The class abstracts some operations of value / policy iteration in order to generalize to various types of robust MDPs. More...
#include <robust_values.hpp>
Inheritance diagram for craam::algorithms::PolicyNature< T >:
Public Types | |
| using | solution_type = SolutionRobust |
 Public Types inherited from craam::algorithms::PolicyDeterministic | |
| using | solution_type = Solution |
Public Member Functions | |
| PolicyNature (indvec policy, vector< NatureInstance< T >> natspec) | |
| Constructs the object from a policy and a specification of nature. | |
| PolicyNature (vector< NatureInstance< T >> natspec) | |
| Constructs the object from a policy and a specification of nature. | |
| SolutionRobust | new_solution (size_t statecount, numvec valuefunction) const |
| Constructs a new robust solution. | |
| template<class SType > | |
| prec_t | update_solution (SolutionRobust &solution, const SType &state, long stateid, const numvec &valuefunction, prec_t discount) const |
| Computes the Bellman update and updates the solution to the best response It does not update the value function. More... | |
| template<class SType > | |
| prec_t | update_value (const SolutionRobust &solution, const SType &state, long stateid, const numvec &valuefunction, prec_t discount) const |
| Computes a fixed Bellman update using the current solution policy. More... | |
| Public Member Functions inherited from craam::algorithms::PolicyDeterministic | |
| PolicyDeterministic () | |
| All actions will be optimized. | |
| PolicyDeterministic (indvec policy) | |
| A partial policy that can be used to fix some actions policy[s] = -1 means that the action should be optimized in the state policy of length 0 means that all actions will be optimized. | |
| Solution | new_solution (size_t statecount, numvec valuefunction) const |
| template<class SType > | |
| prec_t | update_solution (Solution &solution, const SType &state, long stateid, const numvec &valuefunction, prec_t discount) const |
| Computed the Bellman update and updates the solution to the best response It does not update the value function. More... | |
| template<class SType > | |
| prec_t | update_value (const Solution &solution, const SType &state, long stateid, const numvec &valuefunction, prec_t discount) const |
| Computes a fixed Bellman update using the current solution policy. More... | |
Public Attributes | |
| vector< NatureInstance< T > > | natspec |
| Specification of natures response (the function that nature computes, could be different for each state) | |
| Public Attributes inherited from craam::algorithms::PolicyDeterministic | |
| indvec | policy |
| Partial policy specification (action -1 is ignored and optimized) | |
Additional Inherited Members | |
| Protected Member Functions inherited from craam::algorithms::PolicyDeterministic | |
| void | process_valuefunction (size_t statecount, numvec &valuefunction) const |
| indvec | process_policy (size_t statecount) const |
Detailed Description
template<class T>
class craam::algorithms::PolicyNature< T >
The class abstracts some operations of value / policy iteration in order to generalize to various types of robust MDPs.
It can be used in place of response in mpi_jac or vi_gs to solve robust MDP objectives.
Member Function Documentation
◆ update_solution()
template<class T>
template<class SType >
|
inline |
Computes the Bellman update and updates the solution to the best response It does not update the value function.
- Returns
- New value for the state
◆ update_value()
template<class T>
template<class SType >
|
inline |
Computes a fixed Bellman update using the current solution policy.
- Returns
- New value for the state
The documentation for this class was generated from the following file:
- craam/algorithms/robust_values.hpp
