August 2018
Intermediate to advanced
272 pages
7h 2m
English
TensorFlow provides a nice way to profile and see how long your whole graph takes to execute through its timeline tracing tool. This is a great tool to see what parts of your graph are slowing down your training and to spot any inefficiencies in your data pipeline.
We will start by giving you an example of how to trace your graph. It is quite simple: you just add a couple of extra lines to your normal code and a JSON file will be produced that we can load in the Google Chrome browser to see all the timings of the graph execution:
from tensorflow.python.client import timeline .... # Your model and training code here with tf.Session() as sess: # We set some options to give to the session so graph execution is profiled. options ...
Read now
Unlock full access