5 folders: 4 for running code, 1 for dataset;

Folder:

LSTM-LSTM: encoder as LSTM, decoder as LSTM;
LSTM-Transformer: encoder as LSTM, decoder as Transformer;
Transformer-LSTM: encoder as Transformer, decoder as LSTM;
Transformer-Transformer: encoder as Transformer, decoder as Transformer

Running code:
Entering into each running folder:
run:  

CUDA_VISIBLE_DEVICES=0 python run.py 
--train_dataset_path 
    train.pt (generate training dataset in dataset folder) 
--eval_dataset_path 
    eval.pt (generate testing dataset in dataset folder) 


For other experiments, please generate related dataset using:

dataset/dataset_generator.py (ASAP)
dataset/dataset_generator_scheduling_reversed.py (ALAP)


The command in dataset generator is:

myDataset = TopoSortDataset(size=50, num_samples=10240, in_degree_fixed=3,
in_degree_total=6, resource_constraint_level=eval_lvl, level_range=[16, 40],
weight_multiply=5., weight_constraint=35.)

size: number of nodes;
num_samples: number of graphs in the dataset;
in_degree_fixed: incoming edges per node;
in_degree_total: maximum incoming edges embedding can hold;
resource_constraint_level: default
level_range: depth of graphs;
weight_multiply: memory for each node;
weight_constraint: pipeline memory constraint.


