def get_performance(user_pos_test, r, auc, Ks):
'''
:param user_pos_test: user 测试集中真实交互的item
:param r: r = [1,0,1] 表示预测TOP-K是否命中
:param auc: auc =0 标量
:param Ks: TOP-K
:return:
'''
precision, recall, ndcg, hit_ratio, MAP = [], [], [], [], []
for K in Ks:
precision.append(metrics.precision_at_k(r, K))
recall.append(metrics.recall_at_k(r, K, len(user_pos_test)))
ndcg.append(metrics.ndcg_at_k(r, K))
hit_ratio.append(metrics.hit_at_k(r, K))
MAP.append(metrics.AP_at_k(r, K, len(user_pos_test)))
return {'recall': np.array(recall), 'precision': np.array(precision), 'ndcg': np.array(ndcg),
'hit_ratio': np.array(hit_ratio), 'MAP': np.array(MAP), 'auc': auc}
val_u_is[key] = random.sample(val, max_i_num)
preds = {}
for u in val_u_is.keys():
val_u_is[u] = list(val_u_is[u])
pred_rates = [algo.predict(u, i)[0] for i in val_u_is[u]]
rec_idx = np.argsort(pred_rates)[::-1][:args.topk]
top_n = np.array(val_u_is[u])[rec_idx]
preds[u] = list(top_n)
# get actual interaction info. of validation users
ur = defaultdict(list)
for u in val_set.user.unique():
ur[u] = val_set.loc[val_set.user == u, 'item'].values.tolist()
for u in preds.keys():
preds[u] = [1 if e in ur[u] else 0 for e in preds[u]]
val_kpi_k = np.mean(
[precision_at_k(r, args.topk) for r in preds.values()])
val_kpi.append(val_kpi_k)
# get top-N list for test users
preds = {}
for u in test_u_is.keys():
test_u_is[u] = list(test_u_is[u])
pred_rates = [algo.predict(u, i)[0] for i in test_u_is[u]]
rec_idx = np.argsort(pred_rates)[::-1][:args.topk]
top_n = np.array(test_u_is[u])[rec_idx]
preds[u] = list(top_n)
# get actual interaction info. of test users
test_ur = defaultdict(list)
for u in test_set.user.unique():
test_ur[u] = test_set.loc[test_set.user == u,
'item'].values.tolist()
item_pool = list(range(dataset.item_num))
for u in tqdm(val_user_set):
if len(candidates[u]) < max_i_num:
actual_cands = set(candidates[u])
neg_item_pool = set(range(dataset.train_list[fold].shape[1])) - set(ur[u])
neg_cands = random.sample(neg_item_pool, max_i_num - len(candidates[u]))
cands = actual_cands | set(neg_cands)
else:
cands = random.sample(candidates[u], max_i_num)
pred_rates = algo.user_vec[u, :].dot(algo.item_vec).toarray()[0, list(cands)]
rec_idx = np.argsort(pred_rates)[::-1][:args.topk]
preds[u] = list(np.array(list(cands))[rec_idx])
for u in preds.keys():
preds[u] = [1 if i in ur[u] else 0 for i in preds[u]]
val_kpi_k = np.mean([precision_at_k(r, args.topk) for r in preds.values()])
val_kpi.append(val_kpi_k)
print('Start test kpi calculation......')
# genereate top-N list for test user set
test_user_set = dataset.test_users
test_ur = defaultdict(list) # u的实际交互item
index = dataset.test.nonzero()
for u, i in zip(index[0], index[1]):
test_ur[u].append(i)
candidates = defaultdict(list)
for u in test_user_set:
unint = np.where(dataset.train_list[fold][u, :].toarray().reshape(-1) == 0)[0] # 未交互的物品
candidates[u] = list(set(unint) & set(test_ur[u])) # 未交互的物品中属于后续已交互的物品
max_i_num = 1000
for fold in range(len(train_set_list)):
print(f'Start train validation [{fold + 1}]')
reco = MostPopRecommender(k)
reco.fit(train_set_list[fold])
# get top-N list for test users
preds = reco.predict(test_set)
# get actual interaction info. of test users
test_ur = defaultdict(list)
for u in test_set.user.unique():
test_ur[u] = test_set.loc[test_set.user == u,
'item'].values.tolist()
for u in preds.keys():
preds[u] = [1 if e in test_ur[u] else 0 for e in preds[u]]
# calculate metrics
precision_k = np.mean([precision_at_k(r, k) for r in preds.values()])
fnl_precision.append(precision_k)
recall_k = np.mean(
[recall_at_k(r, len(test_ur[u]), k) for u, r in preds.items()])
fnl_recall.append(recall_k)
map_k = map_at_k(list(preds.values()))
fnl_map.append(map_k)
ndcg_k = np.mean([ndcg_at_k(r, k) for r in preds.values()])
fnl_ndcg.append(ndcg_k)
hr_k = hr_at_k(list(preds.values()), list(preds.keys()), test_ur)
fnl_hr.append(hr_k)