diffalign.models.epsnet.diffusion

Functions

cosine_beta_schedule(num_timesteps[, s])	Nichol & Dhariwal (2021): https://arxiv.org/abs/2102.09672 Returns betas of length T (float32).
get_distance(pos, edge_index)
linear_beta_schedule(num_timesteps[, ...])
merge_graphs_in_batch(batch1, batch2[, device])	Merge as [Q1,R1,Q2,R2,...] and attach graph_idx (even=query, odd=ref) and pair-level batch.

Classes

DDPMTimeEncoder(embed_dim[, activation])	SinusoidalPosEmb + MLP for timestep embeddings.
DiffAlign([node_feature_dim, ...])	Isotropic Gaussian Diffusion (v-parameterization; T steps) - Backbone: EGNN + CrossGraphAligner (only query coordinates move) - Output: v_t in merged (Q,R) order - Loss: v MSE + x0 anchor + optional repulsion
SinusoidalPosEmb(dim)	Sine/cosine timestep embedding (float32).