TSP-Hopfield-Tank

Contents

• Hopfield-Tank network
• 'Heatmap image' generation for given step
• 'Heatmap over time' video generation (very usefull for debuging)

Deployment

Running

Programs supports commandline interface which allows to run network once, or multiple times in a row. Properties of the network can be configured with arguments

  -h, --help            show this help message and exit
  --steps [STEPS]       Number of steps to take.
  --freq [FREQ]         Frequency of taking snapshots.
  --seeds [SEEDS [SEEDS ...]]
                        Seed for random. Defines whole run.
  --size-adjs [SIZE_ADJS [SIZE_ADJS ...]]
                        specifies value of size adjustment
  --tag [TAG]           tag added to name

Dependancies

• Matplotlib
• numpy
• ffmpeg

Theoretical remarks

Implementation of Hopfield-Tank model for TSP. Project for my University Course.

Current version requires careful tuning of parameters to create feasible solutions. Look up page 16, second paragraph in following paper: http://www.iro.umontreal.ca/~dift6751/paper_potvin_nn_tsp.pdf

Based on following article by John J Hopfield and D W Tank: https://www.researchgate.net/publication/19135224_Neural_Computation_of_Decisions_in_Optimization_Problems Mainly equations and constants from 4'th paragraph were used.

Contributors

• Maciej Staniuk
• Mateusz Albecki