def get_run_loop(self, show_pbar: bool = None) -> Iterable[int]:
"""
Return a tqdm progress bar or a regular range iterator.
If the code is running in an IPython kernel it will also display the \
internal ``_notebook_container``.
Args:
show_pbar: If ``False`` the progress bar will not be displayed.
Returns:
A Progressbar if ``show_pbar`` is ``True`` and the code is running \
in an IPython kernel. If the code is running in a terminal the logging \
level must be set at least to "INFO". Otherwise return a range iterator \
for ``self.max_range`` iteration.
"""
show_pbar = show_pbar if show_pbar is not None else self.show_pbar
use_tqdm = (
show_pbar if running_in_ipython() else self._log.level < logging.WARNING and show_pbar
)
loop_iterable = (
trange(self.max_epochs, desc="%s" % self.__class__.__name__)
if use_tqdm
else range(self.max_epochs)
)
if running_in_ipython() and self._use_notebook_widget:
from IPython.core.display import display
display(self._notebook_container)
return loop_iterable
def report_progress(self):
"""Report information of the current run."""
if running_in_ipython() and self._use_notebook_widget:
line_break = '<br style="line-height:1px; content: " ";>'
html = str(self).replace("\n\n", "\n").replace("\n", line_break)
# Add strong formatting for headers
html = html.replace("Walkers States", "<strong>Walkers States</strong>")
html = html.replace("Model States", "<strong>Model States</strong>")
html = html.replace("Environment States", "<strong>Environment Model</strong>")
self._notebook_container.value = "%s" % html
elif not running_in_ipython():
self._log.info(repr(self))
def setup_notebook_container(self):
"""Display the display widgets if the Swarm is running in an IPython kernel."""
if running_in_ipython() and self._use_notebook_widget:
from ipywidgets import HTML
from IPython.core.display import display, HTML as cell_html
# Set font weight of tqdm progressbar
display(cell_html("<style> .widget-label {font-weight: bold !important;} </style>"))
self._notebook_container = HTML()
def __init__(self,
n_walkers: int,
env: Callable[[], BaseEnvironment],
model: Callable[[BaseEnvironment], BaseModel],
walkers: Callable[..., Walkers] = Walkers,
reward_scale: float = 1.0,
distance_scale: float = 1.0,
tree: Callable[[], HistoryTree] = None,
report_interval: int = numpy.inf,
show_pbar: bool = True,
use_notebook_widget: bool = True,
force_logging: bool = False,
*args,
**kwargs):
"""
Initialize a :class:`Swarm`.
Args:
n_walkers: Number of walkers of the swarm.
env: A callable that returns an instance of an :class:`Environment`.
model: A callable that returns an instance of a :class:`Model`.
walkers: A callable that returns an instance of :class:`BaseWalkers`.
reward_scale: Virtual reward exponent for the reward score.
distance_scale: Virtual reward exponent for the distance score.
tree: class:`StatesTree` that keeps track of the visited states.
report_interval: Display the algorithm progress every ``report_interval`` epochs.
show_pbar: If ``True`` A progress bar will display the progress of \
the algorithm run.
use_notebook_widget: If ``True`` and the class is running in an IPython \
kernel it will display the evolution of the swarm \
in a widget.
force_logging: If ``True``, disable al ``ipython`` related behaviour.
*args: Additional args passed to init_swarm.
