Table lookup based on 3 variables

Hello,

I am trying to take a large set of data and first create a table based on that data then perform a lookup to find my desried values based on 3 variables of the table.

The data begins as:

A: [10 89 23 4 700 8]

B: [1 234 34 6 89 77]

C: [-12 23 9 -400 62]

D:Z: Variables that will need to be looked up (All variables have the same number of columns as A, B & C)

Then I space out the vectors A, B & C and grid them:

A_min = min(A); A_max = max(A); gridsize = 250;

a = linspace(A_min, A_max, gridsize)

b = linspace(B_min, B_max, gridsize)

c = linspace(C_min, C_max, gridsize)

[A_grid, B_grid, C_grid] = meshgrid(a, b, c)

Then I try to grid the rest of the data to form a table of all the values based on the 3 gridded variables

output_D = griddata(A', B', C', D', A_grid, B_grid, C_grid, 'linear')

Which results in a output that is all NaN

I have been successful in doing this based off 2 variables

A_min = min(A); A_max = max(A); gridsize = 250;

a = linspace(A_min, A_max, gridsize)

b = linspace(B_min, B_max, gridsize)

[A_grid, B_grid] = meshgrid(a, b)

output_D = griddata(A', B', D', A_grid, B_grid, 'linear')

[John Doe]

If i've understood correctly I've solved a similar problem doing this:

Where you create a second table (nt) where the rows in nt are only those in t that satisfy the criteria for columns A,B,C you could expand this further if needed also.

You can also use < or > as necessary in place of ==.

I'm not sure if there is a faster or better way, but this has worked well for me. I usually use this for filtering a table to a specific range of data. If there is more than one result it would show all the possible combinations in table nt.

%Create random 5x5 array
x = magic (5);
%Conver array to table
t = array2table(x);
% Give table variable names
t.Properties.VariableNames = {'A','B','C','D','E'};
% Create new table based on criteria 
nt = t(t.A == 17 & t.B == 24 & t.C == 1,:);
% Results
nt.D
nt.E

[Stephen23]

"Which results in a output that is all NaN"

How did you check this? When I run your code (after a few minor bug fixes, e.g.C size, missing D, etc.), about sixteen percent of the values are not NaN:

>> numel(output_D)
ans =
    15625000
>> nnz(~isnan(output_D))
ans =
     2483115  

Most of your grid values are outside the convex hull of the input data, which according to the griddata documentation should therefore be returned as NaN. As far as I can tell, everything is working exactly as expected and documented.

Note that you should probably use ndgrid rather that meshgrid.

Note that scatteredInterpolant is required if you want to perform extrapolation.

[John]

Hi John & Stephen,

Appreciate you both looking into my issue. I am going for a solution that allows for interpolation between points, not just a lookup on values contained in the data sets so I do not think the first solution is appropriate.

The second solution made me go back and review my code and it does indeed agree with what you are saying.

My original question had A, B, C, D vectors as an example and the data in them is not th actual datasets I am using. My apologies for the confusion.

For consistancy of this question, let's say.

C: [-12 23 9 -400 62 -1]

D: [1:6]

E: [10:10:60]

F: [-1 -14 -16 -200 -30 -10]

Now that my outputs have been created, I want to be able to interpolate between values of A, B and C to find values of D, E, & F. Another desire is to create a human readable table of the newly gridded data. Are either of these possible? My inclination is the interpolation should be but creating a 2-D table from a 3-D grid wouldn't be.