[PYTHON] Differences between numpy and pandas methods for finding variance

TL;DR

I'm doing distributed processing of numpy and pandas, and they don't match, so why? I will leave a note because it became.

The result of the method for finding var in numpy and pandas does not match the default value

Test with a simple, randomly generated matrix. It doesn't really match.

import numpy as np
import pandas as pd

X = np.random.randn(10, 10)
df = pd.DataFrame(data=X)

np.allclose(X, df.values)
# True

X_var = np.var(X, axis=1)
df_var = df.var(axis=1)

np.allclose(X_var, df_var.values)
# False

When I actually check the documentation, the default is ddof = 0 in numpy.var. , Pandas.DataFrame.var defaults to ddof = 1 ..

If you align the default values, the results will match.

X_var_ddof1 = np.var(X, ddof=1, axis=1)
df_var_ddof1 = df.var(axis=1)

np.allclose(X_var_ddof1, df_var_ddof1.values)
# True

I thought that the calculation results wouldn't match, but in fact there was a slight difference between numpy and pandas. I'd like you to unify it, but I'll publish a memo in case someone was addicted to it.