Feature rejection does not take into account negative correlations

[dmfolgado]

Describe the bug

The profilers seem not taking into account negative correlations to remove redundant features. Despite pandas-profiling does a good job detecting positive correlations on the dataset, negative correlations are not being detected. I do believe the reason lies the condition to remove a given variable assumes that the correlation is above a given positive threshold value.

To Reproduce

import numpy as np
import pandas as pd
import pandas_profiling

# Let's generate a dataset where the first and second variables are negative correlated and the third variable is random.
X=np.array([np.arange(100), np.arange(100)[::-1], np.random.randn(100)]).T
df = pd.DataFrame(X, columns=["1", "2", "3"])

profile=df.profile_report()
profile.get_rejected_variables()    # []