, else how can one unambiguously compute y(x)?

Perhaps you see the problem. You are apparently trying to treat SBP as the independent variable x, and time as the dependent variable x. This makes NO sense at all physically, since you appear to have sampled the blood pressure as a function of time. Here time is the independent variable.

Of course, interp1 gets upset, because you have MANY replicate values for SBP. I have no idea what the meaning of those zeros is, but surely it makes no sense to try to predict time as a function of SBP!

Perhaps your goal was to predict SBP(time). This makes sense, since you compute a start time and an end time for the interpolation. However, interp1 cannot know what you WANT to do, only what you TELL it to do! Computers are only capable of reading your mind in the sci-fi shows.

Even then, I think it makes no real sense to try to interpolate this as a function of time, unless you appreciate that the zeros are surely meaningless.

So finally, IF I had to guess, what you really wanted to do is to interpolate SBP(time) but without those spurious zeros in the data. The crystal ball (even MY crystal ball) is foggy though.

[Monty]

Thanks loads. Yup, that is what I was after. I put the zeros in as matlab was treating the gaps as missing data. Should I just remove them?

[Monty]

I have also switched x and y which makes total sense!

[John D'Errico]

Good. I was pretty sure that x and y were swapped as you were using them. If the zeros are just there as missing values, then drop those elements when you do the interpolation. interp1 will think they really are values, and your function is then a nasty saw-toothed one, going up and down from 0 to the real value. Anyway, I always thought that if my blood pressure ever went to zero, it was a cause for serious concern.

xi=(start:0.001:finish);
ind = (SBP ~= 0);
yi = interp1(time(ind),SBP(ind),xi,'linear');
plot(time(ind),SBP(ind),'ro',xi,yi,'g-')

As it is, even this curve is a somewhat noisy one. You may or may not wish to do some smoothing. My SLM toolbox (for download from the file exchange) may help you here.

spl = slmengine(time(ind),SBP(ind),'plot','on','knots',10);

[Monty]

Thanks John! Extremely helpful and worked great on that dataset. Have another dataset with same columns (and BP data only 10 minutes as before) and having same interp1 error message re not being monotonic increasing. Have run as you suggested above. Any ideas?

   xi=(start:0.001:finish);
   ind = (SBP ~= 0);
   yi = interp1(time(ind),SBP(ind),xi,'linear');
Error using griddedInterpolant
The grid vectors are not strictly monotonic increasing.

 Error in interp1 (line 191)
        F = griddedInterpolant(X,V,method);

[John D'Errico]

LOOK at your data. Read the error message. Think about what it is telling you. Don't just give up.

time'
ans =
  Columns 1 through 22
     0     5    10    15    20    25    30    35    40    45    50    55    60    65    70    75    95    85    90    95   100   105
  Columns 23 through 44
   110   115   120   125   130   135   140   145   150   155   160   165   170   175   180   185   190   195   200   205   210   215
  Columns 45 through 49
   220   225   230   235   240

>> diff(time')
ans =
  Columns 1 through 22
     5     5     5     5     5     5     5     5     5     5     5     5     5     5     5    20   -10     5     5     5     5     5
  Columns 23 through 44
     5     5     5     5     5     5     5     5     5     5     5     5     5     5     5     5     5     5     5     5     5     5
  Columns 45 through 48
     5     5     5     5

You should see that the time vector is not an increasing sequence, something that interp1 needs. I'm not sure why it is that your sampling times vary like this. Perhaps you are a sloppy typist. Perhaps there is a local time warp in your vicinity. Perhaps your experiment was run on the deck of the starship Enterprise at an inopportune time. Have you recently traveled faster than light? Faster than sound? Faster than a rubber chicken on steroids?

Regardless, you need to correct the problem to be able to use interp1. Proper use of the function sort may be all it takes, or maybe you just need to be more careful about how you enter your data.

[Monty]

Great - thanks loads!

Answer 2

[nl2605]

Ohh I see now. Many of your data points are the same and not monotonically increasing or so.

http://www.mathworks.de/matlabcentral/newsreader/view_thread/324991

Perhaps this will help. 'unique' function will omit the repeated values. But all the data points are equally important for you I guess?

Then maybe 'griddatan' would work for you.

[Monty]

Thanks - have tried that and it needs X to have two columns.... any ideas?

[John D'Errico]

griddatan does NOT work with replicates.