[Python] How to extract auc scores from scikitplot plot_precision_recall method?

I am trying to get the micro-average score as displayed in:

from scikitplot.metrics import plot_precision_recall

plot_precision_recall(np.array(y_test), y_pred_proba)
plt.show()


For example, I have the plot:



with micro-average of 0.975

I've tried using the auc method with precision_recall_curve and average_precision_score among other methods, but the results are not the same. If I could just extract the desired auc score from the plot_precision_recall method itself, that would be easiest.

I believe the issue is that, as per the documentation, use of the average='micro' argument "Will be ignored when y_true is binary" and thus the macro average is being reported, at least for roc_auc_score. For average_precision_score it's reporting only the PR AUC for the positive class.

NB: I am not including a reproducible example because I only want to get the respective scores from the plot_precision_recall method.

I've Googled up a storm with negative results.

Continue reading...