mpcrl.UpdateStrategy#
- class mpcrl.UpdateStrategy(frequency, hook='on_timestep_end', skip_first=0)[source]#
Bases:
objectA class holding information on the update strategy to be used by the learning algorithm.
- Parameters:
- frequencyint
Frequency at which, each time the hook is called, an update should be carried out.
- skip_firstint, optional
Skips the first
skip_firstupdates. By default0, so no update is skipped. This is useful when, e.g., the agent has to wait for the experience buffer to be filled before starting to update.- hook{“on_episode_end”, “on_timestep_end”}, optional
Specifies to which callback to hook, i.e., when to check if an update is due according to the given frequency. The options are:
"on_episode_end"checks if an update is due after each episode ends"on_timestep_end"checks for an update after each simulation’s time step.
By default,
"on_timestep_end"is selected.
Methods
Returns whether an update must be carried out now, at the current instant, according to the specified strategy.
- can_update()[source]#
Returns whether an update must be carried out now, at the current instant, according to the specified strategy.
- Returns:
- bool
Trueif the agent should update according to this strategy; otherwise,False.
- Return type:
Notes
This methods steps the internal iterators to check whether an update is due with
next. This means that calling this method has a side effect on the state of these iterators, and calling immediately again can result in a different outcome.
