def log(self,Xk,yk,Zk,muk,wk,dt,logpath="."):
#主双対内点法による処理を記録
"""
mu_ | 主変数xと双対変数zの平均値xT*z/N
w_ | 双対ギャップcT*x - bT*y
cx_ | 主問題の目的関数cT*x
by_ | 双対問題の目的関数bT*y
dt | 1反復分の計算時間
x_ | k反復目の主変数xk
y_ | k反復目の主変数yk
z_ | k反復目の主変数zk
"""
mu_= muk.reshape(1,muk.size)[0]
w_ = wk.reshape(1,wk.size)[0]
cx_= np.trace(np.dot(self.C,Xk.transpose()))
by_= np.dot(self.b,yk)
x_ = Xk.reshape(1,Xk.size)[0]
y_ = yk.reshape(1,yk.size)[0]
z_ = Zk.reshape(1,Zk.size)[0]
print(mu_,w_,cx_,by_,dt,x_,y_,z_)
d_ = np.hstack([mu_,w_,cx_,by_,dt,x_,y_,z_])
savecsv(d_,logpath)
return
with tf.Session() as sess:
sess.run(init)
xW, xb = sess.run([W, b])
print("init weight:{} \nbias:{}".format(xW, xb))
# train
BATCHES = math.ceil(len(real_X) / BATCH_SIZE)
for e in range(EPOCHES):
for i in range(BATCHES):
bX = real_X[(i * BATCH_SIZE):(i + 1) * BATCH_SIZE]
bY = real_Y[(i * BATCH_SIZE):(i + 1) * BATCH_SIZE]
sess.run(train, feed_dict={X: bX, Y: bY})
j = e * BATCHES + i
if j % 50 == 0 or (j == EPOCHES * BATCHES - 1 and j % 50 != 0):
xW, xb = sess.run([W, b])
print(
"I,E,B:{:0>4d},{:0>4d},{:0>4d} weight:{} \nbias:{}".format(
j, e, i, xW, xb))
savecsv("{}-{:0>4d}-{}.csv".format(csvprefix, j, "W"), xW)
savecsv("{}-{:0>4d}-{}.csv".format(csvprefix, j, "b"), xb)
# predict
Y_pred = sess.run(pred_Y, feed_dict={X: real_X, Y: real_Y})
print("Y_pred:", Y_pred)
savecsv("{}-pred-{}.csv".format(csvprefix, "Y"), Y_pred)
# save real y for evaluation
savecsv("{}-real-{}.csv".format(csvprefix, "Y"), real_Y)
with tf.Session() as sess:
sess.run(init)
rW, rb = sess.run([reveal_W, reveal_b])
print("init weight:{} \nbias:{}".format(rW, rb))
# train
BATCHES = math.ceil(len(real_X) / BATCH_SIZE)
for e in range(EPOCHES):
for i in range(BATCHES):
bX = real_X[(i * BATCH_SIZE):(i + 1) * BATCH_SIZE]
bY = real_Y[(i * BATCH_SIZE):(i + 1) * BATCH_SIZE]
sess.run(train, feed_dict={X: bX, Y: bY})
j = e * BATCHES + i
if j % 50 == 0 or (j == EPOCHES * BATCHES - 1 and j % 50 != 0):
rW, rb = sess.run([reveal_W, reveal_b])
print(
"I,E,B:{:0>4d},{:0>4d},{:0>4d} weight:{} \nbias:{}".format(
j, e, i, rW, rb))
if mpc_player_id == 0:
savecsv("{}-{:0>4d}-{}.csv".format(csvprefix, j, "W"),
rW.astype(float))
savecsv("{}-{:0>4d}-{}.csv".format(csvprefix, j, "b"),
rb.astype(float))
# predict
Y_pred = sess.run(reveal_Y, feed_dict={X: real_X, Y: real_Y})
if mpc_player_id == 0:
print("Y_pred:", Y_pred)
savecsv("{}-pred-{}.csv".format(csvprefix, "Y"), Y_pred.astype(float))
print(init)
# #############################################################
# save to csv for comparing, for debug
scriptname = os.path.basename(sys.argv[0]).split(".")[0]
#csvprefix = "/root/x/000/cb/example/linear_regression/log/" + scriptname
csvprefix = "./log/" + scriptname
