[Python] scikitplot metric visualization (binary case)

확률값을 기반으로 시각화

scikitplot을 사용해서 이진 분류 관련된 메트릭들 시각화하기

 

1. Confusion Matrix
2. Roc Curve
3. KS-Test(Kolmogorov-Smirnov)
4. Precision-Recall Curve
5. Cumulative Gains Curve
6. Lift Curve

 

 

def metric_vis_binary(probs , y_label, classes=[1,2]) :
    import scikitplot as skplt
    plt.style.use('classic')
    fig , ax = plt.subplots(nrows=3 , ncols=2,figsize=(15,15))
    axes = ax.flatten()
    skplt.metrics.plot_confusion_matrix(y_label ,
                                        np.where(probs[:,1] > 0.5 , 
                                                 classes[0], classes[1]),
                                        ax=axes[0])
    skplt.metrics.plot_roc(y_label, probs,ax=axes[1])
    skplt.metrics.plot_ks_statistic(y_label, probs,ax=axes[2])
    skplt.metrics.plot_precision_recall(y_label, probs,ax=axes[3])
    skplt.metrics.plot_cumulative_gain(y_label, probs,ax=axes[4])
    skplt.metrics.plot_lift_curve(y_label, probs,ax=axes[5])
    plt.show()
metric_vis_binary(valid_proba , valid_y, classes=[1,2])

 

scikit-plot.readthedocs.io/en/stable/metrics.html

Metrics Module (API Reference) — Scikit-plot documentation