pytorch_pfn_extras.training.extensions.Slack#

class pytorch_pfn_extras.training.extensions.Slack(channel, msg=None, *, start_msg='{default}', end_msg='{default}', error_msg='{default}', thread=True, filenames=None, upload_trigger=None, context=None, token=None)#

Bases: _SlackBase

An extension to communicate with Slack.

Example

>>> ppe.training.extensions.Slack(
...     channel="experiment-progress",
...     msg="Epoch #{manager.epoch}: loss = {val/loss}",
...     end_msg="{default} \n <@username> Check out the result!",
...
...     # Upload files at the end of the experiment.
...     filenames=["result/statistics.png"],
...     upload_trigger=(max_epoch, 'epoch'),
... )

This extension posts a message when:

  • start_msg: The training has started

  • msg: The extension is triggered, usually at the end of each epoch

  • end_msg: The training has finished

  • error_msg: An exception has raised during the training

These messages can be specified as a format string, a callable that returns a string, or None to disable posting on that event.

When using a format string, the following variables are available for use:

  • manager: an ExtensionsManager object

  • default: the default message string

  • context: an arbitrary object passed to this extension

  • error: an Exception object (for error_msg only)

  • All reported values (manager.observations)

When using a callable, it should take (ExtensionsManager, context) or (ExtensionsManager, Exception, context) (for error_msg) and return a string.

This extension can upload files along with the message when triggered. filenames can be a list of filenames (the same formatting rule as msg apply), or a callable taking (ExtensionsManager, context) and returning a list of filenames.

To use this extension, you must create a Slack app, then specify the token via an environment variable SLACK_BOT_TOKEN or token option.

Parameters:

  • channel (str) – The channel where messages and files will be sent. This can be a channel name or a channel ID.

  • msg (str, callable, or None) – A message to be sent when triggered. It can be a string to be formatted using .format or a callable that returns a string.

  • start_msg (str, callable, or None) – A message to be sent at the beginning of the experiment.

  • end_msg (str, callable, or None) – A message to be sent at the completion of the experiment.

  • error_msg (str, callable, or None) – A message to be sent when an exception is raised during the experiment.

  • thread (bool) – When True, subsequent messages will be posted as a thread of the original message. Default is True.

  • filenames (list of str or callable) – A list of files that will be uploaded. These are string templates that can take values in the same way as msg, or a callable that returns a list of filenames.

  • upload_trigger (trigger or None) – Used to upload files at certain events. If not specified, files will be uploaded in every call.

  • context (Any) – Any arbitrary user object you will need when generating a message.

  • token (str) – Slack bot token. If None, the environment variable SLACK_BOT_TOKEN will be used. Optional, default is None.

Methods

__init__(channel[, msg, start_msg, end_msg, ...])

default_end_msg(context)

default_error_msg(exc, context)

default_msg(context)

default_start_msg(context)

finalize(manager)

Finalizes the extension.

initialize(manager)

Initializes up the manager state.

load_state_dict(to_load)

on_error(manager, exc, tb)

Handles the error raised during training before finalization.

state_dict()

Serializes the extension state.

Attributes

default_name

Default name of the extension.

is_async

name

needs_model_state

priority

trigger

__init__(channel, msg=None, *, start_msg='{default}', end_msg='{default}', error_msg='{default}', thread=True, filenames=None, upload_trigger=None, context=None, token=None)#
Parameters:
Return type:

None

trigger: Optional[Union[Trigger, Callable[[ExtensionsManagerProtocol], bool], Tuple[float, str]]] = (1, 'epoch')#