HybridNet: Learning Coordination Policies over Heterogeneous Graphs for Human-Robot Teams via Recurrent Neural Schedule Propagation

The Conference Paper link for Human Robot Team Scheduling Pipeline using Graph Attention Networks for embedding generation and RNN for sequential action schedule generation for Multi-Round Scheduling

File Structure

• ./requirements.yaml -- conda environment requirements
• ./hetnet.py -- the gnn model
• ./testlayer.py -- sanity check on model implementation
• ./utils.py -- utility functions and Scheduling Env class SchedulingEnv(object) handles the scheduling actions and updates constraints but is a highly customized one, so we want to implement a similar one under openai gym structure
• ./graph/hetgat.py -- GAT layer implementation
• ./benchmark/ -- Evaluation code for ScheduleNet and baseline heuristics edfutils.py has some simple class for robot teams
• ./gen/ -- Generate random problem instances mrc.py details how the constraints are sampled
• ./env/ -- OpenAI based Environment for Single Round and Multi-Round Scheduling
• ./evolutionary_algorithm.py -- Genetic Algorithm functions for generating schedules using multi-generational optimization

Environment Setup

Create the environment using the requirements.yaml file using:

conda env create -f requirements.yaml

Activate environment using:

conda activate hybridnet

Generate Task Allocation Problems:

robot: 2 human: 2 task_min: 9 task_max: 11

# Small Training Set
python generate_problems.py --folder=data/small_training_set/problems --human=2 --robot=2 --task-min=9 --task-max=11 --start-problem=1 --end-problem=2000
# Small Test Set
python generate_problems.py --folder=data/small_test_set/problems --human=2 --robot=2 --task-min=9 --task-max=11 --start-problem=1 --end-problem=200
# Medium Training Set
python generate_problems.py --folder=data/medium_training_set/problems --human=2 --robot=2 --task-min=18 --task-max=22 --start-problem=1 --end-problem=2000
# Medium Test Set
python generate_problems.py --folder=data/medium_test_set/problems --human=2 --robot=2 --task-min=18 --task-max=22 --start-problem=1 --end-problem=200
# Large Test Set
python generate_problems.py --folder=data/large_training_set/problems --human=2 --robot=2 --task-min=36 --task-max=44 --start-problem=1 --end-problem=200

Training Models

HybridNet Model

To train the HybridNet model, use the trainer.py file with the matching arguments

• --checkpoint: folder to store the checkpoints of the training, stores every 10 epoch of training
• --mode: training mode, for Policy Gradient use pg and for Greedy Baseline Policy Gradient use gb
• --folder: the folder that contains the training problem sets
• --start-problem: The Smallest problem number to train with (inclusive)
• --end-problem: The Largest problem id number to train with (inclusive). Training iterates through each problem and resets the training problem counter to start-problem when completing the end-problem
• --epoch: The number of epochs to train the model
• --weight-decay: The Weight Decay of the Neural Network to prevent overfitting.
• --infeasible-coefficient: The weight of the infeasible task completion, defaults to 2.0.
• --entropy-coefficient: Not used in the version of the IROS2022 submission
• --nn: Model type to train hybridnet for Model using LSTM Propagator and hetgat for interactive sequential HetGAT Model.
• --resume-cp: Add to select the epoch to resume training for. Leave out to start training from scratch.
• --human-noise: Boolean flag. Include in run to enable Stochastic Mode
• --estimator: Boolean flag. Include in run to enable estimator use. If not set, uses the actual human task duration (sampled if --human-noise is set)
• --estimator-noise: Boolean flag. Include in run to enable noise within the estimator. If not set, the estimator will produce mean of task-agent duration for given instance.

