Loss takes a negative value

I installed GluonTS from source code on 2019/10/2.

Hello, everyone.

I am trying to use DeepAREstimator for multivariate time series forecasting.

INFO:root:Epoch[49] Elapsed time 10.459 seconds
INFO:root:Epoch[49] Evaluation metric 'epoch_loss'=-3.829468
INFO:root:Loading parameters from best epoch (49)
INFO:root:Final loss: -3.8294676637649534 (occurred at epoch 49)
INFO:root:End model training

Loss takes a negative value.
Is this correct?

Yes for the DeepAR Estimator on gluon-ts it’s possible for the loss to be negative.

Hi sad, thanks so much for your reply.

Is the smaller the better?
Or should ‘loss’ be close to zero?