Tensorflow AutoComplete is a small tool which can assist in writing tensorflow code, based on RNN.
This project consists of two parts: server side and client side. On the server side, we developed a RNN, which can learn from tensorflow code in token level. The dataset is pulled from github, filtered and pre-processed, then input into the network. Then the RNN can predict tensorflow tokens, with 71% accuracy. As for the client side, we developed an Eclipse plugin, which can define a python editor, handle users' actions, send data to the server side, get the response and display to the screen.
The server side is written in tensorflow. The network has two layers, and each layer has 512 LSTM cells. The client side is written in Java, and can embed into Eclipse.
The client side is open-source, and you can also watch the demo video to learn how the tool assists in writing tensorflow code.