Necessity of FFTKDE grid enclosure with reflected boundaries about kdepy HOT 1 CLOSED

I'll give some general comments which I hope can be helpful. Honestly you are kind of on your own here, since (1) I don't have time to work on this and (2) your requirements are a bit exotic, so you'll have to think it through.

If the restriction is removed, the user could unintentionally get bad results. In theory I think it could be removed, but it would make the library less "safe". I think the feature has saved many more users than it has annoyed, so I am in favor of keeping it.

As per your specific problem, I see two solutions:

• If the bandwidth is small compared to the data (e.g. data on [0, 1] and bw = 0.01), then you don't have to mirror all of the data. If you mirror e.g. 3*bw, then you might get away with much less computation. This might be the simplest solution.
• If you want mirroring across each low/high threshold in each dimension, then you could fork the library and adapt it to your specific needs. If you (1) set up a grid that is barely outside of the data and (2) use a convolution that wraps around the boundaries, you should be able to accomplish what you want in a computationally efficient way. After all convolutions wrap around by default, which is exactly what you want. You will have to look at grid validation/creation, and make sure that the convolution wraps around. Try this: create a grid, perform linear binning, adapt this code to convolve a kernel with the binned data.

Best of luck to you! I'll close this issue, since I will not work on it. Let me know if you have questions. If someone else wants to implement a MirroredFFTKDE let me know, and I'll re-open the issue.

from kdepy.