BDG—ODE

Requirements

Before running the code, ensure that you have the required dependencies installed. The key libraries used in this project include:

• Python 3.x
• NumPy
• PyTorch
• TorchDiffeq
• Matplotlib
• NetworkX
• PIL (Python Imaging Library)
• Pandas

You can install the required libraries using the following command:

pip install -r requirements.txt

Usage

To run the training script, use the following command format:

python main.py --T <Terminal_Time> --sampled_time <Time_Sampling_Type> --baseline <Model_Type> --gpu <GPU_ID> --weight_decay <Weight_Decay> --lr <Learning_Rate> --sparse --niters <Number_of_Iterations> --dump --dataset <Dataset>

Example Command

The following command trains the model using dataset f3 with specific hyperparameters:

python main.py --T 5 --sampled_time irregular --baseline BDG --gpu -1 --weight_decay 1e-3 --lr 0.01 --sparse --niters 1500 --dump --dataset f3

Parameters

• --T: Terminal time (the length of the time interval for the dynamics).
• --sampled_time: Time sampling method, options include irregular or equal.
• --baseline: The baseline model to use for training (e.g., BDG).
• --gpu: The GPU ID to use for training. Use -1 for CPU.
• --weight_decay: The L2 regularization weight for the optimizer.
• --lr: Learning rate for the optimizer.
• --sparse: Use sparse matrices for the adjacency matrix (optional).
• --niters: Number of iterations (epochs) for training.
• --dump: Save the training results in the results directory (optional).
• --dataset: Dataset selection. Options are f1, f2, and f3.

Output

If the --dump option is used, the results will be saved in the results/cognitive/ directory.

Dataset Download

GoogleDrive：https://drive.google.com/file/d/11Vh3IK6VIlNUKgmbQkXQwVmVti8EXuhm/view?usp=drive_link