def matrix_factorisation(R, P, Q, K, steps=5000, alpha=0.0002, beta=0.02):
global X
global Y
Q = Q.T
for step in range(steps):
for i in range(len(R)):
for j in range(len(R[i])):
if R[i][j] > 0:
eij = R[i][j] - numpy.dot(P[i, :], Q[:, j])
for k in range(K):
P[i][k] = P[i][k] + alpha * (2 * eij * Q[k][j] -
beta * P[i][k])
Q[k][j] = Q[k][j] + alpha * (2 * eij * P[i][k] -
beta * Q[k][j])
eR = numpy.dot(P, Q)
e = 0
t = 0
for i in range(len(R)):
for j in range(len(R[i])):
if R[i][j] > 0:
e = e + pow(R[i][j] - numpy.dot(P[i, :], Q[:, j]), 2)
for k in range(K):
t = t + (beta / 2) * (pow(P[i][k], 2) +
pow(Q[k][j], 2))
e = e + t
print 'zui:', t
print 'loss', e
if (step > 200):
X.append(step)
Y.append(e)
if step > 4000:
figure.paint1(X, Y)
break
return P, Q.T
def start_sample_matrix(ratio, lambdas, lr, table_name, l):
MAE = 0
RMSE = 0
start_time = time.strftime('%Y-%m-%d-%H-%M-%S',
time.localtime(time.time()))
log.start_log(
start_time, "../matrix/" + table_name.decode('utf-8') + "/" +
"MF矩阵分解结果.txt".decode('utf-8'))
f = log.write_log()
lc_table_name = 'lc_' + table_name
tp_table_name = 'tp_' + table_name
# start预测部分
# # 得到的矩阵是挖取过值的矩阵
# C, original_matrix, changed_zero = dm.getMatrix(table_name, ratio)
C, original_matrix, changed_zero = dm.get_Matrix_from_lc_tp(
lc_table_name, tp_table_name, ratio, 1)
# C = np.array(C)
d = C.shape
U = np.random.rand(d[0], l)
V = np.random.rand(d[1], l)
print "开始矩阵分解"
matrix, X, Y = simple_mat.matrix_factorization(C, U, V, lambdas, step, lr)
# 开始验证
print "开始验证"
matrix0, pre_or_mat, num = de.norma_matrix(matrix, original_matrix)
MAE, RMSE = vali.validate(matrix, original_matrix, changed_zero)
# #end
# end预测部分
file_path = "../matrix/" + table_name.decode('utf-8')
t = str(ratio) + "_" + str(num) + "_" + start_time + ".txt"
# start将矩阵分解后的矩阵保存
np.savetxt(file_path + "/matrix_factorization/MF_matrix_factorization_" +
str(ratio) + ".txt",
matrix,
fmt='%.8f')
# end将矩阵分解后的矩阵保存
# start将原矩阵经预测填充后的矩阵保存
np.savetxt(file_path + "/result/MF_pre_" + str(ratio) + ".txt",
pre_or_mat,
fmt='%.8f')
# end 将原矩阵经预测填充后的矩阵保存
# start将矩阵分解后的矩阵(处理过的，负数变0)保存
np.savetxt(file_path + "/out/MF_" + t.decode('utf-8'), matrix0, fmt='%.8f')
# end 将矩阵分解后的矩阵(处理过的，负数变0)保存
# end
end_time = time.strftime('%Y-%m-%d-%H-%M-%S', time.localtime(time.time()))
print >> f, "开始时间：", start_time
print >> f, "结束时间：", end_time
# 显示梯度下降情况
title = 'lr:{0} alpha:{1} beta:{2} step:{3} lambdas:{4} sim:{5}'
title = title.format(lr, ab[0], ab[1], step, lambdas, simlambda)
print >> f, "参数：", title
figure_path = "../matrix/" + table_name.decode(
'utf-8') + "/figure/MF_" + str(ratio) + "_" + start_time + ".jpg"
figure.paint1(X, Y, title, figure_path)
log.close_log()
return MAE, RMSE
def matrix_factorization(C, U, V, alpha, beta, lamdas, steps, lr,sk):
global X
global Y
X = []
Y = []
f = log.write_log()
step = 0
loss1, def_v, all_sim_u, all_sim_v = getloss(C, U, V, alpha, beta, lamdas,sk)
while step<=steps:
U, V = gardient(C, U, V, def_v, all_sim_u, all_sim_v, alpha, beta, lamdas, lr)
loss2, def_v, all_sim_u, all_sim_v = getloss(C, U, V, alpha, beta, lamdas,sk)
