kde2d error

[trungnth]

Dear respected developer,
I'm calculating the pdf of (x,y) data using kde2d in KDE-diffusion but I got an error that I can't comprehend. My code is below:

import numpy as np
import matplotlib.pyplot as plt
import matplotlib as mpl
from matplotlib import pyplot
import pandas as pd

x = pd.read_excel("x.xlsx")
y = pd.read_excel("y.xlsx")

from kde_diffusion import kde2d

(density, grid, bandwidth) = kde2d(x, y, n=512, limits=None)


# Display estimated density as image.
from matplotlib import pyplot
figure = pyplot.figure()
axes = figure.add_subplot()
axes.imshow(density.T)
pyplot.show()

The same task using matlab script of Zdravko Botev works fine without any error. Attached is my (x,y) data. Thank you for your time and help.

xy.xlsx
xy-matlab result.pdf

[john-hen]

Hi. I'm not sure I can help with that. When I run your code, I get FileNotFoundError: [Errno 2] No such file or directory: 'x.xlsx'. Which is understandable, as you provided no file with that name. You did not mention the error that you encountered.

Also, please understand that if the actual error is some convergence issue, then this library simply relies on SciPy to do its job, and there's not much I can do (or maybe: have time to do) about it.

[trungnth]

Thank you for your reply. It's my fault, I've corrected my mistake and it works perfectly. Do you have a plan to implement this algorithm for multivariable kernel density estimation? There's an implementation for n-d in Matlab by Zdravko Botev. https://www.mathworks.com/matlabcentral/fileexchange/58312-kernel-density-estimator-for-high-dimensions

[john-hen]

No, I have no plan to implement that. In fact, I was aware of that implementation for n dimensions when I migrated the code for 1d and 2d to Python. And there was a reason why I didn't include it. Unfortunately, I forgot what the reason was. 🤷 Maybe that it doesn't reproduce the results of kde2d for n=2? I don't know if that's true, but it must have been something like that. I was only ever interested in the 2-d case anyway. That being said, I no longer use this library in my day job, so it's unlikely I'll invest time in understanding the underlying math.