choice.DeepValueFunction
Guide & worked example · Source
choice.DeepValueFunction(state_dim, n_choices, hidden_dims=[128, 64])Neural network value function approximation for high-dimensional states.
Enables dynamic choice estimation when |X| > 10,000 makes tabular methods intractable. Uses deep Q-network (DQN) style architecture.
References: Rawat & Rust (2025), pp. 16-17: “Deep RL can indeed overcome the curse of dimensionality that limits traditional Dynamic Programming”
Attributes
| Name | Description |
|---|---|
network |
Methods
| Name | Description |
|---|---|
| forward | Compute value function for each choice at given states. |
forward
choice.DeepValueFunction.forward(state)Compute value function for each choice at given states.
Args: state: (batch, state_dim) continuous states
Returns: values: (batch, n_choices) choice-specific values v̄(s,a)