moten.backend package

Submodules

• moten.backend.numpy module
  • column_stack()
  • eigh()
  • to_numpy()
• moten.backend.torch module
• moten.backend.torch_cuda module
• moten.backend.torch_mps module

Module contents

moten.backend.benchmark(backend=None, nimages=100, vdim=96, hdim=128, stimulus_fps=24)

Benchmark motion energy computation across one or more backends.

Runs a small motion energy pyramid projection and reports the wall-clock time for each backend. Useful for comparing CPU vs GPU performance.

Parameters:

• backend (str or None) – Name of a single backend to benchmark, or None to benchmark all available backends.

• nimages (int) – Number of video frames in the test stimulus.

• vdim (int) – Vertical dimension of the test stimulus (pixels).

• hdim (int) – Horizontal dimension of the test stimulus (pixels).

• stimulus_fps (int) – Stimulus frame rate.

Returns:

results – Dictionary mapping backend name to a dict with keys: - duration_seconds: wall-clock time for per-filter projection - duration_batched_seconds: wall-clock time for batched projection - speedup: ratio of per-filter to batched duration - nimages: number of frames processed - vhsize: (vdim, hdim) - nfilters: number of filters in the pyramid

Return type:

dict

Examples

>>> from moten.backend import benchmark
>>> results = benchmark("numpy")
>>> print(f"numpy: {results['numpy']['duration_seconds']:.3f}s")

moten.backend.get_backend()

Get the current backend module.

Returns:

module – Module of the backend.

Return type:

python module

moten.backend.set_backend(backend, on_error='raise')

Set the backend using a global variable, and return the backend module.

Parameters:

• backend (str or module) – Name or module of the backend.

• on_error (str in {"raise", "warn"}) – Define what is done if the backend fails to be loaded. If “warn”, this function only warns, and keeps the previous backend. If “raise”, this function raises on errors.

Returns:

module – Module of the backend.

Return type:

python module