Deterministic Model Training for HybridNet

# Small Policy Gradient
python trainer.py --checkpoint=data/small_training_set/checkpoints_hybridnet_pg --mode=pg --folder=data/small_training_set/problems --start-problem=1 --end-problem=2000 --epoch=12000 --weight-decay=5e-4 --entropy-coefficient=0.0 --infeasible-coefficient=2.0 --nn=hybridnet
# Small Greedy Baseline
python trainer.py --checkpoint=data/small_training_set/checkpoints_hybridnet_gb --mode=gb --folder=data/small_training_set/problems --start-problem=1 --end-problem=2000 --epoch=12000 --weight-decay=5e-4 --entropy-coefficient=0.0 --infeasible-coefficient=2.0 --nn=hybridnet
# Medium Policy Gradient
python trainer.py --checkpoint=data/medium_training_set/checkpoints_hybridnet_pg --mode=pg --folder=data/small_training_set/problems --start-problem=1 --end-problem=2000 --epoch=12000 --weight-decay=5e-4 --entropy-coefficient=0.0 --infeasible-coefficient=2.0 --nn=hybridnet
# Medium Greedy Baseline
python trainer.py --checkpoint=data/medium_training_set/checkpoints_hybridnet_pg --mode=gb --folder=data/small_training_set/problems --start-problem=1 --end-problem=2000 --epoch=12000 --weight-decay=5e-4 --entropy-coefficient=0.0 --infeasible-coefficient=2.0 --nn=hybridnet

Stochastic Model

# Small Policy Gradient
python trainer.py --checkpoint=data/small_training_set/checkpoints_hybridnet_stoc_pg --mode=pg --folder=data/small_training_set/problems --start-problem=1 --end-problem=2000 --epoch=12000 --weight-decay=5e-4 --entropy-coefficient=0.0 --infeasible-coefficient=2.0 --nn=hybridnet --human-noise --estimator --estimator-noise
# Small Greedy Baseline
python trainer.py --checkpoint=data/small_training_set/checkpoints_hybridnet_stoc_gb --mode=gb --folder=data/small_training_set/problems --start-problem=1 --end-problem=2000 --epoch=12000 --weight-decay=5e-4 --entropy-coefficient=0.0 --infeasible-coefficient=2.0 --nn=hybridnet --human-noise --estimator --estimator-noise
# Medium Policy Gradient
python trainer.py --checkpoint=data/medium_training_set/checkpoints_hybridnet_stoc_pg --mode=pg --folder=data/small_training_set/problems --start-problem=1 --end-problem=2000 --epoch=12000 --weight-decay=5e-4 --entropy-coefficient=0.0 --infeasible-coefficient=2.0 --nn=hybridnet --human-noise --estimator --estimator-noise
# Medium Greedy Baseline
python trainer.py --checkpoint=data/medium_training_set/checkpoints_hybridnet_stoc_pg --mode=gb --folder=data/small_training_set/problems --start-problem=1 --end-problem=2000 --epoch=12000 --weight-decay=5e-4 --entropy-coefficient=0.0 --infeasible-coefficient=2.0 --nn=hybridnet --human-noise --estimator --estimator-noise

HetGAT Model

Deterministic HetGAT Model Training for performance comparison

# Small Policy Gradient
python trainer.py --checkpoint=data/small_training_set/checkpoints_hetgat_pg --mode=pg --folder=data/small_training_set/problems --start-problem=1 --end-problem=2000 --epoch=12000 --weight-decay=5e-4 --entropy-coefficient=0.0 --infeasible-coefficient=2.0 --nn=hetgat
# Small Greedy Baseline
python trainer.py --checkpoint=data/small_training_set/checkpoints_hetgat_gb --mode=gb --folder=data/small_training_set/problems --start-problem=1 --end-problem=2000 --epoch=12000 --weight-decay=5e-4 --entropy-coefficient=0.0 --infeasible-coefficient=2.0 --nn=hetgat
# Medium Policy Gradient
python trainer.py --checkpoint=data/medium_training_set/checkpoints_hetgat_pg --mode=pg --folder=data/small_training_set/problems --start-problem=1 --end-problem=2000 --epoch=12000 --weight-decay=5e-4 --entropy-coefficient=0.0 --infeasible-coefficient=2.0 --nn=hetgat
# Medium Greedy Baseline
python trainer.py --checkpoint=data/medium_training_set/checkpoints_hetgat_pg --mode=gb --folder=data/small_training_set/problems --start-problem=1 --end-problem=2000 --epoch=12000 --weight-decay=5e-4 --entropy-coefficient=0.0 --infeasible-coefficient=2.0 --nn=hetgat

