def get_flag(m, n, sum11, l_array):
r = np.load('E:\\limbi\\n.npy')
c = np.load('E:\\limbi\\m.npy')
sum_ratings = len(r)
c_flag = []
for i in range(len(r)):
c_flag.append(1)
cf.show_time()
sum1 = 0
i = 0
for sum1 in range(sum_ratings):
if r[sum1] == i:
if c[sum1] in l_array[i]:
c_flag[sum1] = 1
else:
c_flag[sum1] = 0
else:
i += 1
if c[sum1] in l_array[i]:
c_flag[sum1] = 1
else:
c_flag[sum1] = 0
cf.show_time()
print(len(c_flag))
print('\n' + f'I have done it! {sum11}')
np.save(f'E:\\limbi\\c_flag{sum11}', c_flag)
def d_book0(n_number, m_number):
data = np.zeros((n_number, m_number), np.float16)
n_a = np.load('E:\\book\\n.npy')
m_a = np.load('E:\\book\\m.npy')
r_a = np.load('E:\\book\\r.npy')
t = len(n_a)
for i in range(t):
s_t = int(n_a[i])
l_t = int(m_a[i])
data[s_t][l_t] = r_a[i]
print('set the old dateset')
cf.show_time()
n_a = []
m_a = []
r_a = []
m_click = 0
for i in range(m_number):
s_t = 0
for j in range(n_number):
ratings_s = data[j][i]
if ratings_s != 0:
s_t += 1
n_a.append(j)
m_a.append(m_click)
r_a.append(ratings_s)
if s_t != 0:
m_click += 1
print(m_click)
np.save('E:\\book\\n', n_a)
np.save('E:\\book\\m', m_a)
np.save('E:\\book\\r', r_a)
print('order the dataset')
def local_differencial_privacy(eta, p):
r = np.load('D:\Labor\\n.npy')
c = np.load('D:\Labor\m.npy')
ratings = np.load('D:\Labor\\r.npy')
D = np.load('D:\Labor\D.npy')
D3 = np.load('D:\Labor\D3.npy')
U = np.load('D:\\U.npy')
V = np.load('D:\V.npy')
m = 131262
n = 138492
sum = 150629
print(sum)
q = 2700
k = 10
d = 15
# 规则化参数
lu = 10 ** -8
lv = 10 ** -8
for s1 in range(1):
print('This is the ' + str(s1 + 1) + 'iteration!')
filename = f'D:\V{p}'
filename1 = f'D:\\U{p}'
# save(filename1, U)
for it in range(k):
# 初始化U，V的偏导矩阵
print('The ' + str(it) + ' iteration!')
common_function.show_time()
# dV = np.zeros((m, d), float16)
rt = 1 / (it + 1) / (k ** 2)
dB = np.zeros((q, d))
dU = np.zeros((n, d))
sum_N = np.zeros((q, d))
sum1 = 0
i = 0
while i < n:
dV = np.zeros((m, d))
N = np.random.laplace(0, 2 * 10 * (15 ** 0.5) / eta, (m, d))
while i < n:
if r[sum1] != i:
i = i + 1
break
j = c[sum1]
T = np.dot(U[i], V[j].T)
dV[j] = -2 * U[i] * (ratings[sum1] - T)
dU[i] = dU[i] + (-2) * V[j] * (ratings[sum1] - T)
sum1 += 1
x = np.dot(D, dV + N)
y = np.dot(D, N)
sum_N += y
dB += x
dB = (dB - sum_N) / sum
dU /= sum
U = U - rt * (dU + 2 * lu * U)
V = V - rt * (np.dot(D3, dB) + 2 * lv * V)
common_function.show_time()
print(V)
print('Get V！')
np.save(filename, V)
np.save(filename1, U)
def get_items(r, c, ratings, m, n, sum11, k, index, eta):
index = index + "iR//"
sum_ratings = len(ratings)
print('计算相关物品编号...')
cf.show_time()
eta_k = eta / k
pr = math.exp(eta_k) / (math.exp(eta_k) + 1)
sum1 = 0
i = 0
# 首先选择k个用户相关物品：1若相关物品大于k，则随机从中选k个；
# 2若相关物品小于k，则不足的数量由非相关物品编号替换
user_list = []
other_list = []
sum_list = []
user_array = np.zeros((n, k))
for number_i in range(m):
sum_list.append(number_i)
other_list = sum_list[:]
while sum1 < sum_ratings:
j = c[sum1]
user_list.append(j)
other_list.remove(j)
sum2 = sum1 + 1
if sum2 < sum_ratings:
try:
test = r[sum2]
except Exception as er:
print(er, test)
break
if r[sum2] != i:
user_list_len = len(user_list)
user_list_select = []
if user_list_len > k:
user_list = sample_list(user_list, k)
user_list_len = len(user_list)
for user_list_number in range(user_list_len):
p = np.random.random()
if p > pr or p == pr:
user_list_select.append(user_list[user_list_number])
user_list_len = len(user_list_select)
if user_list_len < k:
t = k - user_list_len
other_list_select = sample_list(other_list, t)
user_list_select += other_list_select
user_list_len = len(user_list_select)
if user_list_len > k:
user_list_select = sample_list(user_list_select, k)
# print(len(user_list_select))
# print(i)
user_array[i] = user_list_select[:]
user_list = []
other_list = sum_list[:]
i += 1
sum1 += 1
print(f'完成物品编号选择！! {sum11}')
cf.show_time()
np.save(index + f'array_list{sum11}', user_array)
return user_array
def d_book1(n_number, m_number):
n_a = np.load('E:\\book\\n.npy')
m_a = np.load('E:\\book\\m.npy')
r_a = np.load('E:\\book\\r.npy')
t = len(n_a)
m_number = 26509
cf.show_time()
data = np.zeros((n_number, m_number))
for i in range(t):
s_t = int(n_a[i])
l_t = int(m_a[i])
data[s_t][l_t] = r_a[i]
print('get new dataset')
cf.show_time()
n_a = []
m_a = []
r_a = []
for i in range(n_number):
for j in range(m_number):
ratings_s = data[i][j]
if ratings_s != 0:
n_a.append(i)
m_a.append(j)
r_a.append(ratings_s)
print(i)
print('get the order list')
cf.show_time()
m_adjust = []
m_adjust.append(m_a[0])
for i in range(len(m_a)):
if m_a[i] not in m_adjust:
m_adjust.append(m_a[i])
np.save('E:\\book\\m_adjust', m_adjust)
