Implementation for learning curve extrapolation using graph-aware ordinary differential equations

The code is adapted from https://github.com/google-research/torchsde.

• To train and test the neural ODE model for learning curve extrapolation, run python latent_de.py with desired configuration. An example command is

  python3 latent_de.py --latent_dim 16 --cuda_id 3 --data cnn --batch-size 3 --diff_eq ode --seed $i --source_data cifar10 --pause-iters 50 --train_test_ratio 0.75

• To train and test LC-GODE for learning curve extrapolation, run python latent_de_graph.py with desired configuration.

  python3 latent_de_graph.py --latent_dim 16  --cuda_id 7 --data cnn --batch-size 3 --diff_eq ode --seed $i --source_data cifar10 --pause-iters 50 --train_test_ratio 0.75

The final evaluation metrics (e.g., mape, rmse, training runtime, test runtime) and the trained model will be saved in args.train_dir.