Evaluation

Evalutation of a Range of Trained Checkpoints

Used to select a trained model among the stored epochs that can perform the best.

python pg_eval.py --mode=best --nn=hybridnet --cp=data/small_training_set/checkpoints_hetgat_pg --data-folder=data/small_test_set --save-folder=data/small_training_set/checkpoints_hetgat_pg/results --start-cp=0 --end-cp=12000 --cp-period=500 --batch-size=8

Evaluation of a single Checkpoint with multiple iterations:

• --mode: Selection of Tasks mode. Use best for the best of batch model showing improved result through ensemble schedule generation. sample generates a single schedule based on sampling with probability distribution for the model output. argmax selects the model output with highest probability for a single schedule.
• --nn: Model for evaluation set to hybridnet or hetgat for the models
• --specific-cp: The Checkpoint that is specifically
• --cp: Checkpoint Folder Location to iterate through for sequential evaluation. Gets disabled if --specific-cp is set.
• --start-cp: First Checkpoint epoch to be evaluated. Gets disabled if --specific-cp is set.
• --end-cp: Last Checkpoint epoch to be evaluated (inclusive). Gets disabled if --specific-cp is set.
• --cp-period: The increment for the epoch count to evaluated with. Gets disabled if --specific-cp is set.
• --data-folder: Folder of the test problems
• --save-folder: Save location of the output files
• --batch: The number of schedules generatated for each problem instance for evaluation used for evaluation. For best, the best schedule performance among the batch number of schedules is recorded and propagated in consecutive rounds
• --repeat: The number of repetitions for the test set to generate a performance distribution.

Deterministic Performance Evaluation

HybridNet Models Trained on Small Scale:

The following Codes generate the performance results for the Small Trained Models for Small, Medium and Large Data Scales:

# Trained on Small, Test on Small Test Set PG 8
python pg_eval.py --mode=best --nn=hybridnet --specific-cp=data/final_trained_models/checkpoint_hybridnet_small_pg.tar --data-folder=data/small_test_set --save-folder=data/final_trained_models/hybridnet_small_small_pg --estimator --batch-size=8 --repeat=10
# Trained on Small, Test on Small Test Set PG 16
python pg_eval.py --mode=best --nn=hybridnet --specific-cp=data/final_trained_models/checkpoint_hybridnet_small_pg.tar --data-folder=data/small_test_set --save-folder=data/final_trained_models/hybridnet_small_small_pg --estimator --batch-size=16 --repeat=10
# Trained on Small, Test on Small Test GB 8
python pg_eval.py --mode=best --nn=hybridnet --specific-cp=data/final_trained_models/checkpoint_hybridnet_small_gb.tar --data-folder=data/small_test_set --save-folder=data/final_trained_models/hybridnet_small_small_pg --estimator --batch-size=8 --repeat=10
# Trained on Small, Test on Small Test GB 16
python pg_eval.py --mode=best --nn=hybridnet --specific-cp=data/final_trained_models/checkpoint_hybridnet_small_gb.tar --data-folder=data/small_test_set --save-folder=data/final_trained_models/hybridnet_small_small_pg --estimator --batch-size=16 --repeat=10

# Trained on Small, Test on Medium Test Set PG 8
python pg_eval.py --mode=best --nn=hybridnet --specific-cp=data/final_trained_models/checkpoint_hybridnet_small_pg.tar --data-folder=data/medium_test_set --save-folder=data/final_trained_models/hybridnet_small_medium_pg --estimator --batch-size=8 --repeat=10
# Trained on Small, Test on Medium Test Set PG 16
python pg_eval.py --mode=best --nn=hybridnet --specific-cp=data/final_trained_models/checkpoint_hybridnet_small_pg.tar --data-folder=data/medium_test_set --save-folder=data/final_trained_models/hybridnet_small_medium_pg --estimator --batch-size=16 --repeat=10
# Trained on Small, Test on Medium Test GB 8
python pg_eval.py --mode=best --nn=hybridnet --specific-cp=data/final_trained_models/checkpoint_hybridnet_small_gb.tar --data-folder=data/medium_test_set --save-folder=data/final_trained_models/hybridnet_small_medium_pg --estimator --batch-size=8 --repeat=10
# Trained on Small, Test on Medium Test GB 16
python pg_eval.py --mode=best --nn=hybridnet --specific-cp=data/final_trained_models/checkpoint_hybridnet_small_gb.tar --data-folder=data/medium_test_set --save-folder=data/final_trained_models/hybridnet_small_medium_pg --estimator --batch-size=16 --repeat=10

