利用tensorflow2.0组建了神经网络，代码如下：

 		datax=self.data.values.astype('float64')  #是一个(55, 3)的dataframe
        datay=self.result.values.astype('float64') #是一个(55,)的dataframe
        datax=tf.expand_dims(datax,-1)
        model_tf=keras.Sequential()
        layers1=keras.layers.Dense(10,activation='relu')
        layers2=keras.layers.Dense(5,activation='relu')
        layers3=keras.layers.LSTM(5)
        layers4=keras.layers.Dropout(0.3)
        output=keras.layers.Dense(1)
        model_tf=keras.Sequential([layers1,layers2,layers3,layers4,output])
        model_tf.compile(optimizer=tf.keras.optimizers.Adam(),loss="mse")
        model_tf.fit(datax,datay,epochs=5000)

训练后，loss没有下降，最后几个loss，如下：

Epoch 4997/5000
2/2 [==============================] - 0s 2ms/step - loss: 1455905.5000

Epoch 4998/5000
2/2 [==============================] - 0s 1ms/step - loss: 1455843.7500

Epoch 4999/5000
2/2 [==============================] - 0s 2ms/step - loss: 1458519.7500

Epoch 5000/5000
2/2 [==============================] - 0s 1ms/step - loss: 1453022.7500

请问一下各位大神，这可能是什么原因？