def Update(self, R, k, r, lamb1, lamb2, aerfa, aerfa1):
'''
:param R: user-item matrix
:param k: The number of iterations
:param r: r-rank factors
:param lamb1: used to calculate U
:param lamb2: used to calculate V
:param aerfa:
:param aerfa1:
:return:
'''
print("R:\n", R)
I = copy.deepcopy(R)
I[I > 0] = 1
m, n = R.shape
U = np.array(np.random.random((r, m)), dtype='float64')
V = np.array(np.random.random((r, n)), dtype='float64')
#这里可以通过KNN找到user的朋友矩阵
simiX = trainW(R)
W = myKNN(simiX, 5)
# print("w:",W)
# updating formulas
for i in range(k):
# U
for i_u in range(m):
subU1 = np.zeros((r, 1), dtype='float64')
for j_u in range(n):
# print(np.array(U[:,i_u].T).shape, np.array(V[:,j_u]).shape)
# print(U[:,j_u].T)
# print(V[:,j_u])
# print(I[i_u][j_u])
subU1 = subU1 + (I[i_u][j_u] *
(np.dot(U[:, i_u].T, V[:, j_u]) -
R[i_u][j_u])) * V[:, j_u]
subU1 = subU1 + lamb1 * U[:, i_u]
subU2on = np.zeros((r, 1))
Fri = np.argwhere(W[i_u] == 1)
# print(len(Fri))
for f in range(len(Fri)):
# print(Fri[f][0])
# print("i_u\t",i_u)
# print("Fri[%d][0]\t" % f,Fri[f][0])
# print(U[:,Fri[f][0]].shape)
# print("asdasdd:\n",subU2on)
subU2on = subU2on + (metrices.VectorSpaceSimilarity(
R, I, i_u, Fri[f][0])) * (
np.array(U[:, i_u], dtype='float64') -
np.array(U[:, Fri[f][0]], dtype='float64'))
# print("subU2on:\n", subU2on)
subU2 = aerfa * subU2on[0]
subU = subU1 + 2 * subU2
# print("subU1:\n",subU1)
# print("subU2:\n",subU2)
# print("subU:\n",subU)
del subU1
del subU2
U[:, i_u] = U[:, i_u] - aerfa1 * np.array(subU[0])
#V
for i_v in range(n):
subV = np.zeros((1, r))
for j_v in range(m):
subV = subV + (I[j_v][i_v] *
(np.dot(U[:, j_v].T, V[:, i_v]) -
R[j_v][i_v])) * U[:, j_v]
subV = subV + lamb2 * V[:, i_v]
# print("subV:\n", subV)
V[:, i_v] = V[:, i_v] - aerfa1 * subV[0]
print("run%d" % i)
return U, V
def Update(self, R, k, r, lamb1, lamb2, aerfa, aerfa1):
'''
:param R: user-item matrix
:param k: The number of iterations
:param r: r-rank factors
:param lamb1: used to calculate U
:param lamb2: used to calculate V
:param aerfa:
:return:
'''
print("R:\n", R)
I = copy.copy(R)
I[I > 0] = 1
m, n = R.shape
U = np.array(np.random.random((r, m)), dtype='float64')
V = np.array(np.random.random((r, n)), dtype='float64')
#这里可以通过KNN找到user的朋友矩阵
simiX = trainW(R)
W = myKNN(simiX, 5)
# print("w:",W)
# updating formulas
for i in range(k):
# U
for i_u in range(m):
subU1 = np.zeros((r, 1), dtype='float64')
for j_u in range(n):
# print(np.array(U[:,i_u].T).shape, np.array(V[:,j_u]).shape)
# print(U[:,j_u].T)
# print(V[:,j_u])
# print(I[i_u][j_u])
subU1 = subU1 + (I[i_u][j_u] *
(np.dot(U[:, i_u].T, V[:, j_u]) -
R[i_u][j_u])) * V[:, j_u]
subU1 = subU1 + lamb1 * U[:, i_u]
subU2on = np.zeros((r, 1))
subU2down = 0
Fri = np.argwhere(W[i_u] == 1)
# print(len(Fri))
for f in range(len(Fri)):
# print(Fri[f][0])
# print("i_u\t",i_u)
# print("Fri[%d][0]\t" % f,Fri[f][0])
# print(U[:,Fri[f][0]].shape)
# print("asdasdd:\n",subU2on)
subU2on = np.round(
subU2on + (metrices.VectorSpaceSimilarity(
R, I, i_u, Fri[f][0])) * np.array(U[:, Fri[f][0]]),
6)
# print("asd",U[:,Fri[f][0]])
# print("")
subU2down = np.round(
subU2down +
metrices.VectorSpaceSimilarity(R, I, i_u, Fri[f][0]),
6)
# print("subU2on:\n", subU2on)
# print("subU2down:\n", subU2down)
subU2 = aerfa * np.array(U[:, i_u] - (subU2on[0] / subU2down))
subU3 = np.zeros((r, 1))
subU3on = np.zeros((r, 1))
subU3down = 0
subU3onon = np.zeros((r, 1))
subU3downdown = 0
for g in range(len(Fri)):
# for f in range(len(Fri)):
# subU3onon = subU3onon + (metrices.VectorSpaceSimilarity(R, I, Fri[g][0], Fri[f][0])) * U[:,Fri[f][0]]
# subU3downdown = subU3downdown + metrices.VectorSpaceSimilarity(R, I, Fri[g][0], Fri[f][0])
# subU3down = subU3down + metrices.VectorSpaceSimilarity(R, I, Fri[g][0], Fri[f][0])
subU3on = np.round(
subU3on +
-(metrices.VectorSpaceSimilarity(R, I, i_u, Fri[g][0])
* np.array(U[:, g] - (subU2on[0] / subU2down))), 6)
subU3 = subU3 + (subU3on / subU2down)
# print("subU3 run\t\t:",g)
subU3 = aerfa * np.array(subU3)
subU = subU1 + subU2 + subU3
# print("subU1:\n",subU1)
# print("subU2:\n",subU2)
# print("subU3:\n",subU3)
# print("subU:\n",subU)
U[:, i_u] = U[:, i_u] - aerfa1 * np.array(subU[0])
#V
for i_v in range(n):
subV = np.zeros((1, r))
for j_v in range(m):
uv = np.dot(U[:, j_v].T, V[:, i_v]) - R[j_v][i_v]
uv1 = I[j_v][i_v] * uv
subV = np.round(subV + uv1 * U[:, j_v], 6)
subV = subV + lamb2 * V[:, i_v]
# print("subV:\n", subV)
V[:, i_v] = V[:, i_v] - aerfa1 * subV[0]
print("run%d" % i)
return U, V