pyhgf.updates.learning.learning_weights#

pyhgf.updates.learning.learning_weights(attributes, node_idx, edges, kind='precision_weighted', lr=None, adam_beta1=None, adam_beta2=None, adam_epsilon=None)[source]#

Unified weights update.

The gradient is first computed according to kind, then scaled by lr:

  • standard (kind="standard"): raw prediction-error outer product, no

precision weighting.

\(g_i = \text{PE} \cdot g(\text{parent}_i)\)

  • precision_weighted (kind="precision_weighted"): gradient weighted by the child posterior precision. \(g_i = \text{PE} \cdot \pi_\text{child} \cdot g(\text{parent}_i)\)

  • precision_ratio (kind="precision_ratio"): Kalman-gain-weighted PE using the posterior precisions of child and parent. \(K_i = \pi_\text{child} / (\pi_{\text{parent}_i} + \pi_\text{child})\) \(g_i = K_i \cdot \text{PE} \cdot g(\text{parent}_i)\)

lr controls how the gradient is applied (same semantics for all three kinds):

  • Adam (adam_beta1 is a float): gradient filtered through Adam; step size

controlled by lr. - Fixed (adam_beta1 is None): \(\Delta w_i = \text{lr} \cdot g_i\).

To recover the old “full Kalman step” behaviour for kind="precision_ratio", pass lr=1.0.

Parameters:

  • attributes (dict[int | str, dict]) – The attributes of the probabilistic network.

  • node_idx (int) – Pointer to the input node.

  • edges (tuple[AdjacencyLists, ...]) – The edges of the probabilistic nodes as a tuple of pyhgf.typing.Indexes. The tuple has the same length as node number. For each node, the index list value and volatility parents and children.

  • kind (str) – Gradient computation mode: "standard", "precision_weighted" (default), or "precision_ratio".

  • lr (float | None) – Fixed learning rate or Adam step size. Applied uniformly across all kind values, including "precision_ratio".

  • adam_beta1 (float | None) – Adam first moment decay rate. When None (default) Adam is not used.

  • adam_beta2 (float | None) – Adam second moment decay rate.

  • adam_epsilon (float | None) – Adam numerical stability constant.

Returns:

The attributes of the probabilistic network.

Return type:

attributes