print('Get new m order')
cf.show_time()
for i in range(len(m_a)):
for j in range(len(m_adjust)):
if m_a[i] == m_adjust[j]:
m_a[i] = j
print(j)
break
cf.show_time()
print('order the m')
n_adjust = [0]
ss = 0
for i in range(len(n_a) - 1):
j = i + 1
if n_a[j] == n_a[i]:
n_adjust.append(ss)
else:
ss += 1
n_adjust.append(ss)
print('order the n')
np.save('E:\\book\\n', n_a)
np.save('E:\\book\\m', m_a)
np.save('E:\\book\\r', r_a)
def machine_learning(sum11, U, V, r, c, ratings, m, n, d, lu, lv, k, sum, D,
D3):
r = np.load('D:\Labor\\n.npy')
c = np.load('D:\Labor\m.npy')
ratings = np.load('D:\Labor\\r.npy')
D = np.load('D:\Labor\D.npy')
D3 = np.load('D:\Labor\D3.npy')
m = 131262
# n = 138493
n = 138493
d = 15
lu = 10**-8
lv = 10**-8
k = 10
sum = 20000263
n_number = 1000
# U = load("D:\\U8.npy")
# print(U[:3], V[:3])
# V = load('D:\V8.npy')
for it in range(k):
rt = 1 / (it + 1)
print(str(it) + ' iterations！')
common_function.show_time()
dU = np.zeros((n, d))
dB = np.zeros((m, d))
sum1 = 0
i = 0
dV = np.zeros((m, d))
while sum1 < sum:
j = c[sum1]
T = np.dot(U[i], V[j].T)
dV[j] = -2 * U[i] * (ratings[sum1] - T)
dU[i] = dU[i] + (-2) * V[j] * (ratings[sum1] - T)
sum1 += 1
if sum1 < sum:
try:
test = r[sum1]
except Exception as er:
print(er)
break
if r[sum1] != i:
i += 1
dB = dB + dV
# laplace_n = np.random.laplace(0, 10 * 15 / sum11, (m, d))
dV = np.zeros((m, d), float16)
else:
dB = dB + dV
laplace_n = np.random.laplace(0, 10 * 10 / (20 - sum11 * 2), (m, d))
dV = dB / n + laplace_n
dU /= n
U = U - rt * (dU + 2 * lu * U)
V = V - rt * (dV + 2 * lv * V)
common_function.show_time()
print('\n' + f'I have done it! {sum11}')
save(f'D:\\U7_{sum11}', U)
save(f'D:\V7_{sum11}', V)
def machine_learning(list_in_order, sum11, U, V):
r = np.load('E:\\book\\n.npy')
c = np.load('E:\\book\m.npy')
ratings = np.load('E:\\book\\r.npy')
# D = np.load('E:\book\D.npy')
# D3 = np.load('E:\book\D3.npy')
length_list = len(list_in_order)
print('length_list', length_list)
m = 131262
# n = 138493
n = 138493
d = 15
lu = 10**-8
lv = 10**-8
k = 10
sum_set_number = 20000263
# U = load("E:\\U8.npy")
# print(U[:3], V[:3])
# V = load('E:\V8.npy')
for it in range(k):
rt = 1 / (it + 1)
print(str(it) + '次迭代！')
cf.show_time()
dU = np.zeros((n, d))
dB = np.zeros((m, d))
sum1 = 0
i = 0
dV = np.zeros((m, d))
while sum1 < sum_set_number:
j = c[sum1]
if c[sum1] in list_in_order[:]:
T = np.dot(U[i], V[j].T)
dV[j] = -2 * U[i] * (ratings[sum1] - T)
dU[i] = dU[i] + (-2) * V[j] * (ratings[sum1] - T)
sum1 += 1
if sum1 < sum_set_number:
try:
test = r[sum1]
except Exception as er:
print(er, test)
break
if r[sum1] != i:
i += 1
dB = dB + dV
dV = np.zeros((m, d))
else:
dB = dB + dV
dV = dB / n
dU /= n
U = U - rt * (dU + 2 * lu * U)
V = V - rt * (dV + 2 * lv * V)
cf.show_time()
print('\n' + f'I have done it! {sum11}')
np.save(f'E:\\U8_{sum11}', U)
np.save(f'E:\V8_{sum11}', V)
def Hua(r, c, ratings, lu, lv, sub_index, U, V, eta, m, n, d, index, k):
# lu = 10 ** -8
# lv = 10 ** -8
# k = 10
"""
U = np.random.rand(n, d)
V = np.random.rand(m, d)
np.save(index + 'u_4', U)
np.save(index + 'v_4', V)
U = np.load(index + 'u_4.npy')
V = np.load(index + 'v_4.npy')
"""
sum_ratings = len(r)
index = index + "hua\\"
# k = 1
# rt = (2 ** -5)
for it in range(k):
# rt = 1 / (it + 1)
rt_u = 1 / (it + 1)
rt_v = 1 / (it + 1)
# rt_u = (2 ** -5)
# rt_v = rt_u
print(str(it) + ' iterations！')
cf.show_time()
dU = np.zeros((n, d))
dV = np.zeros((m, d))
sum1 = 0
i = 0
while sum1 < sum_ratings:
j = int(c[sum1])
T = np.dot(U[i], V[j].T)
v_grant = -2 * U[i] * (ratings[sum1] - T)
dV[j] += v_grant
dU[i] = dU[i] + (-2) * V[j] * (ratings[sum1] - T)
click = sum1 + 1
if click < sum_ratings:
try:
test = r[click]
except Exception as er:
print(er, test)
break
if r[click] != i:
# dV = np.zeros((m, d))
i += 1
sum1 += 1
dV = dV + np.random.laplace(0, 10 * (d**0.5) / eta,
(m, d)) * 200 * (4 / 3)
dV = dV / n
# dV = dV + np.random.laplace(0, 2 / eta, (m, d))
dU /= n
U = U - rt_u * (dU + 2 * lu * U)
V = V - rt_v * (dV + 2 * lv * V)
cf.show_time()
print('\n' + f'I have done it! {sub_index} Hua')
np.save(index + f'U_4{sub_index}', U)
np.save(index + f'V_4{sub_index}', V)
def getTopK(u, v, n, index):
cf.show_time()
user_top_items = np.zeros((n, 10))
for i in range(n):
user_score_0 = np.dot(u[i], v.T)
list_0 = np.argsort(-user_score_0)
user_top_items[i] = list_0[:10]
np.save(index + 'user_top_items', user_top_items)
cf.show_time()
return user_top_items
def machine_learning_LDP(sum11, U, V):
r = np.load('E:\book\\n.npy')
c = np.load('E:\book\m.npy')
ratings = np.load('E:\book\\r.npy')
# D = np.load('E:\book\D.npy')
# D3 = np.load('E:\book\D3.npy')
m = 131262
