How do I plot data in a single figure instead of 2 in Tensorboard generated using tf.summary.scalar() on metrics procured from a custom callback?

Question

I have used the following code to plot 2 scalar plots in Tensorboard on metrics procured from Custom ROC-AUC score callback.

for epoch in range(len(history.history['roc_train'])):
    with train_summary_writer.as_default():
        tf.summary.scalar('roc_auc_train', history.history['roc_train'][epoch], step=epoch)
        tf.summary.scalar('roc_auc_val', history.history['roc_val'][epoch], step=epoch)

plots in Tensorboard:

How do I get them both in the same figure? Please help.

Answer 1

Create two different file_writer instances, and using the same name (roc_auc in this example) call the with statement on the context of the desired plot.

import tensorflow as tf
import numpy as np
import os
import time

now = time.localtime()
subdir = time.strftime("%d-%b-%Y_%H.%M.%S", now)

summary_train = os.path.join("stackoverflow", subdir, "train")
summary_val = os.path.join("stackoverflow", subdir, "val")

writer_train = tf.summary.create_file_writer(summary_train)
writer_val = tf.summary.create_file_writer(summary_val)

for epoch in range(200):
    with writer_train.as_default():
        tf.summary.scalar(name="roc_auc", data=-np.math.exp(-(epoch/30))+1, step=epoch)
    with writer_val.as_default():
        tf.summary.scalar(name="roc_auc", data=-np.math.exp(-(epoch/15))+1, step=epoch)