scikit-learn MLPRegressor预测结果总是变化和不一致

我正在尝试使用MLPRegressor训练和测试我的数据集 . 我有两个数据集（训练数据集和测试数据集），它们都具有完全相同的特征和标签列 . 这是我的数据集的示例：

训练数据集：

Full,Gold Standard
1.176,3.571
4.231,3.467
3.75,4.333
3.519,3.5
1.154,2.833
3.2,3.643
2.692,3.4
3.611,2.733
4.0,2.393
2.727,1.933
3.529,3.423
2.647,2.733
1.538,2.786
2.0,2.967
2.647,2.533
1.786,2.552
5.0,5.0
3.158,4.6
1.875,2.733

测试数据集：

Full,Gold Standard
1.667,2.345
3.056,1.9
1.765,2.2
0.714,0.0
1.538,2.586
2.188,1.667
3.333,2.8
2.5,2.481
1.667,2.433
1.842,0.0
2.381,0.793
0.588,1.0
1.176,1.433
1.538,2.3
0.588,1.655
0.909,2.333
0.833,3.333
1.111,2.5
0.0,2.067

这是我的代码：

import csv
import numpy as np 
import random
import os.path

from sklearn import preprocessing as pre
from sklearn.neural_network import MLPRegressor

with open('FullFeatures2017.csv') as f:
    reader = csv.reader(f)
    next(reader, None)
    training = list(reader)
f.closed
#print("Row number of data training : ", len(training))

print

#------------training-------------------
train_data = [list(map(float, training[i])) for i in range(len(training))]
data1 = np.array(train_data)
print("Row number of training data : ", len(train_data))
X_train = data1[:, :-1]
y_train = data1[:, -1:]

scaler = pre.StandardScaler()
X_train_scaled = scaler.fit_transform(X_train)

print

#------------open csv data testing-------------------
with open('FullFeatures2016.csv') as f:
    reader = csv.reader(f)
    next(reader, None)
    testing = list(reader)
f.closed
#print("Row number data testing : ", len(testing))

print

#------------testing-------------------
test_data = [list(map(float, testing[i])) for i in range(len(testing))]
data2 = np.array(test_data)
print("Row number of testing data : ", len(test_data))
X_test = data2[:, :-1]
y_test = data2[:, -1:]

X_test_scaled = scaler.fit_transform(X_test)

print

#------------Model Training-------------------
mlp = MLPRegressor(max_iter=500, learning_rate_init=0.1, random_state=1, solver='lbfgs', tol=0.001 )
y_train2 = np.ravel(y_train)
mlp.fit(X_train_scaled, y_train2)
#print(mlp.fit(X_train, y_train2))

print

#------------Model Testing or Prediction-------------------
prediction = mlp.predict(X_test_scaled)
print len(prediction)
print
print prediction
print

它工作但每次运行程序时结果总是会改变 . 我给了random_state数字，如1,2或3（不是0），但仍然总是改变 .

谁知道如何使用MLPRegressor预测相同且一致的结果？

谢谢