n = 138493
d = 15
lu = 10**-8
lv = 10**-8
k = 10
sum = 20000263
n_number = 1000
print('n_number', n_number)
# U = load("E:\\U8.npy")
# print(U[:3], V[:3])
# V = load('E:\V8.npy')
for it in range(k):
rt = 1 / (it + 1)
print(str(it) + '次迭代！')
cf.show_time()
dU = np.zeros((n, d))
dB = np.zeros((m, d))
sum1 = 0
i = 0
dV = np.zeros((m, d))
while sum1 < sum:
j = c[sum1]
T = np.dot(U[i], V[j].T)
dV[j] = -2 * U[i] * (ratings[sum1] - T)
dU[i] = dU[i] + (-2) * V[j] * (ratings[sum1] - T)
sum1 += 1
if sum1 < sum:
try:
test = r[sum1]
except Exception as er:
print(er, test)
break
if r[sum1] != i:
i += 1
dB = dB + dV
# laplace_n = np.random.laplace(0, 10 * 15 / sum11, (m, d))
dV = np.zeros((m, d))
else:
dB = dB + dV
dV = dB / n
dU /= n
U = U - rt * (dU + 2 * lu * U)
V = V - rt * (dV + 2 * lv * V)
cf.show_time()
print('\n' + f'I have done it! {sum11}')
np.save(f'E:\\U9_{sum11}', U)
np.save(f'E:\V9_{sum11}', V)
def get_frency_mf(sum11, U, V):
r = np.load('D:\Labor\\n.npy')
c = np.load('D:\Labor\m.npy')
ratings = np.load('D:\Labor\\r.npy')
m = 131262
order_list = np.load('D:\RR_order_list.npy')
# n = 138493
n = 138493
d = 15
lu = 10 ** -8
lv = 10 ** -8
k = 10
sum = 20000263
n_number = 1000
# U = load("D:\\U8.npy")
# print(U[:3], V[:3])
# V = load('D:\V8.npy')
for it in range(k):
rt = 1 / (it + 1)
print(str(it) + '次迭代！')
common_function.show_time()
dU = np.zeros((n, d))
dB = np.zeros((m, d))
sum1 = 0
i = 0
dV = np.zeros((m, d))
while sum1 < sum:
j = c[sum1]
T = np.dot(U[i], V[j].T)
dV[j] = -2 * U[i] * (ratings[sum1] - T)
dU[i] = dU[i] + (-2) * V[j] * (ratings[sum1] - T)
sum1 += 1
if sum1 < sum:
try:
test = r[sum1]
except Exception as er:
print(er, test)
break
if r[sum1] != i:
i += 1
dB = dB + dV
dV = np.zeros((m, d))
else:
dB = dB + dV
dV = dB / n
dU /= n
U = U - rt * (dU + 2 * lu * U)
V = V - rt * (dV + 2 * lv * V)
common_function.show_time()
print('\n' + 'I have done it!')
np.save(f'D:\\U{sum11}', U)
np.save(f'D:\V{sum11}', V)
def f_score(user_len, user_top_items, u1, v1):
cf.show_time()
f_score_sum = 0
for i in range(user_len):
user_score_1 = np.dot(u1[i], v1.T)
list_1 = np.argsort(-user_score_1)
temp_count = 0
for j in range(10):
if list_1[j] in user_top_items[i]:
temp_count += 1
f_score_sum += temp_count / 10
f_score_final = f_score_sum / user_len
print(f_score_final)
cf.show_time()
return f_score_final
def machine_learning(m, n, d, sum11, U, V, k):
r = np.load('E:\\movielens\\n.npy')
c = np.load('E:\\movielens\\m.npy')
# c = np.load('E:\\movielens\column_list.npy')
ratings = np.load('E:\\movielens\\r.npy')
sum_ratings = len(ratings)
lu = 10**-8
lv = 10**-8
# k = 10
# k = 1
for it in range(k):
rt = 1 / (it + 1)
# 验证实验
# rt = 1 / (it+1) / (k ** 2)
print(str(it) + '次迭代！')
cf.show_time()
dU = np.zeros((n, d))
dB = np.zeros((m, d))
dV = np.zeros((m, d))
sum1 = 0
i = 0
while sum1 < sum_ratings:
j = c[sum1]
T = np.dot(U[i], V[j].T)
dV[j] = -2 * U[i] * (ratings[sum1] - T)
dU[i] = dU[i] + (-2) * V[j] * (ratings[sum1] - T)
sum2 = sum1 + 1
if sum2 < sum_ratings:
try:
test = r[sum2]
except Exception as er:
print(er, test)
break
if r[sum2] != i:
dB = dB + dV
dV = np.zeros((m, d))
i += 1
else:
dB = dB + dV
sum1 += 1
dV = dB / n
dU /= n
U = U - rt * (dU + 2 * lu * U)
V = V - rt * (dV + 2 * lv * V)
cf.show_time()
print('\n' + f'I have done it! {sum11}')
np.save(f'E:\\movielens\\U_0{sum11}', U)
np.save(f'E:\\movielens\\V_0{sum11}', V)
def zh_mf_without_third_part(sum11, U, V, ratings, tipss):
m = np.alen(ratings[0])
# n = 138493
n = np.alen(ratings)
d = 15
lu = 10 ** -8
lv = 10 ** -8
k = 10
if tipss == 0 or tipss == 1:
s = np.load('D:\RR_order_list.npy')
# print(s)
# can = 15000 * (15 ** 0.5)
# n_number = 1000
# rr_order_list = load('D:\RR_order_list.npy')
# lens = len(rr_order_list)
# U = load("D:\\U8.npy")
# print(U[:3], V[:3])
# V = load('D:\V8.npy')
for it in range(1):
rt = 1 / (it + 1)
print(str(it) + '次迭代！')
common_function.show_time()
dU = np.zeros((n, d))
dV = np.zeros((m, d))
for i in range(n):
if tipss == 3 or tipss == 2:
s = np.random.randint(0, m, 5000)
if tipss == 1 or tipss == 3:
laplace_noisy = np.random.laplace(0, 10 * 15 * 5000 / (0.2 * (2 * tipss + 1)), (5000, d))
for number_tipsss in range(5000):
dV[int(s[number_tipsss])] += laplace_noisy[number_tipsss]
for o in range(5000):
j = int(s[o])
if ratings[i][j] != 0:
T = np.dot(U[i], V[j].T)
dV[j] = dV[j] + (-2) * U[i] * (ratings[i][j] - T)
dU[i] = dU[i] + (-2) * V[j] * (ratings[i][j] - T)
dV /= n
if tipss == 0 or tipss == 2:
laplace_noisy = np.zeros((5000, d))
dU /= n
U = U - rt * (dU + 2 * lu * U)
V = V - rt * (dV + 2 * lv * V)
common_function.show_time()
print('\n' + 'I have done it!')
