def test_boxcox_transform(R, metadata, Lambda, threshold, zerovalue, inverse, expected):
"""Test the boxcox_transform."""
assert_array_almost_equal(
transformation.boxcox_transform(
R, metadata, Lambda, threshold, zerovalue, inverse
)[0],
expected,
)
#
# To find a suitable lambda, we will experiment with a range of values
# and select the one that produces the most symmetric distribution, i.e., the
# lambda associated with a value of skewness closest to zero.
# To visually compare the results, the transformed data are standardized.
#
# .. _`Box and Cox (1964)`: https://doi.org/10.1111/j.2517-6161.1964.tb00553.x
data = []
labels = []
skw = []
# Test a range of values for the transformation parameter Lambda
Lambdas = np.linspace(-0.4, 0.4, 11)
for i, Lambda in enumerate(Lambdas):
R_, _ = transformation.boxcox_transform(R, metadata, Lambda)
R_ = (R_ - np.mean(R_)) / np.std(R_)
data.append(R_)
labels.append("{0:.2f}".format(Lambda))
skw.append(skew(R_)) # skewness
# Plot the transformed data distribution as a function of lambda
plot_distribution(data, labels, skw)
plt.title("Box-Cox transform")
plt.tight_layout()
plt.show()
# Best lambda
idx_best = np.argmin(np.abs(skw))
Lambda = Lambdas[idx_best]
