ATIS.keras

A short Keras tutorial that runs RNNs over the ATIS airline query dataset, pairing each word in a request with a slot tag such as B-dept for a departure city or I-arr for an arrival.

The repository is the code side of a spoken language understanding writeup. It implements RNNs in Keras to solve the Airline Travel Information System dataset, which is the slot filling part of SLU, and the README points to a blog post for the full explanation. What ships here is the Python and Keras implementation that accompanies that post rather than a standalone library, and it is MIT licensed.

A small sample shows the training format. The words show, flights, from, Boston, to, New, York, today carry tags O, O, O, B-dept, O, B-arr, I-arr, B-date. O marks tokens outside any slot, B-dept flags the departure city, and the arrival is split across B-arr and I-arr. That token labeling is exactly the task the RNN learns on the ATIS data, and it is the part of spoken language understanding where a model pulls structured information out of a natural language query.

The README stays short. It covers the dataset, the model type, and the blog reference, then stops. There is no training walkthrough or tuning guide, so anyone working through it will lean on the linked post for the details. As a tutorial companion it is deliberately bare, which makes the blog post the real documentation.

The repo pairs the Keras implementation with a blog post, keeping the README to the dataset, the model type, and the word-and-tag example.

• Source 1 · Project repository
• Source 2 · README