def get_metric(score_label):
Precision = np.zeros(20)
NDCG = np.zeros(20)
AUC = 0.
score_df = pd.DataFrame(score_label, columns=['uid', 'score', 'label'])
num = 0
score_label_all = []
for uid, hist in score_df.groupby('uid'):
if hist.shape[0] < 10:
continue
score = hist['score'].tolist()
label = hist['label'].tolist()
score_label_u = []
for i in range(len(score)):
score_label_u.append([score[i], label[i]])
score_label_all.append([score[i], label[i]])
precision, ndcg, auc, mae, mrse = calc_metric(score_label_u)
Precision += precision
NDCG += ndcg
AUC += auc
num += 1
score_label_all = sorted(score_label_all, key=lambda d: d[0], reverse=True)
GPrecision = np.array([
eval.precision_k(score_label_all,
k * len(score_label_all) / 100) for k in range(1, 21)
])
GAUC = eval.auc(score_label_all)
MAE = eval.mae(score_label_all)
MRSE = eval.mrse(score_label_all)
return Precision / num, NDCG / num, AUC / num, GPrecision, GAUC, MAE, MRSE
def calc_metric(score_label_u):
score_label_u = sorted(score_label_u, key=lambda d: d[0], reverse=True)
precision = np.array(
[eval.precision_k(score_label_u, k) for k in range(1, 21)])
ndcg = np.array([eval.ndcg_k(score_label_u, k) for k in range(1, 21)])
auc = eval.auc(score_label_u)
mae = eval.mae(score_label_u)
mrse = eval.mrse(score_label_u)
return precision, ndcg, auc, mae, mrse