def test_one_user(x):
# user u's ratings for user u
rating = x[0]
#uid
u = x[1]
#user u's items in the training set
item_num = x[2]
data_generator = data_generator_all[x[3]]
training_items = data_generator.train_items[u]
#user u's items in the test set
user_pos_test = data_generator.test_set[u]
all_items = set(range(item_num))
test_items = list(all_items - set(training_items))
item_score = []
for i in test_items:
item_score.append((i, rating[i]))
item_score = sorted(item_score, key=lambda x: x[1])
item_score.reverse()
item_sort = [x[0] for x in item_score]
r = []
for i in item_sort:
if i in user_pos_test:
r.append(1)
else:
r.append(0)
recall_20 = ut.recall_at_k(r, 20, len(user_pos_test))
recall_40 = ut.recall_at_k(r, 40, len(user_pos_test))
recall_60 = ut.recall_at_k(r, 60, len(user_pos_test))
recall_80 = ut.recall_at_k(r, 80, len(user_pos_test))
recall_100 = ut.recall_at_k(r, 100, len(user_pos_test))
ap_20 = ut.average_precision(r, 20)
ap_40 = ut.average_precision(r, 40)
ap_60 = ut.average_precision(r, 60)
ap_80 = ut.average_precision(r, 80)
ap_100 = ut.average_precision(r, 100)
return np.array([
recall_20, recall_40, recall_60, recall_80, recall_100, ap_20, ap_40,
ap_60, ap_80, ap_100
])
def simple_test_one_user(id, reclist):
"""
测试单个用户推荐的准确度， 结果是一个列表，包括top3, top5, top10的召回率、准确率、MRR、NDCG
:param id: 用户id
:param reclist: 生成的打分列表
:return: 召回率、准确率、MRR、NDCG
"""
global loader
if id in trainset_dict.keys():
for i in trainset_dict[id]:
reclist[i] = 0
recom = np.argsort(-reclist)
gnd = set(testset_dict[id])
pos = np.zeros(n_items)
for i, e in enumerate(recom):
if e in gnd:
pos[i] = 1
result = []
for k in [3, 5, 10]:
r = utils.recall_at_k(pos, k, len(gnd))
p = utils.precision_at_k(pos, k)
mrr = utils.mrr_at_k(pos, k)
ndcg = utils.ndcg_at_k(pos, k)
result.extend([r, p, mrr, ndcg])
return result
def tfidf():
print('Loading data ...')
train_c, train_r, train_l, \
dev_c, dev_r, dev_l, \
test_c, test_r, test_l = load_data(ratio=0.05)
n_dev = dev_c.shape[0]
n_test = test_c.shape[0]
# Transform to text corpus for tfidf vectorizer
print('Transforming to text corpuses ...')
_, inv_word_index, _ = load_dictionary()
train_c_corpus = dataset_to_corpus(train_c, inv_word_index)
train_r_corpus = dataset_to_corpus(train_r, inv_word_index)
dev_c_corpus = dataset_to_corpus(dev_c, inv_word_index)
dev_r_corpus = dataset_to_corpus(dev_r, inv_word_index)
test_c_corpus = dataset_to_corpus(test_c, inv_word_index)
test_r_corpus = dataset_to_corpus(test_r, inv_word_index)
# Transform to tfidf vector
print('Fitting a tfidf model ...')
vectorizer = TfidfVectorizer()
X = vectorizer.fit_transform(train_c_corpus + train_r_corpus)
print('Transforming to tfidf features ...')
X_dev_c = vectorizer.transform(dev_c_corpus)
X_dev_r = vectorizer.transform(dev_r_corpus)
X_test_c = vectorizer.transform(test_c_corpus)
X_test_r = vectorizer.transform(test_r_corpus)
# Tfidf predict (cosine similarity)
print('Predicting ...')
def cos_sim(x, y):
return x * y.transpose() / np.sqrt(
x * x.transpose() * y * y.transpose())
y_pred_dev = np.array(
[cos_sim(X_dev_c[i, :], X_dev_r[i, :]) for i in range(n_dev)])
y_pred_test = np.array(
[cos_sim(X_test_c[i, :], X_test_r[i, :]) for i in range(n_test)])
# Evaluate results
print('Results:\n')
#print('Validation set')
#for group_size in [2, 10]:
# for k in [1, 2, 5]:
# if k >= group_size:
# break
# r = recall_at_k(y_pred_dev, k, group_size)
# print('recall@{} ({} options): {}'.format(k, group_size-1, r))
#print('Testing set')
for group_size in [2, 10]:
for k in [1, 2, 5]:
if k >= group_size:
break
r = recall_at_k(y_pred_test, k, group_size)
print('recall@{} ({} options): {}'.format(k, group_size - 1, r))
def get_full_sort_score(self, epoch, answers, pred_list):
recall, ndcg = [], []
for k in [5, 10, 15, 20]:
recall.append(recall_at_k(answers, pred_list, k))
ndcg.append(ndcg_k(answers, pred_list, k))
post_fix = {
"Epoch": epoch,
"HIT@5": '{:.4f}'.format(recall[0]),
"NDCG@5": '{:.4f}'.format(ndcg[0]),
"HIT@10": '{:.4f}'.format(recall[1]),
"NDCG@10": '{:.4f}'.format(ndcg[1]),
"HIT@20": '{:.4f}'.format(recall[3]),
"NDCG@20": '{:.4f}'.format(ndcg[3])
}
print(post_fix)
with open(self.args.log_file, 'a') as f:
f.write(str(post_fix) + '\n')
return [recall[0], ndcg[0], recall[1], ndcg[1], recall[3],
ndcg[3]], str(post_fix)
def simple_test_one_user(x):
id, lst, idx = x[0], x[1], x[2]
global loader
if id in loader.trainset_dict.keys():
for i in loader.trainset_dict[id]:
lst[i] = 0
recom = np.argsort(-lst)
gnd = set(loader.testset_dict[id])
pos = np.zeros(opt.n_item)
for i, e in enumerate(recom):
if e in gnd:
pos[i] = 1
result = []
for k in [3, 5, 10]:
r = utils.recall_at_k(pos, k, len(gnd))
p = utils.precision_at_k(pos, k)
mrr = utils.mrr_at_k(pos, k)
ndcg = utils.ndcg_at_k(pos, k)
result.extend([r, p, mrr, ndcg])
return result