{% block content %}
The plot shows the correlation between pair-wise features and label. The label is treated as a categorical variable. For two numerical features, the correlation is the absolute value of Pearson correlation. For a numerical feature and a categorical feature, correlation ratio is used for measuring the curvilinear relationship. For two categorical features, Cramér's V value is used which is based on Pearson's chi-squared statistic.
After training an AFD model, you will get the feature importance distribution. You can combine that with feature correlation to identify potential label leakage. For example, if a feature has >0.99 correlation with label and it has significantly higher feature importance than other features, then there's a risk of label leakage on that feature. 
{% if corr_rec.feature_corr %} 
{% endif %} {% endblock %}