#用于显示梯度下降的效果
X.append(step)
Y.append(loss2)
if(step%250==0):
figure.paint1(X,Y)
print >>f,'loss:',loss1
if(loss1-loss2<0.001):
break;
loss1 = loss2
step = step +1
def getresult(l, table_name, lambdas, ab, simlambda, lr, step, sim_k, ratio):
MAE = 0
RMSE = 0
alpha = ab[0]
beta = ab[1]
start_time = time.strftime('%Y-%m-%d-%H-%M-%S',
time.localtime(time.time()))
log.start_log(
start_time, "../matrix/" + table_name.decode('utf-8') + "/" +
"矩阵分解结果.txt".decode('utf-8'))
f = log.write_log()
lc_table_name = 'lc_' + table_name
tp_table_name = 'tp_' + table_name
# start预测部分
# # 得到的矩阵是挖取过值的矩阵
# C, original_matrix, changed_zero = dm.getMatrix(tp_table_name, ratio)
C, original_matrix, changed_zero = dm.get_Matrix_from_lc_tp(
lc_table_name, tp_table_name, ratio, 1)
# C = np.array(C)
d = C.shape
U = np.random.rand(d[0], l)
V = np.random.rand(d[1], l)
print "开始矩阵分解"
matrix, X, Y, loss = de.matrix_factorization(C, U, V, lambdas, step, alpha,
beta, simlambda, lr, sim_k,
tp_table_name)
# 开始验证
print "开始验证"
matrix0, pre_or_mat, num = de.norma_matrix(matrix, original_matrix)
MAE, RMSE = vali.validate(matrix, original_matrix, changed_zero)
# #end
# end预测部分
file_path = "../matrix/" + table_name.decode('utf-8')
t = str(ratio) + "_" + str(num) + "_" + start_time + ".txt"
# start将矩阵分解后的矩阵保存
np.savetxt(file_path + "/matrix_factorization/matrix_factorization_" +
str(ratio) + ".txt",
matrix,
fmt='%.8f')
# end将矩阵分解后的矩阵保存
# start将原矩阵经预测填充后的矩阵保存
# filematrix0 = open(file_path + "/result/pre_" + str(ratio) + "." + "txt", 'w');
# filematrix0.close()
np.savetxt(file_path + "/result/pre_" + str(ratio) + ".txt",
pre_or_mat,
fmt='%.8f')
# end 将原矩阵经预测填充后的矩阵保存
# start将矩阵分解后的矩阵(处理过的，负数变0)保存
# filematrix1 = open(file_path + "/out/" + t.decode('utf-8'), 'w');
# filematrix1.close()
np.savetxt(file_path + "/out/" + t.decode('utf-8'), matrix0, fmt='%.8f')
# end 将矩阵分解后的矩阵(处理过的，负数变0)保存
# end
# k, disease, mitrax, table_name
# top_k = dv.getTopK(k, disease, matrix, table_name)
# 筛选医院，得到与目标人物处在同一个城市的医院
# pre_top = dv.pre_data(top_k)
# 筛选医院
# matrix = dealtxt.load_file_to_array("../matrix/" + table_name.decode('utf-8') + "/result/pre_" + str(ratio) + ".txt");
# filter_hos = dv.filter_hos_By_sorted100(table_name,matrix,disease,people_locat)
# dv.getResult(pre_top,disease)
# dv.getResult1(disease,matrix,table_name,city,people_locat)
end_time = time.strftime('%Y-%m-%d-%H-%M-%S', time.localtime(time.time()))
print >> f, "开始时间：", start_time
print >> f, "结束时间：", end_time
# 显示梯度下降情况
title = 'lr:{0} alpha:{1} beta:{2} step:{3} lambdas:{4} sim:{5} sim_k:{6}'
title = title.format(lr, ab[0], ab[1], step, lambdas, simlambda, sim_k)
print >> f, "参数：", title
figure_path = "../matrix/" + table_name.decode('utf-8') + "/figure/" + str(
ratio) + "_" + end_time + ".jpg"
figure.paint1(X, Y, title, figure_path)
log.close_log()
return MAE, RMSE, loss
def showgard(title, file_path):
global X
global Y
print "图"
figure.paint1(X, Y, title, file_path)
def showgard(x, y, title, file_path):
"""
显示梯度下降情况
"""
print "图"
figure.paint1(x, y, title, file_path)