Recommenders: [ASK] [HELP] NCF Notebook train loss graph

[karthikraja95]

I am using NCF deep dive notebook, I am just wondering is there any way to plot the train loss graph ? is there any inbuilt function to plot this? If not can you tell me how to do it?

Any help is appreciated.

Thanks
K

[loomlike]

You can use tf.summary.scalar() to write your metrics to tensorboard. Here you can find simple example.

For NCF in our repo, you can add summary writing part to:

def fit(self, data):
...
            # output every self.verbose
            if self.verbose and epoch_count % self.verbose == 0:
                logger.info(
                    "Epoch %d [%.2fs]: train_loss = %.6f "
                    % (epoch_count, train_time, sum(train_loss) / len(train_loss))
                )
                # ... write mean(train_loss) into summary writer here ...

We do have a logging hook at reco_utils/common/tf_utils.py which are being used for wide-deep model, but it is for tf.Estimator API. You can refer the module to see how the summary writer is used.

[yueguoguo]

@karthikraja95 does it help solve your problem? If so, please close the issue.

[karthikraja95]

I am trying to use implement it. I will let you know the results. Thanks