To the best of my knowledge, TensorFlow can use multithreading in two ways:
- Perform lengthy I/O operations in separate threads
- Execute sessions with multiple threads to reduce processing time
For #1, I used threads and queues in the past, but now the documentation says I should use Datasets. Does this mean that QueueRunners and Coordinators are effectively deprecated? Are there still uses for them?
For #2, I use the session-configuration options defined in the ConfigProto message. Is there anything else I can do?