os.makedirs(csvprefix, exist_ok=True)
csvprefix = csvprefix + "/tf"
# #############################################################
with tf.Session() as sess:
sess.run(init)
xW, xb = sess.run([W, b])
print("init weight:{} \nbias:{}".format(xW, xb))
for i in range(EPOCHES):
sess.run(train)
if i % 10 == 0 or (i == EPOCHES-1 and i % 10 != 0):
xW, xb = sess.run([W, b])
print("i:{:0>4d} weight:{} \nbias:{}".format(i, xW, xb))
savecsv("{}-{:0>4d}-{}.csv".format(csvprefix, i, "W"), xW)
savecsv("{}-{:0>4d}-{}.csv".format(csvprefix, i, "b"), xb)
# predict
Y_pred = sess.run(pred_Y)
print("Y_pred:", Y_pred)
savecsv("{}-{:0>4d}-{}.csv".format(csvprefix, EPOCHES, "Y"), Y_pred)
# #############################################################
with tf.Session() as sess:
sess.run(init)
rW, rb = sess.run([reveal_W, reveal_b])
print("init weight:{} \nbias:{}".format(rW, rb))
# train
BATCHES = math.ceil(len(real_X) / BATCH_SIZE)
for e in range(EPOCHES):
for i in range(BATCHES):
bX = real_X[(i * BATCH_SIZE):(i + 1) * BATCH_SIZE]
bY = real_Y[(i * BATCH_SIZE):(i + 1) * BATCH_SIZE]
sess.run(train, feed_dict={X: bX, Y: bY})
j = e * BATCHES + i
if j % 50 == 0 or (j == EPOCHES * BATCHES - 1 and j % 50 != 0):
rW, rb = sess.run([reveal_W, reveal_b])
print(
"I,E,B:{:0>4d},{:0>4d},{:0>4d} weight:{} \nbias:{}".format(
j, e, i, rW, rb))
if rtt.mpc_player.id == 0:
savecsv("{}-{:0>4d}-{}.csv".format(csvprefix, j, "W"), rW)
savecsv("{}-{:0>4d}-{}.csv".format(csvprefix, j, "b"), rb)
# predict
Y_pred = sess.run(reveal_Y, feed_dict={X: real_X, Y: real_Y})
if rtt.mpc_player.id == 0:
print("Y_pred:", Y_pred)
savecsv("{}-pred-{}.csv".format(csvprefix, "Y"), Y_pred)
# 1-Y, (1 - Y) * tf.log(1 - sig_v)],
# feed_dict={X:batch_attr, Y: batch_label})
# if i <= 2:
# print("logits:", logit_v, "\n sig:", sig_pv1, "and \n", sig_pv2)
# print("logv1:", log_v1, "\n logv2:", log_v2, "\n ty:", ty, "\n loss:", curr_loss)
avg_loss += curr_loss / total_batch
if (epoch + 1) % DIS_STEP == 0:
print("****Epoch:", epoch+1)
#print(avg_loss)
cW, cb = native_sess.run([W, b])
print("curr_W:\n", cW)
print("curr_b:\n", cb)
#print("i:{:0>4d} weight:{} \nbias:{}".format(epoch+1, cW, cb))
savecsv("{}-{:0>4d}-{}.csv".format(save_file_prefix, epoch+1, "W"), cW)
savecsv("{}-{:0>4d}-{}.csv".format(save_file_prefix, epoch+1, "b"), cb)
#print(f"Epoch: {epoch:5}, loss={float(avg_loss):20.9}")
end_t = time.time()
print("************ Native Trainning Finished!**************")
print("The training cost: ", end_t - start_t, "seconds")
print("W:", native_sess.run(W))
print("b:", native_sess.run(b))
print("************ Native Fitting on TRAIN dataset**************")
is_pass_all = True
self_check_pred = tf.sigmoid(tf.matmul(X,W) + b)
real_label = Y