TensorFlow: How to merge multiple 'collections'?

Question

I have some collections that I would like to track with TensorBoard using a supervisor. In the Supervisor initializer I would like something to the effect

summary_op = tf.summary.merge_all(['test', 'valid'])

But I get the error TypeError: unhashable type: 'list', because the key must be a string (see documentation).

---

Edit:

This doesn't work either:

summary_op = [tf.summary.merge_all('train'), tf.summary.merge_all('valid')]

Answer 1

Try tf.summary.merge(), e.g. like so:

summary_op = tf.summary.merge([
        tf.summary.merge_all('test'),
        tf.summary.merge_all('train')],
    collections='merged')

This would merge all summaries from the test and train collections and add them to a new merged collection. Keep in mind that this will result in strange effects if the same summary is used multiple times during the same time step:

Here I was accidentally (manually!) storing validation summaries during training runs and then again in a separate validation run.

Also I'm not sure if this is the most efficient way to go about it, but it certainly works.