LRGCPND

This is an example implementation of the LRGCPND model.

Environment Settings

Requirement	Version
python	3.8
numpy	1.20
pandas	1.2
scipy	1.6
scikit-learn	0.24
PyTorch	1.8
cudatoolkit	10.2

Usage

Run python run.py -h/--help for more detailed usage:

usage: run.py [-h] [--n_num N_NUM] [--d_num D_NUM] [-K K] [-S S] [-r REG] [-l LR] [-e EPOCHS] [-b BATCH] [-f FOLD] [-t TIME] [--save_models]
              [--have_trained]

optional arguments:
  --n_num N_NUM         Number of ncRNAs.
  --d_num D_NUM         Number of drugs.
  -K K                  Depth of layers.
  -r REG, --reg REG     Coefficient of L2 regularization.
  -l LR, --lr LR        Initial learning rate.
  -e EPOCHS, --epochs EPOCHS
                        Number of epochs to train.
  -b BATCH, --batch BATCH
                        Batch size to train.
  -f FOLD, --fold FOLD  Number of folds for cross validation.
  -t TIME, --time TIME  Timestamp in milliseconds for training.
  --save_models         Save trained models (true or false).
  --have_trained        Have trained models (true or false).

Generate samples for k-fold CV (necessary for the first time)

python split.py

The randomly generated triples will be saved in /data/samples.

Run (train and test)

python run.py

or if you'd like to save the models:

python run.py --save_models

Be aware that we use timestamps to mark models and corresponding results.

Evaluate (load and test)

python run.py --have_trained -t {str_time}

where {str_time} is the timestamp of your trained model.