仅当测试输入值在火车输入值范围内时，使用回归DL才能获得良好的预测结果
我尝试过minmax和标准缩放器都没有帮助实现改进，我这样做是针对两个回归系统：仅继续值以及继续和类别输入数据的混合，两个系统的结果相似。
import pandas as pd
import numpy as np
import tensorflow as tf
from keras import models
from keras.models import Sequential
from keras.layers import Dense, Dropout
from keras import optimizers
from sklearn.model_selection import KFold
from sklearn.preprocessing import MinMaxScaler
tf.reset_default_graph()
from keras.optimizers import SGD
import pickle
siz=100000
inp0=np.random.randint(100, 1000000 , size=(siz,3))
rand0=np.random.randint(-100, 100 , size=(siz,2))
a1=0.2;a2=0.8;a3=2.5;a4=2.6;a5=1.2;a6=0.3
oup1=np.dot(inp0[:,0],a1)+np.dot(inp0[:,1],a2)+np.dot(inp0[:,2],a3)\
+rand0[:,0]
oup2=np.dot(inp0[:,0],a4)+np.dot(inp0[:,1],a5)+np.dot(inp0[:,2],a6)\
+rand0[:,1]
oup0=np.concatenate((oup1.reshape(siz,1),
oup2.reshape(siz,1)),axis=1)
normzer_inp = MinMaxScaler()
inp_norm = normzer_inp.fit_transform(inp0)
inp_scale = 'normzer_inp_1.sav'
pickle.dump(normzer_inp, open(inp_scale, 'wb'))
normzer_oup = MinMaxScaler()
oup_norm = normzer_oup.fit_transform(oup0)
oup_scale = 'normzer_oup_1.sav'
pickle.dump(normzer_oup, open(oup_scale, 'wb'))
X=inp_norm
Y=oup_norm
opti_SGD = SGD(lr=0.1, momentum=0.9)
model = Sequential()
model.add(Dense(16, input_dim=X.shape[1], activation='relu'))
model.add(Dense(16, activation='relu'))
model.add(Dense(Y.shape[1], activation='linear'))
model.compile(loss='mean_squared_error', optimizer=opti_SGD)
history = model.fit(X, Y, epochs=100,batch_size=2048, verbose=2)
model.save('DeepL_SelfTst_ContCtg_toCont_5.h5')
siz=10
inpt=np.random.randint(100, 1000000 , size=(siz,3))
randt=np.random.randint(-100, 100 , size=(siz,2))
oupt1=np.dot(inpt[:,0],a1)+np.dot(inpt[:,1],a2)\
+np.dot(inpt[:,2],a3)+randt[:,0]
oupt2=np.dot(inpt[:,0],a4)+np.dot(inpt[:,1],a5)\
+np.dot(inpt[:,2],a6)+randt[:,1]
oupt0=np.concatenate((oupt1.reshape(siz,1),
oupt2.reshape(siz,1)),axis=1)
inp_scale_2 = 'normzer_inp_1.sav'
inp_norm_do = pickle.load(open(inp_scale_2, 'rb'))
inp_norm_t = inp_norm_do.transform(inpt)
oup_scale_2 = 'normzer_oup_1.sav'
oup_norm_do = pickle.load(open(oup_scale_2, 'rb'))
new_model = models.load_model('DeepL_SelfTst_ContCtg_toCont_5.h5')
predictions = new_model.predict(inp_norm_t)
pred_real=oup_norm_do.inverse_transform(predictions)
pre_sult=pred_real.copy()
pre_sult=np.concatenate((pre_sult, oupt0), axis=1)
df = pd.DataFrame({'pred 1': pre_sult[:, 0],\
'real 1': pre_sult[:, 2],\
'pred 2': pre_sult[:, 1],'real 2': pre_sult[:, 3],})
pred 1 real 1 pred 2 real 2
0 2.889692e+06 2889419.3 2.048063e+06 2048469.7
1 2.374380e+06 2374681.2 2.684326e+06 2683353.8
2 2.754795e+06 2754609.5 1.796634e+06 1796445.9
3 4.967822e+05 496659.8 1.372401e+06 1372884.8
4 1.785036e+06 1785088.7 2.686990e+06 2687372.3
5 8.210452e+05 821230.4 1.121044e+06 1120332.6
6 2.124294e+06 2124235.6 9.910031e+05 990209.6
7 1.994700e+06 1995039.0 2.210463e+06 2209663.8
8 3.204532e+06 3204814.4 3.456536e+06 3455951.6
9 2.889887e+06 2889767.8 1.388561e+06 1388345.4
inpt=np.random.randint(1, 100 , size=(siz,3))
pred 1 real 1 pred 2 real 2
0 49198.234375 -23.2 267707.34375 297.0
1 49355.171875 186.6 267697.28125 114.8
2 49330.609375 220.5 267652.18750 302.1
3 49194.960938 104.3 267697.53125 203.5
4 49245.285156 76.0 267710.71875 87.8
5 49286.746094 141.3 267733.25000 55.7
6 49195.906250 153.1 267700.15625 232.9
7 49348.496094 162.7 267646.90625 404.5
8 49375.519531 262.6 267653.40625 101.8
9 49396.093750 275.7 267686.25000 267.5
我如何训练模型更加灵活，或采用其他任何方法解决该问题？