np.save(f'D:\\U{tipss}' + f'_{sum11}', U)
np.save(f'D:\V{tipss}' + f'_{sum11}', V)
def machine_learning_LDP_NJ(sum11, U, V, eta, m, n, sum, d):
r = np.load('E:\\movielens\\n.npy')
c = np.load('E:\\movielens\\m.npy')
# c = np.load('E:\\movielens\column_list.npy')
ratings = np.load('E:\\movielens\\r.npy')
lu = 10**-8
lv = 10**-8
k = 10
# k = 1
for it in range(k):
# rt = 1 / (it + 1)
rt = 1 / (it + 1) / (k**2)
print(str(it) + '次迭代！')
cf.show_time()
dU = np.zeros((n, d))
dB = np.zeros((m, d))
dV = np.zeros((m, d))
sum1 = 0
i = 0
while sum1 < sum:
j = c[sum1]
T = np.dot(U[i], V[j].T)
dV[j] = -2 * U[i] * (ratings[sum1] - T)
dU[i] = dU[i] + (-2) * V[j] * (ratings[sum1] - T)
sum1 += 1
if sum1 < sum:
try:
test = r[sum1]
except Exception as er:
print(er, test)
break
if r[sum1] != i:
i += 1
dB = dB + dV
dV = np.zeros((m, d))
else:
dB = dB + dV
dV = dB / n
dV = dV + np.random.laplace(0, 10 * (d**0.5) / eta, (m, d))
dU /= n
U = U - rt * (dU + 2 * lu * U)
V = V - rt * (dV + 2 * lv * V)
cf.show_time()
print('\n' + f'I have done it! {sum11}')
np.save(f'E:\\movielens\\U_5{sum11}', U)
np.save(f'E:\\movielens\\V_5{sum11}', V)
def machine_learning(m, n, sum, d, sum11, U, V, k, top_k):
r = np.load('E:\\book\\n.npy')
c = np.load('E:\\book\\m.npy')
# c = np.load('E:\\book\column_list.npy')
ratings = np.load('E:\\book\\r.npy')
lu = 10**-8
lv = 10**-8
# k = 10
# k = 1
for it in range(k):
rt = 1 / (it + 1)
print(str(it) + '次迭代！')
cf.show_time()
dU = np.zeros((n, d))
dB = np.zeros((m, d))
dV = np.zeros((m, d))
sum1 = 0
i = 0
while sum1 < sum and i < top_k:
j = c[sum1]
T = np.dot(U[i], V[j].T)
dV[j] = -2 * U[i] * (ratings[sum1] - T)
dU[i] = dU[i] + (-2) * V[j] * (ratings[sum1] - T)
sum1 += 1
if sum1 < sum and i < top_k:
try:
test = r[sum1]
except Exception as er:
print(er, test)
break
if r[sum1] != i:
i += 1
dB = dB + dV
dV = np.zeros((m, d))
else:
dB = dB + dV
dV = dB / n
dU /= n
U = U - rt * (dU + 2 * lu * U)
V = V - rt * (dV + 2 * lv * V)
cf.show_time()
print('\n' + f'I have done it! {sum11}')
np.save(f'E:\\book\\U1000_11{sum11}', U)
np.save(f'E:\\book\\V1000_11{sum11}', V)
def zh_mf(sum11, U, V, ratings, tipss):
m = np.alen(ratings[0])
# n = 138493
n = np.alen(ratings)
V0 = np.load('D:\V0.npy')
d = 15
lu = 10 ** -8
lv = 10 ** -8
k = 10
if tipss == 0 or tipss == 1:
s = np.load('D:\RR_order_list.npy')
for it in range(k):
rt = 1 / (it + 1)
print(str(it) + '次迭代！')
common_function.show_time()
dU = np.zeros((n, d))
dV = np.zeros((m, d))
for i in range(n):
if tipss == 3 or tipss == 2:
s = np.random.randint(0, m, 5000)
for o in range(5000):
j = int(s[o])
if ratings[i][j] != 0:
T = np.dot(U[i], V[j].T)
dV[j] = dV[j] + (-2) * U[i] * (ratings[i][j] - T) / n
dU[i] = dU[i] + (-2) * V[j] * (ratings[i][j] - T)
if tipss == 1 or tipss == 3:
laplace_noisy = np.random.laplace(0, 10 * 15 * 5000 / (0.2 * (2 * tipss + 1)), (5000, d))
for number_tipsss in range(5000):
dV[int(s[number_tipsss])] += laplace_noisy[number_tipsss] / n
# dV /= n
if tipss == 0 or tipss == 2:
laplace_noisy = np.zeros((5000, d))
dU /= n
U = U - rt * (dU + 2 * lu * U)
V = V - rt * (dV + 2 * lv * V)
common_function.show_time()
print(np.linalg.norm(V - V0, np.inf))
print('\n' + 'I have done it!')
