4.1.8 pde.backends.registry module

Defines the registry for managing backends.

BackendRegistry	Class handling all backends and their configurations.
load_default_config	Load a default configuration from a module.
get_backend	Return backend specified by string of instance.
registered_backends	Returns all registered backends.
available_backends	Returns all backends that can be used in the current environment.

class BackendRegistry

Bases: object

Class handling all backends and their configurations.

Backends can exist in three different states in registry: * Registered meta-information on how to load a backend package * Loaded backend module, so the class is available * Fully instantiated BackendBase classes

get_backend(name: str, *, config: ConfigLike | None = None, **kwargs) → pde.backends.base.BackendBase

Return backend object, potentially loading the respective package.

The returned backend is cached if config is not specified. Consequently, the same object will be returned for repeated calls to get_backend, which allows sharing configuration parameters. Moreover, if name only specifies a backend, i.e., does not contain a colon :, the configuration of this backend is linked with the global configuration pde.config, such that changes to the config are reflected globally.

Parameters:

• name (str) – Name of the backend to be loaded.

• config (dict) – Additional configuration options for this specific backend. The full configuration will be taken from the global configuration and merged with the given options here.

• **kwargs – Additional options of the backend

Returns:

An instance of the backend with the particular configuration

Return type:

BackendBase

is_available(name: str) → bool

Check whether a backend can be used in the current environment.

The check is based on the python modules that were given as requires when the backend was registered, so no backend is imported by this method.

Parameters:

name (str) – Name of the backend. Additional information given after a colon, e.g., in torch:cuda, is ignored, so this method only checks whether the backend itself could be loaded.

Returns:

Whether the backend can be used. Backends that are not registered at all are reported as being unavailable.

Return type:

bool

register_backend(backend: BackendBase) → None

Register a loaded backend object.

Parameters:

backend (BackendBase) – Implementation of the backend

register_class(name: str, cls: type[BackendBase])

Register a backend class.

Parameters:

• name (str) – Name of the backend

• cls (subclass of BackendBase) – The class for creating a backend

register_package(name: str, package_path: str, *, config: ConfigLike | Sequence[Parameter] | None = None, requires: Sequence[str] = ()) → None

Register a backend python package (without loading it yet)

Parameters:

• name (str) – Name of the backend

• package_path (str) – Import path for the package

• config (list) – Configuration options for the package

• requires (sequence of str) – Names of the python modules that need to be installed for this backend to work. They are used by BackendRegistry.is_available() to check whether the backend can be used without importing it.

values() → Iterator[pde.backends.base.BackendBase]

Iterate over all backends that can be imported.

available_backends() → list[str]

Returns all backends that can be used in the current environment.

In contrast to registered_backends(), this omits backends whose python packages are not installed, e.g., the torch backend if torch is missing.

Returns:

The names of all backends that can be loaded

Return type:

list of str

get_backend(backend: str | pde.backends.base.BackendBase, config: ConfigLike | None = None) → pde.backends.base.BackendBase

Return backend specified by string of instance.

The returned backend is cached if config is not specified. Consequently, the same object will be returned for repeated calls to get_backend, which allows sharing configuration parameters. Moreover, if name only specifies a backend, i.e., does not contain a colon :, the configuration of this backend is linked with the global configuration pde.config, such that changes to the config are reflected globally.

Parameters:

• backend (str or BackendBase) – Backend specified by name given as a string. If the string contains a colon, the first part determines the backend, whereas the second part can be used to convey additional information. For example, torch:cuda may load a torch backend and use a cuda device. As a special case, we also allow full backend objects, which are simply returned. This is a simple way to allow providing full backend objects in places where we otherwise would expect a backend name.

• config (dict) – Additional configuration options for this specific backend. The full configuration will be taken from the global configuration and merged with the given options here. This option is only permitted if backend is a string since there otherwise might be unintended side effects of modifying an existing backend.

Returns:

An initialized backend

Return type:

BackendBase

load_default_config(module_path: str | Path) → list[Parameter] | None

Load a default configuration from a module.

Parameters:

module_path (str) – String to the module to be loaded

registered_backends() → list[str]

Returns all registered backends.