data_handling#

Example Usage#

Full Contents:#

data_handling.inpaint_nans(A, method)#

INPAINT_NANS: in-paints over nans in an array usage: B=INPAINT_NANS(A) % default method usage: B=INPAINT_NANS(A,method) % specify method used

Solves approximation to one of several pdes to interpolate and extrapolate holes in an array

arguments (input):

A - nxm array with some NaNs to be filled in

method - (OPTIONAL) scalar numeric flag - specifies

which approach (or physical metaphor to use for the interpolation.) All methods are capable of extrapolation, some are better than others. There are also speed differences, as well as accuracy differences for smooth surfaces.

methods {0,1,2} use a simple plate metaphor. method 3 uses a better plate equation,

but may be much slower and uses more memory.

method 4 uses a spring metaphor. method 5 is an 8 neighbor average, with no

rationale behind it compared to the other methods. I do not recommend its use.

method == 0 –> (DEFAULT) see method 1, but

this method does not build as large of a linear system in the case of only a few NaNs in a large array. Extrapolation behavior is linear.

method == 1 –> simple approach, applies del^2

over the entire array, then drops those parts of the array which do not have any contact with NaNs. Uses a least squares approach, but it does not modify known values. In the case of small arrays, this method is quite fast as it does very little extra work. Extrapolation behavior is linear.

method == 2 –> uses del^2, but solving a direct

linear system of equations for nan elements. This method will be the fastest possible for large systems since it uses the sparsest possible system of equations. Not a least squares approach, so it may be least robust to noise on the boundaries of any holes. This method will also be least able to interpolate accurately for smooth surfaces. Extrapolation behavior is linear.

Note: method 2 has problems in 1-d, so this method is disabled for vector inputs.

method == 3 –+ See method 0, but uses del^4 for

the interpolating operator. This may result in more accurate interpolations, at some cost in speed.

method == 4 –+ Uses a spring metaphor. Assumes

springs (with a nominal length of zero) connect each node with every neighbor (horizontally, vertically and diagonally) Since each node tries to be like its neighbors, extrapolation is as a constant function where this is consistent with the neighboring nodes.

method == 5 –+ See method 2, but use an average

of the 8 nearest neighbors to any element. This method is NOT recommended for use.

arguments (output):

B - nxm array with NaNs replaced

Example

[x,y] = meshgrid(0:.01:1); z0 = exp(x+y); znan = z0; znan(20:50,40:70) = NaN; znan(30:90,5:10) = NaN; znan(70:75,40:90) = NaN;

z = inpaint_nans(znan);

data_handling#

Example Usage#

Full Contents:#

Indices and tables#