params.py 文件源码

def compute_score(self, conf, hy):
        RS = recall_score(self.y, hy, average=None)
        conf['_all_f1'] = M = {str(self.le.inverse_transform([klass])[0]): f1 for klass, f1 in enumerate(f1_score(self.y, hy, average=None))}
        conf['_all_recall'] = {str(self.le.inverse_transform([klass])[0]): f1 for klass, f1 in enumerate(RS)}
        conf['_all_precision'] = N = {str(self.le.inverse_transform([klass])[0]): f1 for klass, f1 in enumerate(precision_score(self.y, hy, average=None))}
        conf['_macrorecall'] = np.mean(RS)
        if len(self.le.classes_) == 2:
            conf['_macrof1'] = np.mean(np.array([v for v in conf['_all_f1'].values()]))
            conf['_weightedf1'] = conf['_microf1'] = f1_score(self.y, hy, average='binary')
        else:
            conf['_macrof1'] = f1_score(self.y, hy, average='macro')
            conf['_microf1'] = f1_score(self.y, hy, average='micro')
            conf['_weightedf1'] = f1_score(self.y, hy, average='weighted')
        conf['_accuracy'] = accuracy_score(self.y, hy)
        if self.score.startswith('avgf1:'):
            _, k1, k2 = self.score.split(':')
            conf['_' + self.score] = (M[k1] + M[k2]) / 2
        elif self.score.startswith('avgf1f0:'):
            _, k1, k2 = self.score.split(':')
            pos = (M[k1] + N[k1]) / 2.
            neg = (M[k2] + N[k2]) / 2.
            conf['_' + self.score] = (pos + neg) / 2.
        conf['_score'] = conf['_' + self.score]