gradnet.gradnet.DenseParameterization

class gradnet.gradnet.DenseParameterization(num_nodes, budget, mask, cost_matrix, *, delta_sign='nonnegative', directed=False, strict_budget=False, cost_aggr_norm=1, rand_init_weights=True)[source]

Dense parameterization of a delta adjacency matrix.

Maintains a dense, trainable delta_adj_raw and projects it to a constrained perturbation delta through the following pipeline:

raw -> (symmetrize?) -> (square?) -> mask -> normalize (when budget is set)
Parameters:

  • num_nodes (int) – Number of nodes (matrix dimension).

  • budget (float | None) – Target cost-weighted p-norm for delta. If None, skip normalization.

  • mask (torch.Tensor) – Dense mask selecting active entries (1 for active, 0 for inactive). Nonzero diagonal entries are allowed but typically masked out by users.

  • cost_matrix (torch.Tensor) – Per-entry cost tensor for the normalization.

  • delta_sign (str, optional) – Sign constraint for the perturbation. One of {"free", "nonnegative", "nonpositive"}.

  • directed (bool, optional) – If False, symmetrize before masking/normalizing.

  • strict_budget (bool, optional) – If True, always scale up/down to the exact budget; if False, scale down only.

  • cost_aggr_norm (int, optional) – Aggregation norm p for the cost-weighted p-norm.

  • rand_init_weights (bool | float, optional) – Initialization mix coefficient a. Cast to float and clamped to [0,1]. Initial raw parameters are set to (1 - a) * base + a * U(0,1), where base is ones for strict budget mode and zeros otherwise.

Initialize internal Module state, shared by both nn.Module and ScriptModule.

Methods

extra_repr()

Return the extra representation of the module.

forward([noise_amplitude])

Project raw parameters to a constrained delta matrix.

renorm_params()

Renormalize the raw parameters to a DOF-aware scale.

set_initial_state(delta_adj_raw_0)

Set the internal raw parameter and re-normalize.

Attributes

device

dtype

training