Pointer Networks in Tensorflow
This repo has sample code for pointer networks (Vinyals 2015) over a toy problem indexing a high-value segment embedded between two low-value segments.
Run the sample code by typing:
...and you'll train a simple pointer network on a low/high/low sequence indexing task, then test is on a separate data set with different segment lengths. The loss should drop to below 1% after about 2000 training steps (restarting if it stagnates), and will then attempt the test.