2D 数组的 median_absolute_deviation 抛出 AxisError: axis 1 is out of bounds for array of dimension 1

Question

The chunk of code is:

import scipy as sp 
import numpy as np

print(temps.shape)
print(temps.ndim)
mad = sp.stats.median_absolute_deviation(temps, axis=1, nan_policy='omit')
med = np.median(temps, axis=1)
mean = np.mean(temps,axis=1)

temps.shape is (992,2048) and temps.ndim is 2

but line 3 using median_absolute_deviation throws AxisError : axis 1 is out of bounds for array of dimension 1. If I comment out the line, med and mean run fine with no axis error. Why is this happening and how do I get it to compute the median absolute deviation along the axis of length 2048?

Answer 1

I do not have your exact array, but I have tried replicating your code to produce your error:

from scipy.stats import median_absolute_deviation
import numpy as np

# mimicking your data with random numbers
temps = np.random.normal(0.0, 1.0, size=(992,2048))
# setting values of axis 1 to NaN
temps[:,1] = np.nan

print(temps.shape)
print(temps.ndim)
mad = median_absolute_deviation(temps, axis=1, nan_policy='omit')
med = np.median(temps, axis=1)
mean = np.mean(temps,axis=1)

With the argument nan_policy='omit' in median_absolute_deviation() you are effectively asking the function to remove all the np.nan values from the array temps .

As I have shown above, if there is an entire column of these values, they are removed by calling nan_policy='omit' and so the array temps is reduced to an array of dimension 1 and the errors you see arise.