PR curve, threshold

https://hmkim312.github.io/posts/%EC%A0%95%EB%B0%80%EB%8F%84(Precision)%EC%99%80_%EC%9E%AC%ED%98%84%EC%9C%A8(recall)/ 

정밀도(Precision)와 재현율(Recall)

import matplotlib.pyplot as plt
from sklearn.metrics import precision_recall_curve
import numpy as np
import os
def PR_curve(target, pred, iter, cfg):
    out_dir = os.path.join("./PR", cfg.TRAIN.exp_name.split('/')[-1])
    if not os.path.isdir(out_dir):
        os.mkdir(out_dir)
    
    plt.close()
    #import pdb; pdb.set_trace()
    plt.figure(figsize = (8, 6))
    #pred = lr.predict_proba(X_test)[:,1]
    
    # count_big, count_small = [], []
    # for index, i in enumerate(pred[:,1]):
    #     if i >0.99:
    #         count_big.append(index)
            
    #     elif i < 0.01:
    #         count_small.append(index)
        
        # if i >=1.0:
        #     pred[index, 1] = 0.9995 
 

    plt.ylim(0,1.01)
    plt.xlabel('num of images')
    plt.ylabel('prediction score')

    plt.plot(np.sort(pred[:,1]))
    plt.savefig(f"{out_dir}/element_{iter}.png", dpi = 600)
    plt.close()
    
    plt.ylim(0,1.01)
    plt.xlim(0,1.01)
    
    precision, recalls, thresholds = precision_recall_curve(target, pred[:,1])
    
    plt.plot(thresholds, precision[:len(thresholds)], label = 'precision')
    plt.plot(thresholds, recalls[:len(thresholds)], label = 'recall')
    plt.xlabel('threshold')
    plt.ylabel('P/R score')
    
    plt.grid()
    plt.legend()
    plt.show()
    plt.savefig(f"{out_dir}/PR_cruve_{iter}.png", dpi = 600)
    plt.close()