There's no single "best" test for an outlier. Ideally, you should incorporate a-priori information (e.g. "This parameter shouldn't be over x because of blah...").
Most tests for outliers use the median absolute deviation, rather than the 95th percentile or some other variance-based measurement. Otherwise, the variance/stddev that is calculated will be heavily skewed by the outliers.
Here's a function that implements one of the more common outlier tests.
def is_outlier(points, thresh=3.5):
"""
Returns a boolean array with True if points are outliers and False
otherwise.
Parameters:
-----------
points : An numobservations by numdimensions array of observations
thresh : The modified z-score to use as a threshold. Observations with
a modified z-score (based on the median absolute deviation) greater
than this value will be classified as outliers.
Returns:
--------
mask : A numobservations-length boolean array.
References:
----------
Boris Iglewicz and David Hoaglin (1993), "Volume 16: How to Detect and
Handle Outliers", The ASQC Basic References in Quality Control:
Statistical Techniques, Edward F. Mykytka, Ph.D., Editor.
"""
if len(points.shape) == 1:
points = points[:,None]
median = np.median(points, axis=0)
diff = np.sum((points - median)**2, axis=-1)
diff = np.sqrt(diff)
med_abs_deviation = np.median(diff)
modified_z_score = 0.6745 * diff / med_abs_deviation
return modified_z_score > thresh
As an example of using it, you'd do something like the following:
import numpy as np
import matplotlib.pyplot as plt
# The function above... In my case it's in a local utilities module
from sci_utilities import is_outlier
# Generate some data
x = np.random.random(100)
# Append a few "bad" points
x = np.r_[x, -3, -10, 100]
# Keep only the "good" points
# "~" operates as a logical not operator on boolean numpy arrays
filtered = x[~is_outlier(x)]
# Plot the results
fig, (ax1, ax2) = plt.subplots(nrows=2)
ax1.hist(x)
ax1.set_title('Original')
ax2.hist(filtered)
ax2.set_title('Without Outliers')
plt.show()
与恶龙缠斗过久,自身亦成为恶龙;凝视深渊过久,深渊将回以凝视…