**kwargs: Additional kwargs passed to init_swarm.
"""
self._prune_tree = False
self._epoch = 0
self.show_pbar = show_pbar
self.report_interval = report_interval
super(Swarm, self).__init__(walkers=walkers,
env=env,
model=model,
n_walkers=n_walkers,
reward_scale=reward_scale,
distance_scale=distance_scale,
tree=tree,
*args,
**kwargs)
self._notebook_container = None
self._use_notebook_widget = use_notebook_widget
self._ipython_mode = running_in_ipython() and not force_logging
self.setup_notebook_container()
def __init__(self,
n_walkers: int,
step_epochs: int = None,
root_model: Callable[[], RootModel] = MajorityDiscreteModel,
tree: Callable[[], BaseTree] = None,
report_interval: int = numpy.inf,
show_pbar: bool = True,
walkers: Callable[..., StepWalkers] = StepWalkers,
swarm: Callable[..., Swarm] = Swarm,
reward_limit: Scalar = None,
max_epochs: int = None,
accumulate_rewards: bool = True,
minimize: bool = False,
use_notebook_widget: bool = True,
force_logging: bool = False,
*args,
**kwargs):
"""
Initialize a :class:`StepSwarm`.
This class can be initialized the same way as a :class:`Swarm`. All the \
parameters except ``max_epochs`` and ``tree`` will be used to initialize \
the internal swarm, and the :class:`StepSwarm` will use them when necessary.
The internal swarm will be initialized with no tree, and with its \
notebook widgets deactivated.
Args:
n_walkers: Number of walkers of the internal swarm.
step_epochs: Number of epochs that the internal swarm will be run \
before sampling an action.
root_model: Callable that returns a :class:`RootModel` that will be \
used to sample the actions and dt of the root walker.
tree: Disabled for now. It will be used by the root walker.
report_interval: Display the algorithm progress every \
``report_interval`` epochs.
show_pbar: If ``True`` A progress bar will display the progress of \
the algorithm run.
walkers: A callable that returns an instance of :class:`StepWalkers`.
swarm: A callable that returns an instance of :class:`Swarm` and \
takes as input all the corresponding parameters provided. It \
will be wrapped with a :class:`StoreInitAction` before \
assigning it to the ``internal_swarm`` attribute.
reward_limit: The algorithm run will stop after reaching this \
reward value. If you are running a minimization process \
it will be considered the minimum reward possible, and \
if you are maximizing a reward it will be the maximum \
value.
max_epochs: Maximum number of steps that the root walker is allowed \
to take.
accumulate_rewards: If ``True`` the rewards obtained after transitioning \
to a new state will accumulate. If ``False`` only the last \
reward will be taken into account.
minimize: If ``True`` the algorithm will perform a minimization \
process. If ``False`` it will be a maximization process.
use_notebook_widget: If ``True`` and the class is running in an IPython \
kernel it will display the evolution of the swarm \
in a widget.
force_logging: If ``True``, disable al ``ipython`` related behaviour.
*args: Passed to ``swarm``.
**kwargs: Passed to ``swarm``.
"""
self.internal_swarm = StoreInitAction(
swarm(max_epochs=step_epochs,
show_pbar=False,
report_interval=numpy.inf,
n_walkers=n_walkers,
tree=None,
walkers=walkers,
accumulate_rewards=accumulate_rewards,
minimize=minimize,
use_notebook_widget=False,
*args,
**kwargs))
self.internal_swarm.reset()
self.root_model: RootModel = root_model()
if reward_limit is None:
reward_limit = -numpy.inf if self.internal_swarm.walkers.minimize else numpy.inf
self.accumulate_rewards = accumulate_rewards
self._max_epochs = int(max_epochs)
self.reward_limit = reward_limit
self.show_pbar = show_pbar
self.report_interval = report_interval
self.tree = tree() if tree is not None else tree
self._epoch = 0
self._walkers: StepWalkers = self.internal_swarm.walkers
self._model = self.internal_swarm.model
self._env = self.internal_swarm.env
self.cum_reward = numpy.NINF
self.minimize = minimize
self.root_model_states = self.walkers.model_states[0]
self.root_env_states = self.walkers.env_states[0]
self.root_walkers_states = self.walkers.states[0]
self.root_walker = OneWalker(
reward=self.root_env_states.rewards[0],
observ=self.root_env_states.observs[0],
state=self.root_env_states.states[0],
)
self._notebook_container = None
self._use_notebook_widget = use_notebook_widget
self._ipython_mode = running_in_ipython() and not force_logging
self.setup_notebook_container()
