drytorch.trackers.visdom

Module containing a tracker for visualizations on visdom.

Classes

VisdomOpts

Annotations for optional settings in visdom.

VisdomPlotter([server, port, opts, ...])

Tracker that uses visdom as backend.
class VisdomOpts[source]

Bases: TypedDict

Annotations for optional settings in visdom.

See: https://github.com/fossasia/visdom/blob/master/py/visdom/__init__.py.

class VisdomPlotter(server: str = 'http://localhost', port: int = 8097, opts: VisdomOpts | None = None, model_names: Iterable[str] = (), source_names: Iterable[str] = (), metric_names: Iterable[str] = (), metric_loader: MetricLoader | None = None, start: int = 1)[source]

Bases: BasePlotter[str]

Tracker that uses visdom as backend.

server

the address for the visdom server.

Type:

str

port

the port for the server.

Type:

int

opts

plot options.

Type:

drytorch.trackers.visdom.VisdomOpts

Initialize.

Parameters:

  • server (str) – the address of the server.

  • port (int) – the port for the connection:

  • opts (VisdomOpts) – plot options.

  • model_names (Iterable[str]) – the names of the models to plot. Defaults to all.

  • metric_names (Iterable[str]) – the names of the metrics to plot. Defaults to all.

  • source_names (Iterable[str]) – the names of the sources to plot. Defaults to all.

  • metric_loader (MetricLoader | None) – a tracker that can load metrics from a previous run.

  • start (int) – if positive, the epoch from which to start plotting; if negative, the last number of epochs. Defaults to all.

property viz: Visdom

The active Visdom instance.

Raises:

AccessOutsideScopeError – if no run has been started yet.

clean_up() → None[source]

Override to clean up the tracker.

Return type:

None

notify(event: Event) → None[source]
notify(event: StartExperimentEvent) → None

Notify the tracker of an event.

Parameters:

event (Event) – the event to notify about.

Return type:

None