np.save(f'D:\\U{tipss}' + f'_{sum11}', U)
np.save(f'D:\V{tipss}' + f'_{sum11}', V)
def get_rmse(filename_U, filename_V):
common_function.show_time()
U = np.load(filename_U)
V = np.load(filename_V)
r = np.load('E:\Labor\\n.npy')
c = np.load('E:\Labor\m.npy')
ratings = np.load('E:\Labor\\r.npy')
n = len(r)
rmse = 0
i = 0
for m in range(n):
if r[m] >= 138492:
break
i = r[m]
j = c[m]
rmse += ((ratings[m] - np.dot(U[i], V[j].T)) ** 2)
rmse = (rmse / n) ** 0.5
print(rmse)
common_function.show_time()
return rmse
def machine_learning(r, c, ratings, m, n, d, sum11, U, V, k, index):
sum_ratings = len(ratings)
lu = 10**-8
lv = 10**-8
# k = 10
# k = 1
for it in range(k):
rt_u = 1 / (it + 1)
rt_v = 1 / (it + 1)
# rt_v = 1 / (it + 1) / (k ** 2)
# 验证实验
# rt = 1 / (it+1) / (k ** 2)
print(str(it) + '次迭代！')
cf.show_time()
dU = np.zeros((n, d))
dV = np.zeros((m, d))
sum1 = 0
i = 0
while sum1 < sum_ratings:
j = c[sum1]
T = np.dot(U[i], V[j].T)
dV[j] += -2 * U[i] * (ratings[sum1] - T)
dU[i] += (-2) * V[j] * (ratings[sum1] - T)
sum2 = sum1 + 1
if sum2 < sum_ratings:
try:
test = r[sum2]
except Exception as er:
print(er, test)
break
if r[sum2] != i:
i += 1
sum1 += 1
dV = dV / n
dU /= n
U = U - rt_u * (dU + 2 * lu * U)
V = V - rt_v * (dV + 2 * lv * V)
cf.show_time()
print('\n' + f'I have done it! {sum11}')
np.save(index + f'U_0{sum11}', U)
np.save(index + f'V_0{sum11}', V)
def f_score(user_len, user_top_items, u1, v1, list_top_k):
cf.show_time()
f_score_sum = 0
# user_top_items = np.zeros((user_len, 10))
for i in range(user_len):
# user_score_0 = np.dot(u[i], v.T)
user_score_1 = np.dot(u1[i], v1.T)
# list_0 = np.argsort(-user_score_0)
list_1 = np.argsort(-user_score_1)
# list_0_top10 = list_0[:10]
# user_top_items[i] = list_0[:10]
temp_count = 0
for j in range(10):
if list_1[j] in user_top_items[i]:
# if list_1[j] in list_top_k:
temp_count += 1
f_score_sum += temp_count / 10
f_score_final = f_score_sum / user_len
# np.save('E:\\book\\user_top_items', user_top_items)
print(f_score_final)
cf.show_time()
return f_score_final
else:
dB = dB + dV
dV = dB / n
dV = dV + np.random.laplace(0, 10 * (d**0.5) / eta, (m, d))
dU /= n
U = U - rt * (dU + 2 * lu * U)
V = V - rt * (dV + 2 * lv * V)
cf.show_time()
print('\n' + f'I have done it! {sum11}')
np.save(f'E:\\movielens\\U_5{sum11}', U)
np.save(f'E:\\movielens\\V_5{sum11}', V)
if __name__ == '__main__':
cf.show_title()
cf.show_time()
top_k = 138493
k = 10
# eta = 1
experiment = 1
eta_list = [0.05, 0.1, 0.2, 0.4, 0.8, 1.6]
eta = eta_list[3]
U = np.load('E:\\movielens\\U.npy')
V = np.load('E:\\movielens\V.npy')
# machine_learning(0, U, V, top_k)
# machine_learning_LDP_part(0, U, V, top_k, eta, k)
# machine_learning_LDP(experiment, U, V, top_k, eta)
# machine_learning_LDP_hau(0, U, V, top_k, eta)
# machine_learning_LDP_NJ(1, U, V, top_k, eta)
def get_granted(r, c, ratings, m, n, d, sum11, U, V, k, index, eta, rt,
sum111):
index = index + "iR//"
# user_array = np.load(index + f"array_list{sum11}.npy")
user_array = np.load(index + f"array_list0.npy")
sum_ratings = len(ratings)
lu = 10**-8
lv = 10**-8
t_sum = k * d * ((math.exp(eta / k) + 1) / (math.exp(eta / k) - 1))
t_sum_1 = math.exp(eta / k)
# 考虑迭代次数的影响
# k = 10
# k = 1
for it in range(k):
# 考虑不同的学习速率的影响
rt_u = 1 / (it + 1) / rt
rt_v = 1 / (it + 1) / rt
# rt_v = 1 / (it + 1) / (k ** 2)
# 验证实验
# rt = 1 / (it+1) / (k ** 2)
print(str(it) + '次迭代！')
cf.show_time()
dU = np.zeros((n, d))
dV = np.zeros((m, d))
sum1 = 0
i = 0
item_number = np.random.randint(0, 10)
item_index = int(user_array[i][item_number])
click = 0
while sum1 < sum_ratings:
j = c[sum1]
T = np.dot(U[i], V[j].T)
if j == item_index:
click = 1
granted_vetor = -2 * U[i] * (ratings[sum1] - T)
item_number_g = np.random.randint(0, d)
granted_number = granted_vetor[item_number_g]
# 若选择的物品为相关物品，则进行如下梯度值计算
t = granted_number
if t > 1:
t = 1
elif t < -1:
t = -1
T = (t * (t_sum_1 - 1) + t_sum_1 + 1) / (2 * (t_sum_1 + 1))
random_t = np.random.random()
if random_t <= T:
dV[j][item_number_g] += t_sum
else:
dV[j][item_number_g] -= t_sum
# 遍历所有评分，计算并更新个人侧面向量
dU[i] += (-2) * V[j] * (ratings[sum1] - T)
sum2 = sum1 + 1
if sum2 < sum_ratings:
try:
test = r[sum2]
except Exception as er:
print(er, test)
break
if r[sum2] != i:
# 若选择的物品为不相关物品，则进行如下梯度值计算
if click == 0:
item_number_g = np.random.randint(0, d)
T = 1 / 2
random_t = np.random.random()
if random_t <= T:
dV[item_index][item_number_g] += t_sum
else:
dV[item_index][item_number_g] -= t_sum
# 选择下次的物品编号，更新标记值
item_number = np.random.randint(0, 10)
item_index = int(user_array[i][item_number])
click = 0
# 更新用户编号
i += 1
sum1 += 1
dV = dV / n
dU /= n
