API Reference#

classtorch.monitor.Aggregation#

These are types of aggregations that can be used to accumulate stats.

Members:

VALUE :

VALUE returns the last value to be added.

MEAN :

MEAN computes the arithmetic mean of all the added values.

COUNT :

COUNT returns the total number of added values.

SUM :

SUM returns the sum of the added values.

MAX :

MAX returns the max of the added values.

MIN :

MIN returns the min of the added values.

propertyname#

classtorch.monitor.Stat#

Stat is used to compute summary statistics in a performant way overfixed intervals. Stat logs the statistics as an Event once everywindow_size duration. When the window closes the stats are loggedvia the event handlers as atorch.monitor.Stat event.

window_size should be set to something relatively high to avoid ahuge number of events being logged. Ex: 60s. Stat uses millisecondprecision.

Ifmax_samples is set, the stat will cap the number of samples perwindow by discardingadd calls oncemax_samples adds haveoccurred. If it’s not set, alladd calls during the window will beincluded. This is an optional field to make aggregations more directlycomparable across windows when the number of samples might vary.

When the Stat is destructed it will log any remaining data even if thewindow hasn’t elapsed.

__init__(self:torch._C._monitor.Stat,name:str,aggregations:collections.abc.Sequence[torch._C._monitor.Aggregation],window_size:datetime.timedelta,max_samples:SupportsInt=9223372036854775807)→None#

Constructs theStat.

add(self:torch._C._monitor.Stat,v:SupportsFloat)→None#

Adds a value to the stat to be aggregated according to theconfigured stat type and aggregations.

propertycount#

Number of data points that have currently been collected. Resetsonce the event has been logged.

get(self:torch._C._monitor.Stat)→dict[torch._C._monitor.Aggregation,float]#

Returns the current value of the stat, primarily for testingpurposes. If the stat has logged and no additional values have beenadded this will be zero.

propertyname#

The name of the stat that was set during creation.

classtorch.monitor.data_value_t#

data_value_t is one ofstr,float,int,bool.

classtorch.monitor.Event#

Event represents a specific typed event to be logged. This can representhigh-level data points such as loss or accuracy per epoch or morelow-level aggregations such as through the Stats provided through thislibrary.

All Events of the same type should have the same name so downstreamhandlers can correctly process them.

__init__(self:torch._C._monitor.Event,name:str,timestamp:datetime.datetime,data:collections.abc.Mapping[str,data_value_t])→None#

Constructs theEvent.

propertydata#

The structured data contained within theEvent.

propertyname#

The name of theEvent.

propertytimestamp#

The timestamp when theEvent happened.

classtorch.monitor.EventHandlerHandle#

EventHandlerHandle is a wrapper type returned byregister_event_handler used to unregister the handler viaunregister_event_handler. This cannot be directly initialized.

torch.monitor.log_event(event:torch._C._monitor.Event)→None#

log_event logs the specified event to all of the registered eventhandlers. It’s up to the event handlers to log the event out to thecorresponding event sink.

If there are no event handlers registered this method is a no-op.

torch.monitor.register_event_handler(callback:collections.abc.Callable[[torch._C._monitor.Event],None])→torch._C._monitor.EventHandlerHandle#

register_event_handler registers a callback to be called whenever anevent is logged vialog_event. These handlers should avoid blockingthe main thread since that may interfere with training as they runduring thelog_event call.

torch.monitor.unregister_event_handler(handler:torch._C._monitor.EventHandlerHandle)→None#

unregister_event_handler unregisters theEventHandlerHandle returnedafter callingregister_event_handler. After this returns the eventhandler will no longer receive events.

classtorch.monitor.TensorboardEventHandler(writer)[source]#

TensorboardEventHandler is an event handler that will write known events tothe provided SummaryWriter.

This currently only supportstorch.monitor.Stat events which are loggedas scalars.

Example

>>>fromtorch.utils.tensorboardimportSummaryWriter>>>fromtorch.monitorimportTensorboardEventHandler,register_event_handler>>>writer=SummaryWriter("log_dir")>>>register_event_handler(TensorboardEventHandler(writer))

__init__(writer)[source]#

Constructs theTensorboardEventHandler.