Deep Dynamic Transformer Model (DDTM) for Multi-Start Team Orienteering Problem (MSTOP). Prizes are either constant or uniformly distributed. Training based on REINFORE algorithm with greedy rollout baseline (A), greedy rollout baseline with maximum entropy objective (B), multiple-sample baseline with replacement (C), instance-augmentation baseline (D), and instance-augmentation baseline with maximum entropy objecitve (E).
To run DDTM for solving MSTOP instances, first go to run.py and uncomment line 16 as
from nets.attention_model import AttentionModel
To run vanilla AM for solving TSP/CVRP instances, uncomment line 17 in run.py as
from nets.attention_model_original import AttentionModel
- Python >= 3.8
- Numpy
- SciPy
- Pytorch = 1.9.0
- tqdm
- tensorboard_logger
- Matplotlib = 3.4.3
This repository contains adaptations of the following repositories as basework