# Trained on Small, Test on Large Test Set PG 8
python pg_eval.py --mode=best --nn=hybridnet --specific-cp=data/final_trained_models/checkpoint_hybridnet_small_pg.tar --data-folder=data/large_test_set --save-folder=data/final_trained_models/hybridnet_small_large_pg --estimator --batch-size=8 --repeat=10
# Trained on Small, Test on Large Test Set PG 16
python pg_eval.py --mode=best --nn=hybridnet --specific-cp=data/final_trained_models/checkpoint_hybridnet_small_pg.tar --data-folder=data/large_test_set --save-folder=data/final_trained_models/hybridnet_small_large_pg --estimator --batch-size=16 --repeat=10
# Trained on Small, Test on Large Test GB 8
python pg_eval.py --mode=best --nn=hybridnet --specific-cp=data/final_trained_models/checkpoint_hybridnet_small_gb.tar --data-folder=data/large_test_set --save-folder=data/final_trained_models/hybridnet_small_large_pg --estimator --batch-size=8 --repeat=10
# Trained on Small, Test on Large Test GB 16
python pg_eval.py --mode=best --nn=hybridnet --specific-cp=data/final_trained_models/checkpoint_hybridnet_small_gb.tar --data-folder=data/large_test_set --save-folder=data/final_trained_models/hybridnet_small_large_pg --estimator --batch-size=16 --repeat=10

HybridNet Models Trained on Medium Scale:

The following Codes generate the performance results for the Medium Trained Models for Medium and Large Data Scales:

# Trained on Medium, Test on Medium Test Set PG 8
python pg_eval.py --mode=best --nn=hybridnet --specific-cp=data/final_trained_models/checkpoint_hybridnet_medium_pg.tar --data-folder=data/medium_test_set --save-folder=data/final_trained_models/hybridnet_medium_medium_pg --estimator --batch-size=8 --repeat=10
# Trained on Medium, Test on Medium Test Set PG 16
python pg_eval.py --mode=best --nn=hybridnet --specific-cp=data/final_trained_models/checkpoint_hybridnet_medium_pg.tar --data-folder=data/medium_test_set --save-folder=data/final_trained_models/hybridnet_medium_medium_pg --estimator --batch-size=16 --repeat=10
# Trained on Medium, Test on Medium Test GB 8
python pg_eval.py --mode=best --nn=hybridnet --specific-cp=data/final_trained_models/checkpoint_hybridnet_medium_gb.tar --data-folder=data/medium_test_set --save-folder=data/final_trained_models/hybridnet_medium_medium_pg --estimator --batch-size=8 --repeat=10
# Trained on Medium, Test on Medium Test GB 16
python pg_eval.py --mode=best --nn=hybridnet --specific-cp=data/final_trained_models/checkpoint_hybridnet_medium_gb.tar --data-folder=data/medium_test_set --save-folder=data/final_trained_models/hybridnet_medium_medium_pg --estimator --batch-size=16 --repeat=10

# Trained on Medium, Test on Large Test Set PG 8
python pg_eval.py --mode=best --nn=hybridnet --specific-cp=data/final_trained_models/checkpoint_hybridnet_medium_pg.tar --data-folder=data/large_test_set --save-folder=data/final_trained_models/hybridnet_medium_large_pg --estimator --batch-size=8 --repeat=10
# Trained on Medium, Test on Large Test Set PG 16
python pg_eval.py --mode=best --nn=hybridnet --specific-cp=data/final_trained_models/checkpoint_hybridnet_medium_pg.tar --data-folder=data/large_test_set --save-folder=data/final_trained_models/hybridnet_medium_large_pg --estimator --batch-size=16 --repeat=10
# Trained on Medium, Test on Large Test GB 8
python pg_eval.py --mode=best --nn=hybridnet --specific-cp=data/final_trained_models/checkpoint_hybridnet_medium_gb.tar --data-folder=data/large_test_set --save-folder=data/final_trained_models/hybridnet_medium_large_pg --estimator --batch-size=8 --repeat=10
# Trained on Medium, Test on Large Test GB 16
python pg_eval.py --mode=best --nn=hybridnet --specific-cp=data/final_trained_models/checkpoint_hybridnet_medium_gb.tar --data-folder=data/large_test_set --save-folder=data/final_trained_models/hybridnet_medium_large_pg --estimator --batch-size=16 --repeat=10

