Error Passing Network (EPN)

This repository stores the code implemented to generate the results of the paper:
Development of Error Passing Network for Optimizing the Prediction of VO2 peak in Childhood Acute Leukemia Survivors

The datasets analysed during the study are not publicly available for confidentiality purposes. However, randomly generated datasets with the same format as used in our experiments are publicly shared in the data directory.

Installation

To have all the requirements needed, you must do the following actions:

• Open a terminal
• Clone this repo: git clone git@github.com:Rayn2402/ErrorPassingNetwork.git
• Move into the directory: cd ErrorPassingNetwork/
• Create a virtual environment with all requirements (except Pytorch) by running the following commands:
  • conda create --name epn python=3.10 scipy seaborn pandas numpy tqdm matplotlib scikit-learn plotly python-kaleido
  • conda activate epn
  • conda install -c conda-forge optuna py-xgboost cmaes
  • conda install -c dglteam dgl
  • pip install psycopg2-binary
  • pip install -U "ray[default]"
• Install Pytorch library according to your hardware requirements
  • Select the latest stable Pytorch Build
  • Select the appropriate version for Your OS
  • Select Conda as the Package
  • Select Python as the Language
  • Select CPU or Default as the Compute Platform
  • Copy and paste the command provided, but remove torchvision and torchaudio.

Test the implementation

You can write the following lines in a terminal to replicate the experiments of the manuscript using the randomly generated data stored in the data directory. Records of the experiments will be stored in records/experiments directory as they will be completed. For the manuscript, --nb_trials was set to 500. However, here, we set it to 50 to reduce the execution time of test runs.

Experiments with walk variables

- Labonté equation

time python scripts/experiments/original_equation.py --from_csv

- Labonté equation + EPN

python scripts/experiments/model_evaluations.py \
--from_csv \
--remove_sex_variable \
--epn \
--path records/experiments/labonte/ \
--additional_tag labonte \
--nb_trials 50

Experiments without walk variables

- Linear regression, Random Forest and XGBoost

python scripts/experiments/model_evaluations.py \
--from_csv \
--remove_walk_variables \
--linear \
--random_forest \
--xgboost \
--nb_trials 50

- Linear regression + EPN

python scripts/experiments/model_evaluations.py \
--from_csv \
--remove_walk_variables \
--epn \
--path records/experiments/LR_nw/ \
--additional_tag LR \
--nb_trials 50

- Random Forest + EPN

python scripts/experiments/model_evaluations.py \
--from_csv \
--remove_walk_variables \
--epn \
--path records/experiments/RF_nw/ \
--additional_tag RF \
--nb_trials 50

- XGBoost + EPN

python scripts/experiments/model_evaluations.py \
--from_csv \
--remove_walk_variables \
--epn \
--path records/experiments/XGBoost_nw/ \
--additional_tag XGBoost \
--nb_trials 50

Summarizing all results

To summarize the results of test runs, run the command below. The output will be stored in records/csv/results.csv.

python scripts/utils/get_scores_csv.py --path records/experiments/ --filename results

Specs of our computer and execution times recorded for each experiment are displayed below.

• Computer model: Alienware Aurora Ryzen Edition
• Linux version: Ubuntu 20.04.4 LTS
• CPU: AMD Ryzen 9 3900X 12-Core Processor
• GPU: None were used for our experiments

Experiment (with walk variables)	Time
Labonté	3s
Labonté + EPN	3m12s

Experiment (without walk variables)	Time
Linear Regression	6s
Random Forest	5m55s
XGBoost	17s
Linear Regression + EPN	3m10s
Random Forest + EPN	3m17s
XGBoost + EPN	3m29s

Project Tree

├── checkpoints                   <- Temporary state dictionaries save by the EarlyStopper module
├── data
│   └── vo2_dataset_random.csv    <- Synthetic dataset for the VO2 peak prediction task
|
├── hps                           <- Python files used to store sets of hyperparameter values and search spaces
├── masks                         <- JSON files used to store random stratified sampling masks
├── models                        <- State dictionaries associated to the best models
├── records                       <- Directories in which results and summaries of data analyses are stored
|
├── scripts
│   ├── experiments               <- Scripts to run individual experiments
│   ├── post_analyses             <- Scripts to run post analyses
│   └── utils                     <- Scripts to execute different subtasks
|
├── settings                      <- Files used for the setup of the project environment
|
├── src                           <- All project modules
│   ├── data
│   │   ├── extraction            <- Modules related to data extraction from PostgreSQL
│   │   └── processing            <- Modules related to data processing
│   ├── evaluation                <- Modules related to the evaluation and tuning of the models
│   ├── models
│   │   ├── abstract_models       <- Abstract classes from which new models have to inherit
│   │   ├── blocks                <- Neural network architecture blocks
│   │   └── wrappers              <- Abstract classes used to wrap existing models
│   ├── recording                 <- Recording module
│   └── utils                     <- Modules associated to visualization, metrics, hps and more
└── README.md