tf.Session is a class that TensorFlow provides to represent a connection between the Python program and the C++ runtime.
The tf.Session object is the only object able to communicate directly with the hardware (through the C++ runtime), placing operations on the specified devices, using the local and distributed TensorFlow runtime, with the goal of concretely building the defined graph. The tf.Session object is highly optimized and, once correctly built, caches tf.Graph in order to speed up its execution.
Being the owner of physical resources, the tf.Session object must be used as a file descriptor to do the following:
- Acquire the resources by creating a tf.Session (the equivalent of the open operating system ...
