Platform¶

Generic platform¶

class pfl.internal.platform.generic_platform.Platform¶

Collection of properties and methods related to compute platforms

abstractmethod consume_metrics(metrics, iteration, context=None)¶: Process task status and metrics.

abstractmethod create_checkpoint_directories(subdir_names)¶: Given a list of directories, creates the directories either in the current working directory if executed on a local computer or on a remote compute platform.

abstractmethod get_distributed_addresses(verbose=False)¶: Get all IP addresses of compute nodes when running in a distributed context.

abstractmethod get_platform_context()¶

Get a string-keyed dict of platform context information.

abstractmethod get_default_callbacks()¶

Get a list of callbacks to add to algorithms using this platform.

class pfl.internal.platform.generic_platform.GenericPlatform¶

consume_metrics(metrics, iteration, context=None)¶

Send metrics to platform, which may e.g. output them on the terminal or display on a dashboard.

Parameters:

metrics (Metrics) – Metrics to output.
iteration (int) – Current iteration.
context (Optional[Dict]) – An optional dict defining the context, e.g. hyperparameters used.

create_checkpoint_directories(subdir_names)¶

Given a list of directories, creates the directories in appropriate location on the current platform.

get_distributed_addresses(verbose=False)¶

Get all IP addresses of compute nodes when running in a distributed context.

get_platform_context()¶

Get a string-keyed dict of platform context information.

get_default_callbacks()¶

Get a list of callbacks to add to algorithms using this platform.

pfl.internal.platform.selector.get_platform()¶: Get the current platform. If the platform is not set, return the default platform pfl.internal.platform.GenericPlatform.

pfl.internal.platform.selector.set_platform(module)¶: Manually set the platform module. Can be useful when you are running on an unsupported platform and made a custom integration for it.