 |
Base class for all optimizers.
The Optimizer maintains a list of OptimizerStates which are updated each time the conformation is changed.
The optimizers have one key method Optimizer::optimize() which takes the number of steps to perform. The optimizers can have other stopping conditions as appropriate.
A typical optimization loop proceeds by:
Inheritance diagram for IMP::Optimizer:Public Member Functions | |
| Optimizer (Model *m, std::string name="Optimizer %1%") | |
| double | get_last_score () const |
| Return the score found in the last evaluate. | |
| Model * | get_model () const |
| Get the model being optimized. | |
| bool | get_stop_on_good_score () const |
| double | optimize (unsigned int max_steps) |
| void | set_model (Model *m) |
| Set the model being optimized. | |
| void | set_restraints (const RestraintsTemp &rs) |
| The optimizer can be told to use only a subset of the restraints to compute scores. | |
| void | set_stop_on_good_score (bool tf) |
| virtual void | show (std::ostream &out=std::cout) const |
| Print info about the optimizer state. | |
Macros to aid with implementing object classes | |
These macros are here to aid with implementing classes that inherit from the various abstract base classes in the kernel. Each Each IMP_OBJECT()/IMP_RESTRAINT()-style macro declares/defines the set of needed functions. The declared functions should be defined in the associated All of the IMP_OBJECT()/IMP_RESTRAINT()-style macros define the following methods:
In addition, they all declare:
For all macros, the Name parameter is the name of the class being implemented and the version_info parameter is the IMP::VersionInfo to use (probably get_version_info()). | |
| #define | IMP_OPTIMIZER(Name) |
| Define the basic things you need for an Optimizer. | |
States | |
The stored OptimizerState objects are updated each time the Optimizer decides to accept a new configuration of the Model. To manipulate the list of optimizer states use the methods below. | |
| void | remove_optimizer_state (OptimizerState *d) |
| template<class F > | |
| void | remove_optimizer_states_if (const F &f) |
| template<class List > | |
| void | remove_optimizer_states (List d) |
| template<class List > | |
| void | set_optimizer_states (List ps) |
| template<class List > | |
| void | set_optimizer_states_order (List ps) |
| unsigned int | add_optimizer_state (OptimizerState *obj) |
| template<class List > | |
| void | add_optimizer_states (List objs) |
| void | clear_optimizer_states () |
| unsigned int | get_number_of_optimizer_states () const |
| bool | get_has_optimizer_states () const |
| OptimizerState * | get_optimizer_state (unsigned int i) const |
| OptimizerStates | get_optimizer_states () const |
| void | reserve_optimizer_states (unsigned int sz) |
| OptimizerStateIterator | optimizer_states_begin () |
| OptimizerStateIterator | optimizer_states_end () |
| OptimizerStateConstIterator | optimizer_states_begin () const |
| OptimizerStateConstIterator | optimizer_states_end () const |
Methods for getting and setting optimized attributes | |
Optimizers don't have to go through the particles themselves looking for values to optimize unless they care about special properties of the optimized values. Instead they can iterate through the list of optimized attributes, each of which is identified by a FloatIndex. With these FloatIndex objects they can get and set the values and derivatives as needed. | |
| typedef Model::FloatIndex | FloatIndex |
| typedef vector< FloatIndex > | FloatIndexes |
| vector< Model::FloatIndex > | get_optimized_attributes () const |
| void | set_value (Model::FloatIndex fi, double v) const |
| Float | get_value (Model::FloatIndex fi) const |
| Float | get_derivative (Model::FloatIndex fi) const |
Methods to get and set scaled optimizable values | |
Certain optimizers benefit from having all the optimized values scaled to vary over a similar range. These accessors use the Model::get_range ranges to scale the values before returning them and unscale them before setting them. | |
| void | set_scaled_value (Model::FloatIndex fi, Float v) const |
| double | get_scaled_value (Model::FloatIndex fi) const |
| double | get_scaled_derivative (Model::FloatIndex fi) const |
| void | clear_range_cache () |
| Clear the cache of range information. Do this at the start of optimization. | |
Protected Member Functions | |
| virtual double | do_optimize (unsigned int ns)=0 |
| override this function to do actual optimization | |
| double | evaluate (bool compute_derivatives) const |
| double | evaluate_if_below (bool compute_derivatives, double max) const |
| RestraintsTemp | get_restraints () const |
| Return the restraint sets used in evaluation. | |
| void | update_states () const |
| Update optimizer state, should be called at each successful step. | |
| double | width (FloatKey k) const |
| void IMP::Optimizer::clear_range_cache | ( | ) | [protected] |
Clear the cache of range information. Do this at the start of optimization.
| virtual double IMP::Optimizer::do_optimize | ( | unsigned int | ns | ) | [protected, pure virtual] |
override this function to do actual optimization
| double IMP::Optimizer::evaluate | ( | bool | compute_derivatives | ) | const [protected] |
Evaluate the score of the model (or of a subset of the restraints if desired.
| double IMP::Optimizer::evaluate_if_below | ( | bool | compute_derivatives, |
| double | max | ||
| ) | const [protected] |
Evaluate the score of the model (or of a subset of the restraints if desired.
| double IMP::Optimizer::get_last_score | ( | ) | const |
Return the score found in the last evaluate.
| Model* IMP::Optimizer::get_model | ( | ) | const |
Get the model being optimized.
| RestraintsTemp IMP::Optimizer::get_restraints | ( | ) | const [protected] |
Return the restraint sets used in evaluation.
Use IMP::get_restraints() to get the actual restraints used.
| double IMP::Optimizer::optimize | ( | unsigned int | max_steps | ) |
Optimize the model
| [in] | max_steps | The maximum number of iterations of the optimizer to perform. Increasing this number will generally make the optimizer spend more time searching for a solution, but beyond that, the details of what changes will vary. |
| void IMP::Optimizer::set_model | ( | Model * | m | ) |
Set the model being optimized.
| void IMP::Optimizer::set_restraints | ( | const RestraintsTemp & | rs | ) |
The optimizer can be told to use only a subset of the restraints to compute scores.
These restraints must all be part of the model. Passing an empty list will revert to using the whole model.
| void IMP::Optimizer::set_stop_on_good_score | ( | bool | tf | ) |
Optimization can be stopped if all the thresholds in the Model are satisfied.
| virtual void IMP::Optimizer::show | ( | std::ostream & | out = std::cout | ) | const [virtual] |
Print info about the optimizer state.
It should end in a newline
Reimplemented from IMP::base::Object.
| void IMP::Optimizer::update_states | ( | ) | const [protected] |
Update optimizer state, should be called at each successful step.
| #define IMP_OPTIMIZER | ( | Name | ) | [related] |
IMP_IMPLEMENT(virtual Float do_optimize(unsigned int max_steps)); \ IMP_OBJECT(Name)
Define the basic things you need for an Optimizer.
In addition to the methods done by IMP_OBJECT, it declares
