Backends

Check the Current Backend

from hdmrlib import get_backend

print(get_backend())

The library keeps one active backend at a time.

Select a Backend

from hdmrlib import set_backend

set_backend("numpy")
set_backend("torch")
set_backend("tensorflow")

Backend names are case-insensitive.

"tf" can also be used as an alias for "tensorflow".

set_backend("tf")

List Available Backends

from hdmrlib.backends import available_backends

print(available_backends())

This returns the backends that are currently available in the environment.

Default Backend

NumPy is used by default when available. Otherwise, the first available backend is selected.

Input Conversion

The active backend converts input data internally:

  • NumPy backend converts inputs to NumPy arrays

  • PyTorch backend converts inputs to Torch tensors

  • TensorFlow backend converts inputs to TensorFlow tensors

All backend implementations use float64 internally.

Missing Backends

If you select a backend that is not installed, the library raises an import error.

For example:

  • set_backend("torch") raises an error if the Torch backend is not available

  • set_backend("tensorflow") raises an error if the TensorFlow backend is not available

Notes

  • only one backend is active at a time

  • available backends depend on installed dependencies