神经网络没有预测正确的数字

Question

我是 ML 的新手，对于我的项目，我正在尝试使用神经网络制作数字分类器。我制作了一个 GUI，您可以在其中绘制数字，它将 NumPy 数组传递给神经网络。我用 mnist 数字数据集 训练了我的神经网络，模型准确率为 97.70%，但它无法预测输入的数字。

#CODE FOR NEURAL NETWORK
    class mltest():
       def __init__(self):
          self.model = keras.Sequential([  keras.layers.Dense(120,input_shape = (784,),activation = 'relu'),
                                           keras.layers.Dense(10,activation = 'softmax')])
       def train(self):   
          (x_train,y_train),(x_test,y_test) = keras.datasets.mnist.load_data()
          x_train = x_train/255
          x_test = x_test/255
          x_train = x_train.reshape(len(x_train),28*28)
          x_test = x_test.reshape(len(x_test),28*28)
          
                                  
          self.model.compile(optimizer='adam',loss='SparseCategoricalCrossentropy',metrics=['accuracy'])
          self.model.fit(x_train,y_train,epochs=12,batch_size=200)
          self.model.evaluate(x_test,y_test)
          
       def test(self,value):
           y_predicted = self.model.predict(value)
           print(np.argmax(y_predicted))
       
       if __name__ =='__main__':
           obj = mltest()
           obj.train()

Epoch 12/12
300/300 [==============================] - 1s 2ms/step - loss: 0.0338 - accuracy: 0.9908
313/313 [==============================] - 0s 776us/step - loss: 0.0763 - accuracy: 0.9770

GUI 代码

class Window(QMainWindow):
    def __init__(self):
        super().__init__()
  
        self.setWindowTitle("Ai")
  
        self.setGeometry(300,300,300,300)
  
        self.image = QImage(self.size(), QImage.Format_RGB32)

        self.image.fill(Qt.white)
  
        self.drawing = False

        self.brushSize = 25

        self.brushColor = Qt.black

        self.lastPoint = QPoint()

        self.object = mltest()

#To send the picture data to neural network when enter key is pressed
    def keyPressEvent(self, event):
        if event.key() ==  16777220:
            screen = QtWidgets.QApplication.primaryScreen()
            screenshot = screen.grabWindow(QtWidgets.QWidget.winId(self))
            bufffer = QBuffer()
            bufffer.open(bufffer.ReadWrite)
            screenshot = screenshot.save(bufffer,'PNG')
            image = Image.open(io.BytesIO(bufffer.data()))
            image = image.convert('P')
            image = np.array(image)
           
            image =cv2.resize(image,dsize = (28,28))

            image = cv2.blur(image,(2,2))
            image = cv2.bitwise_not(image)
            image[image<=30] = 0 
            print(image)
             
            plt.matshow(image)
            plt.show()
            image = image.reshape(1,28*28)
            image = image/255
           
            self.object.test(image)
 #To draw           
    def mousePressEvent(self, event):
        if event.button() == Qt.LeftButton:
            

            self.drawing = True
            self.lastPoint = event.pos()

    def mouseMoveEvent(self, event):

        if (event.buttons() & Qt.LeftButton) & self.drawing:
              
        
            painter = QPainter(self.image)
            painter.setPen(QPen(self.brushColor, self.brushSize, 
                            Qt.SolidLine, Qt.RoundCap, Qt.RoundJoin))

            painter.drawLine(self.lastPoint, event.pos())
            
            self.lastPoint = event.pos()
        
            self.update()


    def mouseReleaseEvent(self, event):
  
        if event.button() == Qt.LeftButton:
            self.drawing = False
  
    def paintEvent(self, event):
        canvasPainter = QPainter(self)
        canvasPainter.drawImage(self.rect(), self.image, self.image.rect())

      
    if __name__ == '__main__':
        App = QApplication(sys.argv)
        window = Window()
        window.show()
        sys.exit(App.exec())

predicted value is 5

Answer 1

您显然在两个文件中使用了 if __name__ =='__main__':，并且您仅在调用网络文件时训练网络，而当您启动 GUI 应用程序时，您的网络是在 self.object = mltest() 中未经训练创建的。除非你调用 self.object.train()，否则它可能仍未经过训练，因此无法做出好的预测

神经网络没有预测正确的数字

Neural Network is not predicting the correct digit

python

neural-network

mnist

keras

tensorflow