HetGat Models Trained on Small Scale

# Trained on Small, Test on Small Test Set PG 8
python pg_eval.py --mode=best --nn=hetgat --specific-cp=data/final_trained_models/checkpoint_hetgat_small_pg.tar --data-folder=data/small_test_set --save-folder=data/final_trained_models/hetgat_small_small_pg --batch-size=8 --repeat=10
# Trained on Small, Test on Small Test Set PG 16
python pg_eval.py --mode=best --nn=hetgat --specific-cp=data/final_trained_models/checkpoint_hetgat_small_pg.tar --data-folder=data/small_test_set --save-folder=data/final_trained_models/hetgat_small_small_pg --batch-size=16 --repeat=10
# Trained on Small, Test on Small Test GB 8
python pg_eval.py --mode=best --nn=hetgat --specific-cp=data/final_trained_models/checkpoint_hetgat_small_gb.tar --data-folder=data/small_test_set --save-folder=data/final_trained_models/hetgat_small_small_pg --batch-size=8 --repeat=10
# Trained on Small, Test on Small Test GB 16
python pg_eval.py --mode=best --nn=hetgat --specific-cp=data/final_trained_models/checkpoint_hetgat_small_gb.tar --data-folder=data/small_test_set --save-folder=data/final_trained_models/hetgat_small_small_pg --batch-size=16 --repeat=10

# Trained on Small, Test on Medium Test Set PG 8
python pg_eval.py --mode=best --nn=hetgat --specific-cp=data/final_trained_models/checkpoint_hetgat_small_pg.tar --data-folder=data/medium_test_set --save-folder=data/final_trained_models/hetgat_small_medium_pg --batch-size=8 --repeat=10
# Trained on Small, Test on Medium Test Set PG 16
python pg_eval.py --mode=best --nn=hetgat --specific-cp=data/final_trained_models/checkpoint_hetgat_small_pg.tar --data-folder=data/medium_test_set --save-folder=data/final_trained_models/hetgat_small_medium_pg --batch-size=16 --repeat=10
# Trained on Small, Test on Medium Test GB 8
python pg_eval.py --mode=best --nn=hetgat --specific-cp=data/final_trained_models/checkpoint_hetgat_small_gb.tar --data-folder=data/medium_test_set --save-folder=data/final_trained_models/hetgat_small_medium_pg --batch-size=8 --repeat=10
# Trained on Small, Test on Medium Test GB 16
python pg_eval.py --mode=best --nn=hetgat --specific-cp=data/final_trained_models/checkpoint_hetgat_small_gb.tar --data-folder=data/medium_test_set --save-folder=data/final_trained_models/hetgat_small_medium_pg --batch-size=16 --repeat=10

# Trained on Small, Test on Large Test Set PG 8
python pg_eval.py --mode=best --nn=hetgat --specific-cp=data/final_trained_models/checkpoint_hetgat_small_pg.tar --data-folder=data/large_test_set --save-folder=data/final_trained_models/hetgat_small_large_pg --batch-size=8 --repeat=10
# Trained on Small, Test on Large Test Set PG 16
python pg_eval.py --mode=best --nn=hetgat --specific-cp=data/final_trained_models/checkpoint_hetgat_small_pg.tar --data-folder=data/large_test_set --save-folder=data/final_trained_models/hetgat_small_large_pg --batch-size=16 --repeat=10
# Trained on Small, Test on Large Test GB 8
python pg_eval.py --mode=best --nn=hetgat --specific-cp=data/final_trained_models/checkpoint_hetgat_small_gb.tar --data-folder=data/large_test_set --save-folder=data/final_trained_models/hetgat_small_large_pg --batch-size=8 --repeat=10
# Trained on Small, Test on Large Test GB 16
python pg_eval.py --mode=best --nn=hetgat --specific-cp=data/final_trained_models/checkpoint_hetgat_small_gb.tar --data-folder=data/large_test_set --save-folder=data/final_trained_models/hetgat_small_large_pg --batch-size=16 --repeat=10

