Currently, I have a 3-dimensional DataArray using xarray and I am looking to apply a 1-dimensional filter along a specific dimension. Specifically, I aim to utilize the scipy.signal.medfilt() function in a 1-dimensional manner.
My current implementation involves the following code:
for sample in data_raw.coords["sample"]:
for experiment in data_raw.coords["experiment"]:
data_filtered.loc[sample,experiment,:] = signal.medfilt(data_raw.loc[sample,experiment,:], 15)
(The dimensions of my data array are "sample", "experiment", and "wave_number". This code applies a filter along the "wave_number" dimension.)
However, this method is quite time-consuming and I suspect that looping through coordinates may not be the most efficient approach. Therefore, I am considering using the xarray.apply_ufunc() function:
xr.apply_ufunc(np.linalg.norm, data, kwargs={"axis": 2}, input_core_dims=[["wave_number"]])
(This computes the vector's length along the "wave_number" dimension.)
In a previous instance, I looped through coordinates as well.
When I attempt to use:
xr.apply_ufunc(signal.medfilt, data_smooth, kwargs={"kernel_size": 15})
The resulting data array consists only of zeroes, likely due to applying a 3D median filter with NaN entries present in the data array. It seems evident that I need to provide a 1D array to the scipy.signal.medfilt() function, but unfortunately, there is no option to specify an axis for filtering compared to numpy.linalg.norm().
Given these challenges, how can I efficiently apply a 1D median filter without iterating through coordinates?