bootstrap_decision_boundary.py

import numpy as np

def plot_decision_boundaries(X, y, model_class, bootstrap=False, 
                             x_label=None, y_label=None, title=None, **model_params):
    
    # adapted from https://gist.github.com/anandology/772d44d291a9daa198d4
    reduced_data = X[:, :2] # take only the first two feature columns (since we are plotting a 2D contour map)
    
    # we need to recombine the data and target together since we need
    # to bootstrap sample from them with replacement
    combined_data = np.column_stack((reduced_data, y))
    combined_df = pd.DataFrame(combined_data)

    bootstrapped_models = []

    if bootstrap: # bootstrap sample and fit model B times
        for b in range(B):
            bootstrap_model = model_class(**model_params) # instantiate a generic model
            bootstrap_dataset = combined_df.sample(len(combined_df), replace=True) # sample w/ replacement
            bootstrap_X = bootstrap_dataset.iloc[:, :-1] # get all columns except last one
            bootstrap_y = bootstrap_dataset.iloc[:,-1] # get last column (target)
            bootstrap_model.fit(bootstrap_X, bootstrap_y)
            bootstrapped_models.append(bootstrap_model)
    else: # single model use case
        model = model_class(**model_params)
        model.fit(reduced_data, y)

    # Step size of the mesh. Decrease to increase the quality of the VQ.
    h = .1     # point in the mesh [x_min, m_max]x[y_min, y_max].    

    # Plot the decision boundary. For that, we will assign a color to each
    x_min, x_max = reduced_data[:, 0].min() - 1, reduced_data[:, 0].max() + 1
    y_min, y_max = reduced_data[:, 1].min() - 1, reduced_data[:, 1].max() + 1
    xx, yy = np.meshgrid(np.arange(x_min, x_max, h), np.arange(y_min, y_max, h))

    # Obtain labels for each point in mesh using the model.
    x_min, x_max = X[:, 0].min() - 1, X[:, 0].max() + 1
    y_min, y_max = X[:, 1].min() - 1, X[:, 1].max() + 1
    xx, yy = np.meshgrid(np.arange(x_min, x_max, 0.1),
                         np.arange(y_min, y_max, 0.1))

       if bootstrap: # majority vote of the B bootstrapped models
        Z = np.zeros(xx.shape)
        for model in bootstrapped_models:
            Z += model.predict(np.c_[xx.ravel(), yy.ravel()]).reshape(xx.shape)
        # average and then round to either 0 or 1
        Z = np.around(Z / (len(bootstrapped_models) * 1.0))
    else:
        Z = model.predict(np.c_[xx.ravel(), yy.ravel()]).reshape(xx.shape)

    plt.contourf(xx, yy, Z, alpha=0.4, cmap='RdYlGn')
    plt.scatter(X[:, 0], X[:, 1], c=y, alpha=0.8)
    clb = plt.colorbar()
    clb.set_label('# of Predicted Conversions')
    
    # annotate the chart
    if x_label:
        plt.xlabel(x_label)
    if y_label:
        plt.ylabel(y_label)
    if title:
        plt.title(title)
    return plt