def test_Model(U, V):
# test the precision
k_num = len(top_k)
# k_num-length list to record F1 and NDCG
F1 = np.zeros(k_num)
NDCG = np.zeros(k_num)
# test all test samples
user_num = 0
for u in range(M):
# the data in test set is [[i, i, i, i],...,[i, i]]
test_item = Test[u]
if len(test_item) > 0:
user_num += 1
# score all items
score = np.dot(V, U[u])
# order
b = zip(score, range(N))
b.sort(key=lambda x: x[0])
order = [x[1] for x in b]
order.reverse()
# remove the training samples from the recommendations
train_positive = train_data_aux[u]
for item in train_positive:
order.remove(item)
# for each k, calculate top_k
for i in range(len(top_k)):
F1[i] += evaluation_F1(order, top_k[i], test_item)
NDCG[i] += evaluation_NDCG(order, top_k[i], test_item)
# calculate the average
F1 = (F1 / user_num).tolist()
NDCG = (NDCG / user_num).tolist()
return F1, NDCG
def test_DCFA(U, Vu, Vt, T, M, N, F):
# test the effectiveness
U = mat(U)
Vu = mat(Vu)
Vt = mat(Vt)
T = mat(T)
F = mat(F)
M = mat(M)
N = mat(N)
k_num = len(top_k)
# k_num-long lists to record F1 and NDCG
F1 = np.zeros(k_num)
NDCG = np.zeros(k_num)
num_item = len(Test)
# choose batch_size_test test samples randomly
for i in range(batch_size_test):
j = int(math.floor(num_item * random.random()))
# test data: [u, [i, i, i, i], [r, r, r]], where u, i, r are for user, item, time, respectively
u = Test[j][0]
test_item = Test[j][1]
# score for all users
Order = []
for r in Test[j][2]:
# for each r, score all items
UV = U[u] * Vu.T + M[u] * F.T
VT = T[r] * Vt.T + N[r] * F.T
UV = np.array(UV.tolist()[0])
VT = np.array(VT.tolist()[0])
score = (UV * VT).tolist()
# order
b = zip(score, range(len(score)))
b.sort(key=lambda x: x[0])
order = [x[1] for x in b]
order.reverse()
Order.append(order)
# train samples
train_positive = train_data_aux[u][0]
# we have len(train_data_aux[u][1]) k-length recommendation lists for each user,
# to compare fairly with other baselines, we combine len(train_data_aux[u][1]) k-length list to one k-length lists for each user
# we will remove at most len(train_positive) train samples from order, so we return k+len(train_positive) items
order = get_order(Order, top_k[-1] + len(train_positive))
# remove the train samples from the recommendations
for item in train_positive:
try:
order.remove(item)
except:
continue
# we also remove the train samples from test samples
test_item = list(set(test_item) - set(train_positive))
# test F1 and NDCG for each k
for i in range(len(top_k)):
F1[i] += evaluation_F1(order, top_k[i], test_item)
NDCG[i] += evaluation_NDCG(order, top_k[i], test_item)
# calculate the average
F1 = (F1 / batch_size_test).tolist()
NDCG = (NDCG / batch_size_test).tolist()
return F1, NDCG
