qcritic

qcritic#

Classes#

class prt_rl.common.components.networks.qcritic.QCritic(network: Module, critic_head: Module)[source]#

QCritic is a neural network module for estimating Q-values in reinforcement learning.

This class composes a feature extraction network and a critic head to compute Q-values given observations and actions. It is typically used in actor-critic or value-based RL algorithms.

Parameters:

  • network (nn.Module) – Feature extractor network that processes observations.

  • critic_head (nn.Module) – Head network that takes features and actions to output Q-values.

Initialize the QCritic module.

Parameters:

  • network (nn.Module) – Feature extractor for observations.

  • critic_head (nn.Module) – Module that computes Q-values from features and actions.

__init__(network: Module, critic_head: Module)[source]#

Initialize the QCritic module.

Parameters:

  • network (nn.Module) – Feature extractor for observations.

  • critic_head (nn.Module) – Module that computes Q-values from features and actions.

forward(obs, action)[source]#

Forward pass to compute Q-values from observations and actions.

Parameters:

  • obs – Input observations (tensor or compatible type for network).

  • action – Actions to evaluate (tensor or compatible type for critic_head).

Returns:

Q-values estimated by the critic (tensor).

Contents