config.py

import math

class Config:
    """Hyperparameter configuration for dynamic spiking GNN experiments."""

    def __init__(self):
        # Dataset settings
        self.dataset = "Cora"  # or "PubMed"
        self.num_time_steps = 10  # total time steps
        self.train_steps = 7  # first 7 for pre-training
        self.adapt_steps = 2  # steps 8-9 for adaptation (STDP or surrogate)
        self.test_step = 9  # last step (0-indexed: step 9) for evaluation

        # Architecture
        self.node_feat_dim = 1433  # Cora feature dimension
        self.hidden_dim = 64
        self.out_dim = 1  # link prediction score (binary)
        self.num_layers = 2
        self.num_heads = 8  # for GAT
        self.dropout = 0.5

        # Training
        self.lr = 0.001
        self.weight_decay = 5e-4
        self.epochs_static = 200  # for baseline GNNs
        self.epochs_pretrain = 100  # for spiking GNN pre-training
        self.epochs_adapt = 10  # for fine-tuning or mask learning
        self.batch_size = 64
        self.seed = 42
        self.device = "cuda:0"

        # STDP hyperparameters
        self.mask_initial_ratio = 0.95
        self.stdp_learning_rate = 0.01
        self.stdp_window = 5  # temporal window for spike coincidence
        self.stdp_decay = 0.01

        # Surrogate gradient
        self.gumbel_tau = 1.0
        self.mask_lr = 0.01
        self.entropy_reg_lambda = 1e-3

        # Energy estimation
        self.num_macs_per_edge = 2  # approximate MAC cost per edge
        self.spike_energy_pJ = 0.9  # pJ per spike (45nm CMOS)
        self.mac_energy_pJ = 0.1  # pJ per MAC
        self.static_energy_mW = 100  # mW idle

    def to_dict(self):
        """Return configuration as a dictionary."""
        return {k: v for k, v in self.__dict__.items() if not k.startswith('_')}

    def __repr__(self):
        return f"Config({self.to_dict()})"