Determine features and targets to use for your model using pearson correlation coefficient

features_and_targets.py

def get_features_targets(df, target_correlation_map,
                         num_features = 2, correlation_method = 'pearson', correlation_barrier = 0.6):
    from itertools import combinations
    targets = list(df.columns.values)
    target_map = {}
    #construct a target map using the correlation matrix
    for features in combinations(targets, num_features + 1):
        #target is first one, remaining are feature combinations
        target = features[0]
        if target not in target_map:
            target_map[target] = []
        target_map[target].append(features[1:])
        #populate the target map for remaining features in this combination
        for feature_index in range(1, num_features+1):
            target = features[feature_index]
            if target not in target_map:
                target_map[target] = []
            target_map[target].append(features[:feature_index] + features[feature_index+1:])

    #now get the correlation matrix and apply the barrier against the feature combination for the targets
    correlation_matrix = df.corr(method = correlation_method)
    for target, features in target_map.items():
        for feature_set in features:
            feature_correlations = [ f for f in feature_set if abs(correlation_matrix[target][f]) >= correlation_barrier ]
            if len(feature_correlations) == len(feature_set):
                #acceptance
                if target not in target_correlation_map:
                    target_correlation_map[target] = []
                target_correlation_map[target].append(feature_set)
    for target, features in target_correlation_map.items():
        print('Target:%s' %target)
        print('Selected features')
        print('-'*80)
        for f in features:
            print('Features:{}'.format(f))
            print('-'*40)
        print('-'*80)
    return target_correlation_map