HetGat Models Trained on Medium Scale

# Trained on Medium, Test on Medium Test Set PG 8
python pg_eval.py --mode=best --nn=hetgat --specific-cp=data/final_trained_models/checkpoint_hetgat_medium_pg.tar --data-folder=data/medium_test_set --save-folder=data/final_trained_models/hetgat_medium_medium_pg --batch-size=8 --repeat=10
# Trained on Medium, Test on Medium Test Set PG 16
python pg_eval.py --mode=best --nn=hetgat --specific-cp=data/final_trained_models/checkpoint_hetgat_medium_pg.tar --data-folder=data/medium_test_set --save-folder=data/final_trained_models/hetgat_medium_medium_pg --batch-size=16 --repeat=10
# Trained on Medium, Test on Medium Test GB 8
python pg_eval.py --mode=best --nn=hetgat --specific-cp=data/final_trained_models/checkpoint_hetgat_medium_gb.tar --data-folder=data/medium_test_set --save-folder=data/final_trained_models/hetgat_medium_medium_pg --batch-size=8 --repeat=10
# Trained on Medium, Test on Medium Test GB 16
python pg_eval.py --mode=best --nn=hetgat --specific-cp=data/final_trained_models/checkpoint_hetgat_medium_pg.tar --data-folder=data/medium_test_set --save-folder=data/final_trained_models/hetgat_medium_medium_pg --batch-size=16 --repeat=10

# Trained on Medium, Test on Large Test Set PG 8
python pg_eval.py --mode=best --nn=hetgat --specific-cp=data/final_trained_models/checkpoint_hetgat_medium_pg.tar --data-folder=data/large_test_set --save-folder=data/final_trained_models/hetgat_medium_large_pg --batch-size=8 --repeat=10
# Trained on Medium, Test on Large Test Set PG 16
python pg_eval.py --mode=best --nn=hetgat --specific-cp=data/final_trained_models/checkpoint_hetgat_medium_pg.tar --data-folder=data/large_test_set --save-folder=data/final_trained_models/hetgat_medium_large_pg --batch-size=16 --repeat=10
# Trained on Medium, Test on Large Test GB 8
python pg_eval.py --mode=best --nn=hetgat --specific-cp=data/final_trained_models/checkpoint_hetgat_medium_gb.tar --data-folder=data/large_test_set --save-folder=data/final_trained_models/hetgat_medium_large_pg --batch-size=8 --repeat=10
# Trained on Medium, Test on Large Test GB 16
python pg_eval.py --mode=best --nn=hetgat --specific-cp=data/final_trained_models/checkpoint_hetgat_medium_gb.tar --data-folder=data/large_test_set --save-folder=data/final_trained_models/hetgat_medium_large_pg --batch-size=16 --repeat=10

For HetGAT Models, nn argument must be set to hetgat

python pg_eval.py --mode=best --nn=hetgat --specific-cp=data/final_trained_models/checkpoint_hetgat_small_pg.tar --data-folder=tmp/small_test_set --save-folder=tmp/small_test_results --batch-size=8 --repeat=10

Stochastic Performance Evaluation

HybridNet Only

Stochastic HybridNet Models Trained on Small Scale:

The following Codes generate the performance results for the Small Trained Models for Small, Medium and Large Data Scales:

