Reimplementation of the paper Learning to execute in Tensorflow (>=1.4).
Original implementation in Torch: https://github.com/wojciechz/learning_to_execute
A Seq2Seq model is trained to execute simple Python programs.
Example:
Input:
g=6
a=-77
if a>2:
g=a-4
print(a+g)
-----------
Target: -71
Model prediction: -71
The program generation code is simplified compared to the original paper but can easily be extended.
date-normalization.ipynb also contains a simple date format normalization example using the same model code.
This work was presented at PyCon Sweden 2016, Stockholm. Presentation