卷积神经网络预测函数出来的值一模一样是什么问题？

卷积神经网络预测函数出来的值一模一样是什么问题？

预测值predict一列下来一模一样，是过拟合呢还是欠拟合？太小白纠结了一个月了，求大神指导

`
1.测试的代码：

import joblib
import numpy as np
from com.lcg.version11 import model
from com.lcg.version11 import Loading_pictures as lp

# dnn = joblib.load("D:/python/data/model/model1.model")  # 加载模型
# dnn = joblib.load("D:/python/data/model/model2.model")  # 加载模型
dnn = joblib.load("D:/python/data/model/model3.model")  # 加载模型

# 加载数据集
A, B = lp.loaddata("D:/python/data/testdata")
predictB = dnn.BatchPredict(A)

print("predictB=", predictB)
predictBB = np.array([np.argmax(one_hot) for one_hot in predictB])
print("predictBB=", predictBB)
realB = B
print("realB",realB)
realBB = np.array([np.argmax(one_hot) for one_hot in realB])
print("realBB", realBB)
from sklearn.metrics import accuracy_score

# accuracy_score(predictYY, realYY)

print("准确数=",accuracy_score(predictBB, realBB, normalize=False))
print("准确率=",accuracy_score(predictBB, realBB))

2.训练的代码：

#---------训练1--------------
    dnn = md.DNN()
    # dnn.Add(CNN2D(3,3,2,10))
    # 卷积
    dnn.Add(md.CNN2D_MultiLayer(4, 4, 2, 10))
    # 池化
    dnn.Add(md.DMaxPooling2D(2, 2))
    yy = dnn.Forward(X[0])
    dnn.Add(md.CNN2D_MultiLayer(3, 3, 1, 5))
    yy2 = dnn.Forward(X[0])
    dnn.Add(md.DFlatten())
    # yy=dnn.Forward(X[0])
    dnn.Add(md.DDense(10, 'relu', bFixRange=False))
    dnn.Add(md.DDense(1))
    # dnn.Add(DDense(1,'relu'))
    # yy=dnn.BatchPredict(X)
    dnn.Compile()
    # ratio=dnn.AdjustWeightByComputingRatio(X,Y)
    # print(ratio)
    dnn.Fit(X, Y, 50)
    # 将模型持久化保存
    joblib.dump(dnn, "D:/python/data/model/model3.model")

3.包含预测函数的dnn模型：

class DNN:

    def __init__(self):

        self.layers = []

    # 在前一层的基础上继续追加层
    def Add(self, layer):

        self.layers.append(layer)

    def Forward(self, X):

        if myDebug == 1:
            print('dnn forward', X.shape)

        nL = len(self.layers)

        y = X

        for i in range(nL):
            y = self.layers[i].Forward(y)
        return y

        # 批预测
    def BatchPredict(self, X):

        self.predictY = []

        for k in range(X.shape[0]):
            self.predictY.append(self.Forward(X[k]))
            # print(np.array(self.predictY))

        self.predictY = np.array(self.predictY)

        return self.predictY

    def Compile(self, lossMethod='MSE'):
        # self.lossModel=Entropy()

        if lossMethod == 'MSE':
            self.lossModel = MSE()

        if lossMethod == 'CrossEntropy':
            self.lossModel = CrossEntropy()

        if lossMethod == 'SoftmaxCrossEntropy':
            self.lossModel = SoftmaxCrossEntropy()

    def FitOneRound(self, X, Y, iRound, epochs):

        loss = 0

        nL = len(self.layers)

        for k in range(X.shape[0]):

            y = self.Forward(X[k])

            loss += self.lossModel.loss(nx=y, ny=Y[k])

            dy = self.lossModel.backward()

            # print('k=',k,'y=',y,'realY=',Y[k],'dy=',dy)
            # dy=np.min(-1,np.max(dy,1))

            # print('type y=',type(y),'shape y',y.shape)

            # if  (np.linalg.norm(dy,1)>1):
            if (y.shape[0] == 1) and (np.linalg.norm(dy, 1) > 1):

                if y > Y[k]:
                    dy = 1
                else:
                    dy = -1

            step = 0.75 * (epochs - iRound) / epochs + 0.01

            # if iRound%10==0:
            #     print('iter',iteration,'k= ',k,' dy=',dy,'y=',y,'Yk=',Y[k])

            for i in range(nL):
                dy = self.layers[nL - i - 1].Backward(dy)

                self.layers[nL - i - 1].Learn()

        if (iRound % (int(epochs / 10)) == 0) or (iRound == epochs - 1):
            print('round=', iRound, 'loss=', loss)