YeoJonson transformation
YeoJonson transform is one of the transform methods that bring numerical data closer to a normal distribution. Unlike the BoxCox transformation, which cannot be applied to data containing negative values, it can be applied even when negative values are included.
Unlike the box-cox transformation, this method can be used even when negative values are included.
I. Yeo and R.A. Johnson, “A New Family of Power Transformations to Improve Normality or Symmetry”, Biometrika 87.4 (2000):
from scipy import stats
import matplotlib.pyplot as plt
x = stats.loggamma.rvs(1, size=1000) - 0.5
plt.hist(x)
plt.axvline(x=0, color="r") # verify that there is data below 0 as well
plt.show()
import numpy as np
from scipy.stats import yeojohnson
plt.hist(yeojohnson(x))
plt.show()
