有人在Keras / Tensorflow模型训练中看到此问题，它显示RuntimeWarning: Mean of empty slice.，然后损失计算显示 nan ，然后准确性下降，无法恢复。
---这是日志输出-----
Epoch 9701/9701 [==============================] - 1s 111us/step - loss: 0.0830 - acc: 0.9790 - val_loss: 6.6692 - val_acc: 0.3525
Epoch 63/1000
9701/9701 [==============================] - 1s 113us/step - loss: 0.0824 - acc: 0.9784 - val_loss: 6.8110 - val_acc: 0.3469
Epoch 64/1000
/home/anatolii/.local/lib/python3.6/site-packages/numpy/core/fromnumeric.py:2920: **RuntimeWarning: Mean of empty slice**.
out=out, **kwargs)
/home/anatolii/.local/lib/python3.6/site-packages/numpy/core/_methods.py:85: RuntimeWarning: invalid value encountered in true_divide
ret = ret.dtype.type(ret / rcount)
9701/9701 [==============================] - 1s 56us/step - loss: **nan** - acc: 0.9488 - val_loss: 15.7643 - val_acc: **0.0000e+00**
Epoch 65/1000
9701/9701 [==============================] - 0s 44us/step - loss: **nan** - acc: 0.9349 - val_loss: 15.9566 - val_acc: **0.0000e+00**
Epoch 66/1000
任何帮助将不胜感激。