# 一次迭代结束，更新侧面矩阵
U = U - rt_u * (dU + 2 * lu * U)
V = V - rt_v * (dV + 2 * lv * V)
cf.show_time()
print('\n' + f'I have done it! {sum11}')
# 进行k次迭代后，将侧面矩阵写入磁盘
# 0为学习率为1/i，1为学习率1/k/i，2为学习率1/i/k*eta
np.save(index + f'U_{sum111}{sum11}', U)
np.save(index + f'V_{sum111}{sum11}', V)
def ML_LDP_part(m, n, d, sum11, U, V, eta, k, c_flag, l_array, r_q):
r = np.load('E:\\limbi\\n.npy')
c = np.load('E:\\limbi\\m.npy')
ratings = np.load('E:\\limbi\\r.npy')
D = np.load('E:\\limbi\\Dl1.npy')
D3 = np.load('E:\\limbi\\Dl3.npy')
n, l_items = np.shape(l_array)
q = r_q
lu = 10**-8
lv = 10**-8
# k = 10
sum_ratings = len(r)
eta /= d
t_sum = q * ((math.exp(eta / k) + 1) / (math.exp(eta / k) - 1))
t_s = math.exp(eta / k)
print(l_array[:10][:10], c_flag[:10])
for it in range(k):
# rt = 1 / (it + 1) / (k ** 2)
# rt = 1 / (it + 1)
rt = 1 / (it + 1)
print(str(it) + '次迭代！')
cf.show_time()
dU = np.zeros((n, d))
dB = np.zeros((m, d))
dl = np.zeros((l_items, d))
sum1 = 0
i = 0
dV = np.zeros((m, d))
while sum1 < sum_ratings:
j = c[sum1]
T = np.dot(U[i], V[j].T)
if c_flag[sum1] == 1:
dV[j] = -2 * U[i] * (ratings[sum1] - T)
dU[i] = dU[i] + (-2) * V[j] * (ratings[sum1] - T)
sum2 = sum1 + 1
if sum2 < sum_ratings:
try:
test = r[sum2]
except Exception as er:
print(er, test)
break
if r[sum2] != i:
for o in range(l_items):
tts = int(l_array[i][o])
dl[o] = dV[tts]
x = np.dot(D, dl)
for ls in range(d):
s = np.random.randint(0, q)
t = x[s][ls]
if t > 1:
t = 1
elif t < -1:
t = -1
T = (t * (t_s - 1) + t_s + 1) / (2 * (t_s + 1))
random_t = np.random.random()
if random_t <= T:
x[s][ls] = t_sum
else:
x[s][ls] = -t_sum
x_z = np.dot(D3, x)
for o in range(l_items):
tts = int(l_array[i][o])
dB[tts] += x_z[o]
dV = np.zeros((m, d))
i += 1
else:
for o in range(l_items):
tts = int(l_array[i][o])
dl[o] = dV[tts]
x = np.dot(D, dl)
for ls in range(d):
s = np.random.randint(0, q)
t = x[s][ls]
if t > 1:
t = 1
elif t < -1:
t = -1
T = (t * (t_s - 1) + t_s + 1) / (2 * (t_s + 1))
random_t = np.random.random()
if random_t <= T:
x[s][ls] = t_sum
else:
x[s][ls] = -t_sum
x_z = np.dot(D3, x)
for o in range(l_items):
tts = int(l_array[i][o])
dB[tts] += x_z[o]
dV = np.zeros((m, d))
i += 1
sum1 += 1
dV = dB / n
dU /= n
U = U - rt * (dU + 2 * lu * U)
V = V - rt * (dV + 2 * lv * V)
cf.show_time()
print('\n' + f'I have done it! {sum11}')
np.save(f'E:\\limbi\\U_5{sum11}', U)
np.save(f'E:\\limbi\\V_5{sum11}', V)
def zh_select_GD_DR(n, m, r, c, ratings, d, sum11, U, V, eta, k, index, frency,
flag, flag_v, delt):
sum_ratings = len(ratings)
lu = 10**-8
lv = 10**-8
factor = eta * math.sqrt(n) / ((delt * d * k) * math.sqrt(2 * d))
# k = 10
# k = 1
for it in range(k):
rt_u = 1 / (it + 1)
rt_v = 1 / (it + 1) * factor
# rt_v = 1 / (it + 1) / (k ** 2)
# 验证实验
# rt = 1 / (it+1) / (k ** 2)
print(str(it) + '次迭代！')
cf.show_time()
dU = np.zeros((n, d))
dV = np.zeros((m, d))
sum1 = 0
i = 0
f = 0
# factor = eta / ((delt * d * k) * math.sqrt(2*d))
while sum1 < sum_ratings:
j = c[sum1]
T = np.dot(U[i], V[j].T)
if flag[sum1] == 1 and f == 0:
# factor_temp = factor / math.sqrt(frency[i])
dV[j] += np.random.laplace(0, delt * d * k / eta,
d) - 2 * U[i] * (ratings[sum1] - T)
# if factor_temp < 1:
# dV[j] += (np.random.laplace(0, 2*5*d*k/eta, d) - 2 * U[i] * (ratings[sum1] - T)) * factor_temp
# else:
# dV[j] += -2 * U[i] * (ratings[sum1] - T)
f = 1
dU[i] += (-2) * V[j] * (ratings[sum1] - T)
sum2 = sum1 + 1
if sum2 < sum_ratings:
try:
test = r[sum2]
except Exception as er:
print(er, test)
break
if r[sum2] != i:
ss = int(flag_v[i])
if f == 0:
# factor_temp = factor / math.sqrt(frency[i])
dV[ss] += np.random.laplace(0, delt * d * k / eta, d)
# if factor_temp < 1:
# dV[ss] += (np.random.laplace(0, 2*5*d*k/eta, d)) * factor_temp
# else:
# dV[ss] += np.random.laplace(0, 2*5*d*k/eta, d)
# 可以考虑使用Nuyen提出的算法，单个加噪，另外可以尝试看
f = 0
i += 1
sum1 += 1
dV = dV / n
dU /= n
U = U - rt_u * (dU + 2 * lu * U)
V = V - rt_v * (dV + 2 * lv * V)
cf.show_time()
print('\n' + f'I have done it! {sum11}')
np.save(index + f'U_7{sum11}', U)
np.save(index + f'V_7{sum11}', V)
def machine_learning_LDP_part(m, n, sum, d, sum11, U, V, top_k_user, eta, k):
r = np.load('E:\\movielens\\n.npy')
# c = np.load('E:\\movielens\m.npy')
c = np.load('E:\\movielens\\m_adjust.npy')
ratings = np.load('E:\\movielens\\r.npy')
list_in_order4 = np.load('E:\\movielens\\list_in_order4.npy')
D4 = np.load('E:\\movielens\\D4.npy')
D5 = np.load('E:\\movielens\\D5.npy')
m_d = 5000
# q = 2700
q = 500
lu = 10**-8
lv = 10**-8
t_sum = q * d * ((math.exp(eta / k) + 1) / (math.exp(eta / k) - 1))
for it in range(k):
rt = 1 / (it + 1) / (k**2)
print(str(it) + '次迭代！')
cf.show_time()
dU = np.zeros((n, d))