# Trained on Small, Test on Small Test Set PG 8
python pg_eval.py --mode=best --nn=hybridnet --specific-cp=data/final_trained_models/checkpoint_hybridnet_small_pg.tar --data-folder=data/small_test_set --save-folder=data/final_trained_models/hybridnet_stoc_small_small_pg --batch-size=8 --repeat=10 --human-noise --estimator --estimator-noise
# Trained on Small, Test on Small Test Set PG 16
python pg_eval.py --mode=best --nn=hybridnet --specific-cp=data/final_trained_models/checkpoint_hybridnet_small_pg.tar --data-folder=data/small_test_set --save-folder=data/final_trained_models/hybridnet_stoc_small_small_pg --batch-size=16 --repeat=10 --human-noise --estimator --estimator-noise
# Trained on Small, Test on Small Test GB 8
python pg_eval.py --mode=best --nn=hybridnet --specific-cp=data/final_trained_models/checkpoint_hybridnet_small_gb.tar --data-folder=data/small_test_set --save-folder=data/final_trained_models/hybridnet_stoc_small_small_pg --batch-size=8 --repeat=10 --human-noise --estimator --estimator-noise
# Trained on Small, Test on Small Test GB 16
python pg_eval.py --mode=best --nn=hybridnet --specific-cp=data/final_trained_models/checkpoint_hybridnet_small_gb.tar --data-folder=data/small_test_set --save-folder=data/final_trained_models/hybridnet_stoc_small_small_pg --batch-size=16 --repeat=10 --human-noise --estimator --estimator-noise

# Trained on Small, Test on Medium Test Set PG 8
python pg_eval.py --mode=best --nn=hybridnet --specific-cp=data/final_trained_models/checkpoint_hybridnet_small_pg.tar --data-folder=data/medium_test_set --save-folder=data/final_trained_models/hybridnet_stoc_small_medium_pg --batch-size=8 --repeat=10 --human-noise --estimator --estimator-noise
# Trained on Small, Test on Medium Test Set PG 16
python pg_eval.py --mode=best --nn=hybridnet --specific-cp=data/final_trained_models/checkpoint_hybridnet_small_pg.tar --data-folder=data/medium_test_set --save-folder=data/final_trained_models/hybridnet_stoc_small_medium_pg --batch-size=16 --repeat=10 --human-noise --estimator --estimator-noise
# Trained on Small, Test on Medium Test GB 8
python pg_eval.py --mode=best --nn=hybridnet --specific-cp=data/final_trained_models/checkpoint_hybridnet_small_gb.tar --data-folder=data/medium_test_set --save-folder=data/final_trained_models/hybridnet_stoc_small_medium_pg --batch-size=8 --repeat=10 --human-noise --estimator --estimator-noise
# Trained on Small, Test on Medium Test GB 16
python pg_eval.py --mode=best --nn=hybridnet --specific-cp=data/final_trained_models/checkpoint_hybridnet_small_gb.tar --data-folder=data/medium_test_set --save-folder=data/final_trained_models/hybridnet_stoc_small_medium_pg --batch-size=16 --repeat=10 --human-noise --estimator --estimator-noise

# Trained on Small, Test on Large Test Set PG 8
python pg_eval.py --mode=best --nn=hybridnet --specific-cp=data/final_trained_models/checkpoint_hybridnet_small_pg.tar --data-folder=data/large_test_set --save-folder=data/final_trained_models/hybridnet_stoc_small_large_pg --batch-size=8 --repeat=10 --human-noise --estimator --estimator-noise
# Trained on Small, Test on Large Test Set PG 16
python pg_eval.py --mode=best --nn=hybridnet --specific-cp=data/final_trained_models/checkpoint_hybridnet_small_pg.tar --data-folder=data/large_test_set --save-folder=data/final_trained_models/hybridnet_stoc_small_large_pg --batch-size=16 --repeat=10 --human-noise --estimator --estimator-noise
# Trained on Small, Test on Large Test GB 8
python pg_eval.py --mode=best --nn=hybridnet --specific-cp=data/final_trained_models/checkpoint_hybridnet_small_gb.tar --data-folder=data/large_test_set --save-folder=data/final_trained_models/hybridnet_stoc_small_large_pg --batch-size=8 --repeat=10 --human-noise --estimator --estimator-noise
# Trained on Small, Test on Large Test GB 16
python pg_eval.py --mode=best --nn=hybridnet --specific-cp=data/final_trained_models/checkpoint_hybridnet_small_gb.tar --data-folder=data/large_test_set --save-folder=data/final_trained_models/hybridnet_stoc_small_large_pg --batch-size=16 --repeat=10 --human-noise --estimator --estimator-noise

Stochastic HybridNet Models Trained on Medium Scale:

