如何在sklearn.metrics中为函数precision_recall_curve和roc_curve获得相同的阈值

Question

I need to make a table with the TPR and FPR values, as well as precision and recall. I am using the roc_curve and precision_recall_curve functions from sklearn.metrics package in python. My problem is that every function give me a different vector for the thresholds, and I need only one, to merge the values as columns in a single table. Could anyone help me?

Thanks in advance

Answer 1

The threshold values have two major differences.

1. The orders are different. roc_curve has thresholds in decreasing order, while precision_recall_curve has thresholds in increasing order.

2. The numbers are different. In roc_curve , n_thresholds = len(np.unique(probas_pred)) , while in precision_recall_curve the number n_thresholds = len(np.unique(probas_pred)) - 1 . In the latter, the smallest threshold value from roc_curve is not included. In the same time, the last precision and recall values are 1. and 0. respectively with no corresponding threshold. Therefore, the numbers of items for tpr, fpr, precision and recall are the same.

So, back to your question, how to make a table to include tpr, fpr, precision and recall with corresponding thresholds? Here are the steps:

1. Discard the last precision and recall values
2. Reverse the precision and recall values
3. Compute the precision and recall values corresponding to the lowest threshold value from the thresholds of roc_curve
4. Put all the values into the same table