dB = np.zeros((q, d))
sum1 = 0
i = 0
dV = np.zeros((m, d))
while sum1 < sum and i < top_k_user:
j = c[sum1]
T = np.dot(U[i], V[j].T)
if j >= 0:
dV[j] = -2 * U[i] * (ratings[sum1] - T)
dU[i] = dU[i] + (-2) * V[j] * (ratings[sum1] - T)
sum1 += 1
if sum1 < sum and i < top_k_user:
try:
test = r[sum1]
except Exception as er:
print(er, test)
break
if r[sum1] != i:
i += 1
dv_x = np.zeros((m_d, d))
for o in range(m_d):
dv_x[o] = dV[int(list_in_order4[o])]
x = np.dot(D4, dv_x)
s = np.random.randint(0, q)
l = np.random.randint(0, d)
if x[s][l] > 0:
x[s][l] = t_sum
else:
x[s][l] = -t_sum
dB = dB + x
dV = np.zeros((m, d))
else:
dB = dB + x
dV = np.zeros((m, d))
dv_x = dB / n
dv_xt = np.dot(D5, dv_x)
for o in range(m_d):
dV[int(list_in_order4[o])] = dv_xt[o]
dU /= n
U = U - rt * (dU + 2 * lu * U)
V = V - rt * (dV + 2 * lv * V)
cf.show_time()
print('\n' + f'I have done it! {sum11}')
np.save(f'E:\\movielens\\U_3{sum11}', U)
np.save(f'E:\\movielens\\V_3{sum11}', V)
def ML_LDP_part_prove(m, n, sum, d, sum11, U, V, top_k_user, eta, k):
r = np.load('E:\\book\\n.npy')
c = np.load('E:\\book\m.npy')
c_flag = np.load('E:\\book\\m_adjust.npy')
ratings = np.load('E:\\book\\r.npy')
list_in_order4 = np.load('E:\\book\\list_in_order4.npy')
D4 = np.load('E:\\book\\D4.npy')
D5 = np.load('E:\\book\\D5.npy')
m_d = 5000
# q = 2700
q = 500
lu = 10**-8
lv = 10**-8
for it in range(k):
# rt = 1 / (it + 1) / (k ** 2)
rt = 1 / (it + 1) / k
print(str(it) + '次迭代！')
cf.show_time()
dU = np.zeros((n, d))
dB = np.zeros((q, d))
sum1 = 0
i = 0
dV = np.zeros((m, d))
while sum1 < sum and i < top_k_user:
j = c[sum1]
T = np.dot(U[i], V[j].T)
if c_flag[sum1] >= 0:
dV[j] = -2 * U[i] * (ratings[sum1] - T)
dU[i] = dU[i] + (-2) * V[j] * (ratings[sum1] - T)
sum1 += 1
if sum1 < sum and i < top_k_user:
try:
test = r[sum1]
except Exception as er:
print(er, test)
break
if r[sum1] != i:
i += 1
dv_x = np.zeros((m_d, d))
for o in range(m_d):
dv_x[o] = dV[int(list_in_order4[o])]
x = np.dot(D4, dv_x)
s = np.random.randint(0, q)
x[s] += np.random.laplace(0, 2 * d * k / eta, d)
dB = dB + x
dV = np.zeros((m, d))
else:
dB = dB + x
dV = np.zeros((m, d))
dv_x = dB / n
dv_xt = np.dot(D5, dv_x)
for o in range(m_d):
dV[int(list_in_order4[o])] = dv_xt[o]
dU /= n
U = U - rt * (dU + 2 * lu * U)
V = V - rt * (dV + 2 * lv * V)
cf.show_time()
print('\n' + f'I have done it! {sum11}')
np.save(f'E:\\book\\U1000_7{sum11}', U)
np.save(f'E:\\book\\V1000_7{sum11}', V)
def machine_learning_LDP_hau(sum11, U, V, eta, m, n, sum, d):
r = np.load('E:\\movielens\\n.npy')
c = np.load('E:\\movielens\\m.npy')
ratings = np.load('E:\\movielens\\r.npy')
lu = 10**-8
lv = 10**-8
# k = 10
k = 10
# eta = 0.1
t_sum = m * d * ((math.exp(eta / k) + 1) / (math.exp(eta / k) - 1))
t_sum_1 = math.exp(eta / k)
for it in range(k):
rt = 1 / (it + 1) / (k**2)
# rt = 1 / (it + 1)
print(str(it) + '次迭代！')
cf.show_time()
dU = np.zeros((n, d))
dB = np.zeros((m, d))
sum1 = 0
i = 0
dV = np.zeros((m, d))
while sum1 < sum:
j = c[sum1]
T = np.dot(U[i], V[j].T)
# if j >= 0:
dV[j] = -2 * U[i] * (ratings[sum1] - T)
dU[i] = dU[i] + (-2) * V[j] * (ratings[sum1] - T)
sum1 += 1
if sum1 < sum:
try:
test = r[sum1]
except Exception as er:
print(er, test)
break
if r[sum1] != i:
i += 1
s = np.random.randint(0, m)
ls = np.random.randint(0, d)
t = dV[s][ls]
if t > 1:
t = 1
elif t < -1:
t = -1
T = (t * (t_sum_1 - 1) + t_sum_1 + 1) / (2 * (t_sum_1 + 1))
random_t = np.random.random()
if random_t <= T:
dV[s][ls] = t_sum
else:
dV[s][ls] = -t_sum
dB = dB + dV
dV = np.zeros((m, d))
else:
dB = dB + dV
dV = dB / n
dU /= n
# dV_t = np.dot(D, dV)
# dV = np.dot(D3, dV_t)
U = U - rt * (dU + 2 * lu * U)
V = V - rt * (dV + 2 * lv * V)
cf.show_time()
print('\n' + f'I have done it! {sum11}')
np.save(f'E:\\movielens\\U_4{sum11}', U)
np.save(f'E:\\movielens\\V_4{sum11}', V)
def d_limbi(m_number, n_number):
'''
m = np.load('E:\\limbi\\m.npy')
n = np.load('E:\\limbi\\n.npy')
r = np.load('E:\\limbi\\r.npy')
m_array = np.zeros((3, len(m)))
m_array[0] = n[:]
m_array[1] = m[:]
m_array[2] = r[:]
m_list = np.zeros((1, m_number))
for i in range(len(m)):
j = m[i]
m_list[0][j] += 1
long_list = []
for i in range(m_number):
if m_list[0][i] > 140:
long_list.append(i)
print('get score is over 140')
t = len(m)
m_a = []
n_a = []
r_a = []
for i in range(t):
if m[i] in long_list:
m_a.append(m[i])
n_a.append(n[i])
r_a.append(r[i])
print('get the new set')
np.save('E:\\limbi\\n', n_a)
np.save('E:\\limbi\\m', m_a)
np.save('E:\\limbi\\r', r_a)
'''
# test = np.zeros((20, 30))
# print(test)
'''
data = np.zeros((n_number, m_number), np.float16)
n_a = np.load('E:\\limbi\\n.npy')
