Open
Description
Should we add some logging of loss, acc during the training process? As far as I know, we would only output the final loss or auc at the end of training. But some users may want to know the loss and acc per N batches, so that they can decide whether the training process works well or the models are suitable for their data (some models may consume lots of time to train. It is better to kill the training job if the model is extremely bad).