Source code for pyart.util.circular_stats
"""
Functions for computing statistics on circular (directional) distributions.
"""
import numpy as np
# For details on these computation see:
# https://en.wikipedia.org/wiki/Directional_statistics
# https://en.wikipedia.org/wiki/Mean_of_circular_quantities
[docs]def compute_directional_stats(field, avg_type="mean", nvalid_min=1, axis=0):
"""
Computes the mean or the median along one of the axis (ray or range)
Parameters
----------
field : ndarray
the radar field data
avg_type :str
the type of average: 'mean' or 'median'
nvalid_min : int
the minimum number of points to consider the stats valid. Default 1
axis : int
the axis along which to compute (0=ray, 1=range)
Returns
-------
values : ndarray 1D
The resultant statistics
nvalid : ndarray 1D
The number of valid points used in the computation
"""
if avg_type == "mean":
values = np.ma.mean(field, axis=axis)
else:
values = np.ma.median(field, axis=axis)
# Set to non-valid if there is not a minimum number of valid gates
valid = np.logical_not(np.ma.getmaskarray(field))
nvalid = np.sum(valid, axis=axis, dtype=int)
values[nvalid < nvalid_min] = np.ma.masked
return values, nvalid
[docs]def mean_of_two_angles(angles1, angles2):
"""
Compute the element by element mean of two sets of angles.
Parameters
----------
angles1 : array
First set of angles in radians.
angles2 : array
Second set of angles in radians.
Returns
-------
mean : array
Elements by element angular mean of the two sets of angles in radians.
"""
angles1 = np.asanyarray(angles1)
angles2 = np.asanyarray(angles2)
x = (np.cos(angles1) + np.cos(angles2)) / 2.0
y = (np.sin(angles1) + np.sin(angles2)) / 2.0
return np.arctan2(y, x)
[docs]def mean_of_two_angles_deg(angle1, angle2):
"""
Compute the element by element mean of two sets of angles in degrees.
Parameters
----------
angle1 : array
First set of angles in degrees.
angle2 : array
Second set of angles in degrees.
Returns
-------
mean : array
Elements by element angular mean of the two sets of angles in degrees.
"""
return np.rad2deg(mean_of_two_angles(np.deg2rad(angle1), np.deg2rad(angle2)))
[docs]def angular_mean(angles):
"""
Compute the mean of a distribution of angles in radians.
Parameters
----------
angles : array like
Distribution of angles in radians.
Returns
-------
mean : float
The mean angle of the distribution in radians.
"""
angles = np.asanyarray(angles)
x = np.cos(angles)
y = np.sin(angles)
return np.arctan2(np.mean(y), np.mean(x))
[docs]def angular_std(angles):
"""
Compute the standard deviation of a distribution of angles in radians.
Parameters
----------
angles : array like
Distribution of angles in radians.
Returns
-------
std : float
Standard deviation of the distribution.
"""
angles = np.asanyarray(angles)
x = np.cos(angles)
y = np.sin(angles)
norm = np.sqrt(np.mean(x) ** 2 + np.mean(y) ** 2)
return np.sqrt(-2 * np.log(norm))
[docs]def angular_mean_deg(angles):
"""
Compute the mean of a distribution of angles in degrees.
Parameters
----------
angles : array like
Distribution of angles in degrees.
Returns
-------
mean : float
The mean angle of the distribution in degrees.
"""
return np.rad2deg(angular_mean(np.deg2rad(angles)))
[docs]def angular_std_deg(angles):
"""
Compute the standard deviation of a distribution of angles in degrees.
Parameters
----------
angles : array like
Distribution of angles in degrees.
Returns
-------
std : float
Standard deviation of the distribution.
"""
return np.rad2deg(angular_std(np.deg2rad(angles)))
[docs]def interval_mean(dist, interval_min, interval_max):
"""
Compute the mean of a distribution within an interval.
Return the average of the array elements which are interpreted as being
taken from a circular interval with endpoints given by interval_min and
interval_max.
Parameters
----------
dist : array like
Distribution of values within an interval.
interval_min, interval_max : float
The endpoints of the interval.
Returns
-------
mean : float
The mean value of the distribution.
"""
# transform distribution from original interval to [-pi, pi]
half_width = (interval_max - interval_min) / 2.0
center = interval_min + half_width
a = (np.asarray(dist) - center) / (half_width) * np.pi
# compute the angular mean and convert back to original interval
a_mean = angular_mean(a)
return (a_mean * (half_width) / np.pi) + center
[docs]def interval_std(dist, interval_min, interval_max):
"""
Compute the standard deviation of a distribution within an interval.
Return the standard deviation of the array elements which are interpreted
as being taken from a circular interval with endpoints given by
interval_min and interval_max.
Parameters
----------
dist : array_like
Distribution of values within an interval.
interval_min, interval_max : float
The endpoints of the interval.
Returns
-------
std : float
The standard deviation of the distribution.
"""
# transform distribution from original interval to [-pi, pi]
half_width = (interval_max - interval_min) / 2.0
center = interval_min + half_width
a = (np.asarray(dist) - center) / (half_width) * np.pi
# compute the angular standard dev. and convert back to original interval
a_std = angular_std(a)
return a_std * half_width / np.pi