m_a = np.load('E:\\limbi\\m.npy')
r_a = np.load('E:\\limbi\\r.npy')
t = len(n_a)
for i in range(t):
s_t = int(n_a[i])
l_t = int(m_a[i])
data[s_t][l_t] = r_a[i]
print('set the old dateset')
cf.show_time()
n_a = []
m_a = []
r_a = []
m_click = 0
for i in range(m_number):
s_t = 0
for j in range(n_number):
ratings_s = data[j][i]
if ratings_s != 0:
s_t += 1
n_a.append(j)
m_a.append(m_click)
r_a.append(ratings_s)
if s_t != 0:
m_click += 1
print(m_click)
np.save('E:\\limbi\\n', n_a)
np.save('E:\\limbi\\m', m_a)
np.save('E:\\limbi\\r', r_a)
print('order the dataset')
'''
n_a = np.load('E:\\limbi\\n.npy')
m_a = np.load('E:\\limbi\\m.npy')
r_a = np.load('E:\\limbi\\r.npy')
t = len(n_a)
m_number = 26509
cf.show_time()
data = np.zeros((n_number, m_number))
for i in range(t):
s_t = int(n_a[i])
l_t = int(m_a[i])
data[s_t][l_t] = r_a[i]
print('get new dataset')
cf.show_time()
n_a = []
m_a = []
r_a = []
for i in range(n_number):
for j in range(m_number):
ratings_s = data[i][j]
if ratings_s != 0:
n_a.append(i)
m_a.append(j)
r_a.append(ratings_s)
print(i)
print('get the order list')
cf.show_time()
'''
m_adjust = []
m_adjust.append(m_a[0])
for i in range(len(m_a)):
if m_a[i] not in m_adjust:
m_adjust.append(m_a[i])
np.save('E:\\limbi\\m_adjust', m_adjust)
print('Get new m order')
cf.show_time()
for i in range(len(m_a)):
for j in range(len(m_adjust)):
if m_a[i] == m_adjust[j]:
m_a[i] = j
print(j)
break
cf.show_time()
print('order the m')
n_adjust = [0]
ss = 0
for i in range(len(n_a)-1):
j = i + 1
if n_a[j] == n_a[i]:
n_adjust.append(ss)
else:
ss += 1
n_adjust.append(ss)
'''
print('order the n')
np.save('E:\\limbi\\n', n_a)
np.save('E:\\limbi\\m', m_a)
np.save('E:\\limbi\\r', r_a)
def machine_learning_LDP(m, n, sum, d, sum11, U, V, top_k, eta):
r = np.load('E:\\movielens\\n.npy')
c = np.load('E:\\movielens\m.npy')
ratings = np.load('E:\\movielens\\r.npy')
D = np.load('E:\\movielens\D.npy')
D3 = np.load('E:\\movielens\D3.npy')
# m = 131262
# n = 138493
# d = 15
q = 2700
lu = 10**-8
lv = 10**-8
# k = 10
k = 10
# sum = 20000263
# eta = 0.1
t_sum = q * d * ((math.exp(eta / k) + 1) / (math.exp(eta / k) - 1))
t_sum_1 = math.exp(eta / k)
for it in range(k):
rt = 1 / (it + 1) / (k**2)
print(str(it) + '次迭代！')
cf.show_time()
dU = np.zeros((n, d))
dB = np.zeros((q, d))
sum1 = 0
i = 0
dV = np.zeros((m, d))
while sum1 < sum and i < top_k:
j = c[sum1]
T = np.dot(U[i], V[j].T)
# if j >= 0:
dV[j] = -2 * U[i] * (ratings[sum1] - T)
dU[i] = dU[i] + (-2) * V[j] * (ratings[sum1] - T)
sum1 += 1
if sum1 < sum and i < top_k:
try:
test = r[sum1]
except Exception as er:
print(er, test)
break
if r[sum1] != i:
i += 1
x = np.dot(D, dV)
s = np.random.randint(0, q)
ls = np.random.randint(0, d)
t = x[s][ls]
if t > 1:
t = 1
elif t < -1:
t = -1
T = (t * (t_sum_1 - 1) + t_sum_1 + 1) / (2 * (t_sum_1 + 1))
random_t = np.random.random()
if random_t <= T:
x[s][ls] = t_sum
else:
x[s][ls] = -t_sum
dB = dB + x
dV = np.zeros((m, d))
else:
dB = dB + x
dV = dB / n
dU /= n
# dV_t = np.dot(D, dV)
# dV = np.dot(D3, dV_t)
U = U - rt * (dU + 2 * lu * U)
V = V - rt * (np.dot(D3, dV) + 2 * lv * V)
cf.show_time()
print('\n' + f'I have done it! {sum11}')
np.save(f'E:\\movielens\\U_2{sum11}', U)
np.save(f'E:\\movielens\\V_2{sum11}', V)
def s_items(m, n, d, sum11, U, V, eta, k, l_i):
r = np.load('E:\\limbi\\n.npy')
c = np.load('E:\\limbi\\m.npy')
t_sum = math.exp(eta / (2 * l_i))
sum_ratings = len(r)
cf.show_time()
sum1 = 0
user_items = []
sum_items = []
for ii in range(m):
sum_items.append(ii)
other_items = sum_items[:]
s_i = []
l_array = np.zeros((n, l_i))
i = 0
while sum1 < sum_ratings and i < n:
j = c[sum1]
user_items.append(j)
other_items.remove(j)
sum2 = sum1 + 1
if sum2 < sum_ratings and i < n:
try:
test = r[sum1]
except Exception as er:
print(er, test)
break
if r[sum2] != i:
s = len(user_items)
if s >= l_i:
if s > l_i:
for o in range(s - l_i):
l_o = len(user_items)
rand_number = np.random.randint(0, l_o - 1)
del user_items[rand_number]
else:
for o in range(l_i - s):
l_o = len(other_items)
tt = np.random.randint(0, l_o - 1)
user_items.append(other_items[tt])
del other_items[tt]
for o in range(l_i):
rand_number = np.random.random()
if rand_number <= t_sum / (t_sum + 1):
s_i.append(user_items[o])
# l_other = s_r(0, m-l_i-1, l_i-len(s_i))
for o in range(l_i - len(s_i)):
l_o = len(other_items)
tt = np.random.randint(0, l_o - 1)
s_i.append(other_items[tt])
del other_items[tt]
s_i.sort()
l_array[i] = s_i
i += 1
other_items = sum_items[:]
s_i = []
user_items = []
sum1 += 1
cf.show_time()
print(l_array[0])
print('\n' + f'I have done it! {sum11}')
np.save(f'E:\\limbi\\l_array{sum11}', l_array)
