2分类val_loss变化巨大的原因?

keras做图像2分类,lr=0.000000001,optimizer是rmsprop,batch size=32,y_pred>0.1定义为正
第一次跑的结果,:
Epoch 1/5
81/81 [==============================] - 1s 8ms/step - loss: 14.5646 - acc: 0.0864
Epoch 2/5
81/81 [==============================] - 0s 6ms/step - loss: 14.5646 - acc: 0.0864
Epoch 3/5
81/81 [==============================] - 0s 6ms/step - loss: 14.5646 - acc: 0.0864
Epoch 4/5
81/81 [==============================] - 0s 5ms/step - loss: 14.5646 - acc: 0.0864
Epoch 5/5
81/81 [==============================] - 0s 6ms/step - loss: 14.5646 - acc: 0.0864
混淆矩阵:
[[ 0 36]
[ 0 3]]

第2次跑的结果:
Epoch 1/5
81/81 [==============================] - 1s 7ms/step - loss: 3.4956 - acc: 0.4074
Epoch 2/5
81/81 [==============================] - 1s 7ms/step - loss: 1.3929 - acc: 0.9136
Epoch 3/5
81/81 [==============================] - 1s 6ms/step - loss: 1.3929 - acc: 0.9136
Epoch 4/5
81/81 [==============================] - 0s 5ms/step - loss: 1.3929 - acc: 0.9136
Epoch 5/5
81/81 [==============================] - 0s 5ms/step - loss: 1.3929 - acc: 0.9136
混淆矩阵
[[36 0]
[ 3 0]]

2次跑的结果差距这么大的原因?

查看全部
weixin_44347319
C医生
2019/04/03 10:26
  • 机器学习
  • 深度学习
  • 人工智能
  • 神经网络
  • 数据挖掘
  • 点赞
  • 收藏
  • 回答
    私信

0个回复