How to interpret the learnableParameters (Reinforcement Learning Toolbox)?
I was wondering how to interpret the tables that one is receiving when using this line: params = getLearnableParameters (critic); Why are there 6 tables after q learning for example? I expected one.
John Michell answered .
2025-11-20
In MATLAB's Reinforcement Learning Toolbox, the `learnableParameters` are the weights and biases of the neural networks used in the agent, actor, or critic. These parameters are crucial as they determine the behavior of the agent during training and execution.
Interpretation of `learnableParameters`:
1. Weights and Biases: The `learnableParameters` typically consist of weights and biases that are adjusted during the training process to minimize the loss function. These parameters define how the input data is transformed within the neural network.
2. Critic Network: For example, if you're using a critic network, the `learnableParameters` will include the weights and biases of the neural network used to estimate the value function. These parameters are optimized to accurately predict the expected return from each state-action pair.
3. Actor Network: Similarly, for an actor network, the `learnableParameters` will include the weights and biases that determine the policy, i.e., the probability distribution over actions given a state.
4. Modifying Parameters: You can access and modify these parameters using functions like `getLearnableParameters` and `setLearnableParameters`. This allows you to inspect the learned values and make adjustments if necessary.
Example:
% Load a trained agent
load('DoubleIntegDDPG.mat', 'agent');
critic = getCritic(agent);
params = getLearnableParameters(critic);
disp(params);
In this example, `params` will contain the learnable parameters of the critic network, which you can then inspect or modify.