yiminking

Yes and no.

You can input an constraint solving optimization problem (such as TSP-Travelling Salesman Problem , the Vehicle Routing Problem, Employee Rostering, etc) into a Neural Network and after training you’ll get something out of it. Starting from a First Fit solution, expect about a 1% score improvement, which is well above the quality of randomly sampling solution states.

But it’s very much like taking a bicycle to a Formula 1 race.

Any basic constraint solving algorithm (such as Tabu Search, Simulated Annealing, Late Acceptance, …) easily beats those results, on any dataset. They get about a 15% improvement on TSP, etc. Research shows that Neural Nets are consistently inferior to constraint solving algorithms for constraint optimization problems. That being said, they are also consistently better at pattern recognition, such as image recognition and voice recognition. That’s no big surprise, given how good human brains are at pattern recognition and how bad they are at constraint solving.

Referred

yiminking

# Load the module
module load python3/anaconda

# Create an environment
conda create -n myenv python=3.8

# Activate the enviornment, which is in .conda/envs/myenv
source activate myenv

# Install whatever packages you need .. for example TF-gpu and numpy