This is console interface for the HORDA approach presented in the elsevier paper. Over the console interface it is possible to evaluate instance defined in json format via the HORDA with defined neural network.
Due to a lot of version of ML libraries it is necessary to install its separately.
Create conda environment without GPU:
conda create -n schnn python=3.7
conda activate schnn
conda install --file conda_requirements.txt
pip install pymonad==2.4.0You need python in version 3.7. Install requirements from requirements.txt file.
It is possible to give the instance over stdin during the evaluation of the python code.
The code expect the line with processing times of instances and secondly the line with due dates of instances, separated by space.
It is possible to give more than one instance, for finish of input mode write -1 after submitting due dates.
console.py --nn nn/best/nn.json.nn
It is possible to give one instance directly to the python code over the cli.
It is expected to give the instance in json format, as you can see in following example.
console.py --nn nn/out/best/nn.json.nn --instance "{"proc":[90, 51, 2, 90, 37, 16, 96, 45, 60, 1, 99, 32, 86, 55, 29, 26, 96, 100, 9, 7],"due":[367, 460, 368, 371, 423, 403, 375, 409, 323, 466, 313, 488, 429, 352, 318, 440, 373, 315, 494, 491]}"
It is possible to submit the file with more than one instance.
The file should be list of instances.
console.py --nn nn/out/best/nn.json.nn --instances_file test_instances.json
It is possible to store the results to the fie, with flag --output_file.
For example console.py --nn nn/out/best/nn.json.nn --instances_file test_instances.json --output_file out.json.
Warning: This can be time and memory demanding!
You can use switch --optimal in these case, the code also compute the optimal solution by our method.
It is reimplementation of the SDD approach in python, however not time and memory efficient.
The tkindt solver is more power.