Inpaintings.jl provides a Julia version of MATLAB’s inpaint_nans function (originally written by John d’Errico, available on the MathWorks File Exchange website and ported here with his authorization by personal communication).

Simply put, Inpaintings.jl provides a simple inpaint function, which takes an array A as input and inpaints its missing values by solving a simple n-dimensional PDE. The inpaint function can also be used to inpaint NaNs or any other values, thanks to the syntax described below and in the documentation.


Like every Julia package you must first add it via ]add Inpaintings. And every time you want to use Inpaintings.jl, you must start with

julia> using Inpaintings

In order to inpaint an array A’s missing values, simply apply inpaint to your array:

julia> inpaint(A) # will inpaint missing values

The array to be inpainted can be a vector, a matrix, or even an n-dimensional array.

If your array A has some NaN values and is filled with floats otherwise, then

julia> inpaint(A) # will inpaint NaN values

Inpaintings.jl provides a syntax to inpaint any specified value via

julia> inpaint(A, -999) # will inpaint -999 values

(The value to inpaint can be specified as NaN or missing, too!)

Alternatively, Inpaintings.jl also provides a syntax taking a boolean function f as an argument before the array (f will be applied to all the elements of the array and must return a boolean).

julia> inpaint(f, A)

In this case, the values of A for which f returns true will be inpainted. (For example, f can be, e.g., ismissing or isnan, but it can also be x -> x < 0.)

Finally, Inpaintings.jl provides a syntax to allow some dimensions to be assumed cyclic:

julia> inpaint(A, cycledims=[1]) # will inpaint A with dimension 1 as cyclic

(The cyclic dimensions must be an array of Int64 that contains the dimension number of cyclic dimensions.)

See the docs if you want to see more examples.

Benoît Pasquier
Research Associate

My research interests include mathematics, oceanography, and computer science.