search

google / uis-rnn

This is the library for the Unbounded Interleaved-State Recurrent Neural Network (UIS-RNN) algorithm, corresponding to the paper Fully Supervised Speaker Diarization.
https://arxiv.org/abs/1810.04719
Apache License 2.0
1.55k stars 320 forks source link

Interpreting the loss values #12

Closed dalonlobo closed 5 years ago

dalonlobo commented 5 years ago

Hi Team,

I'm training the model on a custom dataset. I'm finding it confusing to interpret the various losses displayed during training. For example, does having large negative Training loss is better or should I concentrate on Negative log likelihood? To sum up, how will I know that the model is converging? Following image is from my training process.

image

Your help is greatly appreciated.

wq2012 commented 5 years ago

You could plot the training loss with matplotlib to see whether it's converging or not. We didn't add it by default because we don't want to have too many dependencies.

But in general, smaller value of loss means better model.