What is fraction_of_32_full in TensorFlow
As you can see in this figure, my graph has: "fraction_of_32_full" output attached to it. I have not done this explicitly, and nowhere in my code have I set any size limit of 32.
The only data I explicitly add to my resume is the cost of each batch, but when I look at the TensorBoard render I see this:
As you can see, it contains three things. The cost I asked for and two other variables I asked for. The share is 25,000 full, and the share is 32 full.
My questions:
- What is it?
- Why are they being added to my resumes without an explicit request?
source to share
I can actually answer my own question. I did some digging and found the answers.
- What is it?
These are indicators of how full your queues are. I have two queues, a line input queue that reads my files, and a batch queue that participates in my writes. Both entries are in the format: fraction of x full, where x is the capacity of the queue.
The reason the input string producer is 32 fraction is because if you look here you will see the default capacity is 32.
- Why are they added to my resumes without explicit request.
It was a little tricky. If you look at the source code for the input string producer here , you can see that although input_string_producer does not explicitly prompt for a short name, it returns input_producer which has a default summary name: summary_name = "fraction_of_% d_full"% capacity. To do this, check line 235. Something similar happens here for the packet queue.
The reason they are written without an explicit request is because they were created without an explicit request, followed by a line of code:
merged_summaries = tf.summary.merge_all()
put all these summaries together, so when I called:
sess.run([optimizer, cost, merged_summaries], ..... )
writer.add_summary(s, batch)
I did ask that they be recorded too.
I hope this answer helped some people.
source to share