Pointer Network for TSP

This repository contains pytorch implementation for Pointer Network[1] and focusses on the usecase for TSP.

Label Creation

Bellman Held Karp Heuristic [2]

Model Details

Encoder

1. Single LSTM Layer
2. 512 hidden units
3. Random Uniform weight initialization (-0.08 to 0.08)

Decoder

1. Single LSTM Layer with hidden state from encoder
2. 512 hidden units
3. Random Uniform weight initialization (-0.08 to 0.08)

Optimizer

1. SGD
   1. learning rate 1.0
   2. L2 Gradient Clipping 2.0
   3. Batch Size 128

Hardware

1. OS: macOS 14.0 23A344 arm64
2. Host: MacBookPro17,1
3. CPU: Apple M1
4. GPU: Apple M1
5. Memory: 16384MiB

Environment

conda env create --name name_of_environment --file=environment.yml

Links

[1] Vinyals, O. et al.: Pointer Networks, http://arxiv.org/abs/1506.03134, (2017). https://doi.org/10.48550/arXiv.1506.03134.

[2] https://github.com/Valdecy/pyCombinatorial