The following Codes generate the performance results for the Medium Trained Models for Medium and Large Data Scales:

# Trained on Medium, Test on Medium Test Set PG 8
python pg_eval.py --mode=best --nn=hybridnet --specific-cp=data/final_trained_models/checkpoint_hybridnet_medium_pg.tar --data-folder=data/medium_test_set --save-folder=data/final_trained_models/hybridnet_stoc_medium_medium_pg --batch-size=8 --repeat=10 --human-noise --estimator --estimator-noise
# Trained on Medium, Test on Medium Test Set PG 16
python pg_eval.py --mode=best --nn=hybridnet --specific-cp=data/final_trained_models/checkpoint_hybridnet_medium_pg.tar --data-folder=data/medium_test_set --save-folder=data/final_trained_models/hybridnet_stoc_medium_medium_pg --batch-size=16 --repeat=10 --human-noise --estimator --estimator-noise
# Trained on Medium, Test on Medium Test GB 8
python pg_eval.py --mode=best --nn=hybridnet --specific-cp=data/final_trained_models/checkpoint_hybridnet_medium_gb.tar --data-folder=data/medium_test_set --save-folder=data/final_trained_models/hybridnet_stoc_medium_medium_pg --batch-size=8 --repeat=10 --human-noise --estimator --estimator-noise
# Trained on Medium, Test on Medium Test GB 16
python pg_eval.py --mode=best --nn=hybridnet --specific-cp=data/final_trained_models/checkpoint_hybridnet_medium_gb.tar --data-folder=data/medium_test_set --save-folder=data/final_trained_models/hybridnet_stoc_medium_medium_pg --batch-size=16 --repeat=10 --human-noise --estimator --estimator-noise

# Trained on Medium, Test on Large Test Set PG 8
python pg_eval.py --mode=best --nn=hybridnet --specific-cp=data/final_trained_models/checkpoint_hybridnet_medium_pg.tar --data-folder=data/large_test_set --save-folder=data/final_trained_models/hybridnet_stoc_medium_large_pg --batch-size=8 --repeat=10 --human-noise --estimator --estimator-noise
# Trained on Medium, Test on Large Test Set PG 16
python pg_eval.py --mode=best --nn=hybridnet --specific-cp=data/final_trained_models/checkpoint_hybridnet_medium_pg.tar --data-folder=data/large_test_set --save-folder=data/final_trained_models/hybridnet_stoc_medium_large_pg --batch-size=16 --repeat=10 --human-noise --estimator --estimator-noise
# Trained on Medium, Test on Large Test GB 8
python pg_eval.py --mode=best --nn=hybridnet --specific-cp=data/final_trained_models/checkpoint_hybridnet_medium_gb.tar --data-folder=data/large_test_set --save-folder=data/final_trained_models/hybridnet_stoc_medium_large_pg --batch-size=8 --repeat=10 --human-noise --estimator --estimator-noise
# Trained on Medium, Test on Large Test GB 16
python pg_eval.py --mode=best --nn=hybridnet --specific-cp=data/final_trained_models/checkpoint_hybridnet_medium_gb.tar --data-folder=data/large_test_set --save-folder=data/final_trained_models/hybridnet_stoc_medium_large_pg --batch-size=16 --repeat=10 --human-noise --estimator --estimator-noise

Comparison Baselines:

Earliest Deadline First:

# Deterministic
python benchmark/edf.py --folder=data/small_test_set --start=1 --end=200  --repeat=10
python benchmark/edf.py --folder=data/medium_test_set --start=1 --end=200  --repeat=10
python benchmark/edf.py --folder=data/large_test_set --start=1 --end=200  --repeat=10

Genetic Algorithm:

python benchmark/genetic_edf.py --folder=data/small_test_set --start=1 --end=200 --generation=10 --base-population=90 --new-random=10 --new-mutation=10  --repeat=10
python benchmark/genetic_edf.py --folder=data/medium_test_set --start=1 --end=200 --generation=10 --base-population=90 --new-random=10 --new-mutation=10  --repeat=10
python benchmark/genetic_edf.py --folder=data/large_test_set --start=1 --end=200 --generation=10 --base-population=90 --new-random=10 --new-mutation=10  --repeat=10

Results