netket.vqs.freeze_parameters

netket.vqs.freeze_parameters(vstate, is_frozen)

Freeze a subset of model parameters in a variational state, removing them from gradient/QGT calculations.

Frozen parameters are moved from vstate.parameters into vstate.model_state['frozen_params'].

This is a thin shim around netket.nn.freeze_parameters(), which reconstructs a new variational state.

Parameters:

• vstate (VariationalState) – A MCState or FullSumState.

• is_frozen (Callable[[tuple[str, ...], Any], bool]) – Callable (path, leaf) -> bool.

Return type:

VariationalState

Returns:

A new variational state of the same type as vstate.

Example:

import netket as nk

frozen_vstate = nk.vqs.freeze_parameters(
    vstate, lambda path, _: "kernel" in path
)