This is an implementation of Pointer Networks to solve such problems. Beam search is supported.

If you don't know what is convex hulls, you could find the definition from here.

Generate dataset

Create data folder for saving dataset, then run node generate_ch_data.js to generate training and test data, you could adjust some parameters inside the source code.

Training and test

Create models folder for saving models, then run python3 train_ch.py to start training, also you could adjust some parameters inside the source code.

Review the incorrect cases

Run visualize_ch_data.html in your browser, then copy & paste the incorrect samples to textbox and visualize it.

The full name of MWT is minimum-weight triangulation, could refer to the wikipedia page.