Surrogate model predicting TSP cost by means of neural network.
Tested on Julia 1.5.1, but should work on older versions, too.
git clone https://github.com/dhonza/CarSharingTSP.jl.git
cd CarSharingTSP.jl.git
julia --project=.
julia> using Pkg; Pkg.instantiate()
Select training dataset in src/carsharing-tsp.jl (TODO: add command line options). Execute:
julia --project=. src/carsharing-tsp.jl
To run on RCI cluster:
cd slurm
sbatch train_gpu.batch
The trained models as well as train logs are stored in the ./exp directory. RCI logs reside in ./logs.
The MultiLayer Perceptron trained in Julia can be converted to Python. See Jupyter notebooks: 1.0-hd-export_model_parameters.ipynb and 2.0-hd-import_python.ipynb.