I am using a seq2seq encoder-decoder neural net with attention to summarize text, but I think it may be overfitting. https://github.com/dongjun-Lee/text-summarization-tensorflow
Is there a way to plot loss vs num_epochs for both the training and validation set? Is this an interpretable measure?
From my understanding ROUGE score is necessary because multiple different summaries can be considered a correct answer, so does that mean you cannot use loss on the validation set vs num_epochs as a measure for over/underfitting? Are there any other ways to check for over/under fitting?