1

I am running an LSTM network on textual data. The code uses the Keras library from python3. I would like to know what it means to have a validation loss that is greater than one. The loss used is categorical_crossentropy

I am trying to make sense of the validation loss which seems to be greater than 1.0 in my case. The other metrics are as follows: acc: 0.6104 - val_loss: 1.0719 - val_acc: 0.7046

Here is the learning curve, in case it helps. LSTM-Acc

Rahul Krishnan
  • 43
  • 2
  • 11
  • https://stackoverflow.com/questions/34518656/how-to-interpret-loss-and-accuracy-for-a-machine-learning-model – PV8 Jun 11 '19 at 09:06
  • In contrast with measures like the accuracy, loss is no bounded in [0, 1], and there is nothing strange with it being greater than 1. Own answer in [Loss & accuracy - Are these reasonable learning curves?](https://stackoverflow.com/questions/47817424/loss-accuracy-are-these-reasonable-learning-curves/47819022#47819022) might be helpful. – desertnaut Jun 11 '19 at 09:11
  • true. not LSTM specific ) actualy it is only a layer, not a network), loss value is just a marker what is bigger if error bigger is. Interesting what do you call an "accuracy" ?? – user8426627 Jun 11 '19 at 09:15
  • 1
    Thanks, that answers what I was looking for. @desertnaut – Rahul Krishnan Jun 11 '19 at 09:17

0 Answers0