According to an article written by a previous Google intern explaining tensorflow eager mode .
Basically, this is a mode in a tensor stream that allows you to write the style of imperative coding, for example, using numpy . Therefore, there should no longer be explicit graph , session , session.run() . The graph is implicitly constructed when the code works as in Chainer / PyTorch .
It is still under active development, and performance is still not as good as it could be. You can try it through the nightly build on pip , expecting some interesting new functions, such as graph function , which allows you to call a subgraph as a function. As the structure tensorflow this may be the default mode for tensorflow .
Update: Tensorflow team officially talked about this .
Expected Execution is a mandatory, defining interface where operations are executed immediately as they are called from Python. This simplifies working with TensorFlow and can research and development more intuitively.
The vast majority of TensorFlow APIs remain the same whether execution is enabled or not. As a result, the same code that TensorFlow graphing (for example, using the API layers) can be executed imperatively, using impatient execution. In contrast, most models written with Eager enabled can be converted to a graph that can be further optimized and / or retrieved for deployment in production without changing the code.
You can learn more about this in a post or README . This is still a preview release, so you can hit some rough edges.
Looking forward to your feedback when you try.
Update: Standby is now officially published in TF 1.7 . There is a good tutorial on its website. I think everyone knows what it is, all the basic structures converge, dynamic neural networks.