Feature Correlation

This visualizer calculates Pearson correlation coefficients and mutual information between features and the dependent variable. This visualization can be used in feature selection to identify features with high correlation or large mutual information with the dependent variable.

Pearson Correlation

The default calculation is Pearson correlation, which is perform with scipy.stats.pearsonr.

from sklearn import datasets
from yellowbrick.target import FeatureCorrelation

# Load the regression data set
data = datasets.load_diabetes()
X, y = data['data'], data['target']
feature_names = np.array(data['feature_names'])

visualizer = FeatureCorrelation(labels=feature_names)
visualizer.fit(X, y)
visualizer.poof()
../../_images/feature_correlation_pearson.png

Mutual Information - Regression

Mutual information between features and the dependent variable is calculated with sklearn.feature_selection.mutual_info_classif when method='mutual_info-classification' and mutual_info_regression when method='mutual_info-regression'. It is very important to specify discrete features when calculating mutual information because the calculation for continuous and discrete variables are different. See scikit-learn documentation for more details.

from sklearn import datasets
from yellowbrick.target import FeatureCorrelation

# Load the regression data set
data = datasets.load_diabetes()
X, y = data['data'], data['target']
feature_names = np.array(data['feature_names'])

discrete_features = [False for _ in range(len(feature_names))]
discrete_features[1] = True

visualizer = FeatureCorrelation(method='mutual_info-regression',
                                labels=feature_names)
visualizer.fit(X, y, discrete_features=discrete_features, random_state=0)
visualizer.poof()
../../_images/feature_correlation_mutual_info_regression.png

Mutual Information - Classification

By fitting with a pandas DataFrame, the feature labels are automatically obtained from the column names. This visualizer also allows sorting of the bar plot according to the calculated mutual information (or Pearson correlation coefficients) and selecting features to plot by specifying the names of the features or the feature index.

from sklearn import datasets
from yellowbrick.target import FeatureCorrelation

# Load the regression data set
data = datasets.load_diabetes()
X, y = data['data'], data['target']
feature_names = np.array(data['feature_names'])
X_pd = pd.DataFrame(X, columns=feature_names)

feature_to_plot = ['alcohol', 'ash', 'hue', 'proline', 'total_phenols']

visualizer = FeatureCorrelation(method='mutual_info-classification',
                                feature_names=feature_to_plot, sort=True)
visualizer.fit(X_pd, y, random_state=0)
visualizer.poof()
../../_images/feature_correlation_mutual_info_classification.png

API Reference

Feature Correlation to Dependent Variable Visualizer.

class yellowbrick.target.feature_correlation.FeatureCorrelation(ax=None, method='pearson', labels=None, sort=False, feature_index=None, feature_names=None, **kwargs)[source]

Bases: yellowbrick.target.base.TargetVisualizer

Displays the correlation between features and dependent variables.

This visualizer can be used side-by-side with yellowbrick.features.JointPlotVisualizer that plots a feature against the target and shows the distribution of each via a histogram on each axis.

Parameters:
ax : matplotlib Axes, default: None

The axis to plot the figure on. If None is passed in the current axes will be used (or generated if required).

method : str, default: ‘pearson’

The method to calculate correlation between features and target. Options include:

  • ‘pearson’, which uses scipy.stats.pearsonr
  • ‘mutual_info-regression’, which uses mutual_info-regression from sklearn.feature_selection
  • ‘mutual_info-classification’, which uses mutual_info_classif from sklearn.feature_selection
labels : list, default: None

A list of feature names to use. If a DataFrame is passed to fit and features is None, feature names are selected as the column names.

sort : boolean, default: False

If false, the features are are not sorted in the plot; otherwise features are sorted in ascending order of correlation.

feature_index : list,

A list of feature index to include in the plot.

feature_names : list of feature names

A list of feature names to include in the plot. Must have labels or the fitted data is a DataFrame with column names. If feature_index is provided, feature_names will be ignored.

kwargs : dict

Keyword arguments that are passed to the base class and may influence the visualization as defined in other Visualizers.

Examples

>>> viz = FeatureCorrelation()
>>> viz.fit(X, y)
>>> viz.poof()
Attributes:
features_ : np.array

The feature labels

scores_ : np.array

Correlation between features and dependent variable.

draw()[source]

Draws the feature correlation to dependent variable, called from fit.

finalize()[source]

Finalize the drawing setting labels and title.

fit(X, y, **kwargs)[source]

Fits the estimator to calculate feature correlation to dependent variable.

Parameters:
X : ndarray or DataFrame of shape n x m

A matrix of n instances with m features

y : ndarray or Series of length n

An array or series of target or class values

kwargs : dict

Keyword arguments passed to the fit method of the estimator.

Returns:
self : visualizer

The fit method must always return self to support pipelines.