Python function to create multiple plots to evaluate the model performance visually.

evaluate_model.py

def evaluate_model(y_train, y_train_pred, y_test, y_test_pred):
    '''Creates multiple plots to evaluate the model performance visually.
    
    Parameters
    ----------
    y_train: array
        The target for the train set.
    y_train_pred: array
        The predicted target for the train set.
    y_test: numpy array
        The target for the test set.
    y_test_pred: array
        The predicted target for the test set.
    
    Returns
    -------
        Matplotlib GridSpec
    
    Notes
    ------
    Requirements: Matplotlib, Seaborn.
    '''
    from matplotlib.ticker import FuncFormatter
    from matplotlib.gridspec import GridSpec
    
    # This is optional, it's made for large values (i.e., money)
    def kilos(x, pos):
        'The two args are the value and tick position'
        return '%3.0fK' % (x * 1e-3)
    formatter = FuncFormatter(kilos)
    # End of optional
    
    fig = plt.figure(constrained_layout=False, figsize=(12, 9))
    widths = [5, 5]
    heights = [3, 5]
    gs = fig.add_gridspec(ncols=2, nrows=2, wspace=0.4, hspace=0.35,
                          width_ratios=widths,
                          height_ratios=heights)
    
    
    # Histogram of Errors
    ax1 = fig.add_subplot(gs[0, :])
    ax1.set_title('Histogram of Errors')
    train_errors = y_train - y_train_pred
    test_errors = y_test - y_test_pred
    rango = (min(np.min(train_errors), 
                 np.min(test_errors)),
             max(np.max(train_errors), 
                 np.max(test_errors)))        
    sns.distplot(train_errors, kde=False, bins=45, label='Train', 
                 hist_kws={'range': rango, 'alpha': 1}, 
                 ax=ax1)
    sns.distplot(test_errors, kde=False, bins=45, label='Test',
                 hist_kws={'range': rango, 'alpha': 1},
                 ax=ax1)
    ax1.set_ylabel('Quantity')
    ax1.set_xlabel('Error')
    ax1.legend(fancybox=True, loc='right')
    ax1.xaxis.set_major_formatter(formatter)
    
    # Residuals plot
    ax2 = fig.add_subplot(gs[1, 0])
    ax2.set_title('Residuals Plot')
    ax2.scatter(y_train_pred,  y_train_pred - y_train,
               s=25, edgecolor='white', linewidths=0.5, alpha=1,
               label='Train')
    ax2.scatter(y_test_pred, y_test_pred - y_test, 
               s=25, edgecolor='white', linewidths=0.5, alpha=1,
               label='Test')
    ax2.hlines(y=0, 
               xmin=np.min(ax2.get_xlim()), 
               xmax=np.max(ax2.get_xlim()), 
               color='black', alpha=0.9,
               lw=2)
    ax2.set_xlabel('Predicted values')
    ax2.set_ylabel('Residuals')
    ax2.legend(fancybox=True, loc='lower right')
    ax2.xaxis.set_major_formatter(formatter)
    ax2.yaxis.set_major_formatter(formatter)
    
    
    # Prediction errors  plot
    ax3 = fig.add_subplot(gs[1, 1])
    ax3.set_title('Prediction Errors Plot')
    ax3.scatter(y_test, y_test_pred, color=red,
                s=25, edgecolor='white', linewidths=0.5, alpha=1)
    lims = [np.min([ax3.get_xlim(), ax3.get_ylim()]),
            np.max([ax3.get_xlim(), ax3.get_ylim()])] 
    ax3.plot(lims, lims, '-k',  alpha=1, lw=2, label='Identity')
    sns.regplot(y_test, y_test_pred, scatter=False, ci=None,
                color='black', label='Best fit',
                line_kws={'ls': '--', 'alpha':0.6, 'lw':2})
    ax3.set_xlabel('Real (test)')
    ax3.set_ylabel('Predicted (test)')
    ax3.legend(fancybox=True, loc='lower right')
    ax3.xaxis.set_major_formatter(formatter)
    ax3.yaxis.set_major_formatter(formatter)