WDM wavelet transform

This package hosts the python version of WDM wavelet used in coherentWaveBurst(cWB) search.

Installation

pip install wdm-wavelet

Example

Generate a timeseries waveform

from pycbc.waveform import get_td_waveform
import matplotlib.pyplot as plt

hp, hc = get_td_waveform(approximant="IMRPhenomTPHM",
                         mass1=20,
                         mass2=20,
                         spin1z=0.9,
                         spin2z=0.4,
                         inclination=1.23,
                         coa_phase=2.45,
                         distance=100,
                         delta_t=1.0/2048,
                         f_lower=20)

Apply WDM wavelet transform

from wdm_wavelet.wdm import WDM

wdm = WDM(32, 64, 6, 10, backend="jax")  # or backend="numba"

tf_map = wdm.t2w(hp)

tf_map.plot_energy()

Inverse WDM wavelet transform

ts = wdm.w2t(tf_map)

plt.plot(ts)

Supported class backends: "jax" (default) and "numba". Both JAX and Numba are required installation dependencies.

For more examples, please refer to the example notebook.

API docs

• API usage guide
• Explicit transform options
• Release integration assessment

Choosing a backend

from wdm_wavelet.wdm import WDM

wdm = WDM(16, 16, 6, 10, backend="numba")  # or backend="jax"

The backend selects the forward and inverse transform kernels for each instance. Numba executes on the CPU; JAX uses its configured device. Both implementations are installed together, so changing backend requires no dependency changes. Missing dependencies fail at import rather than silently selecting slower kernels.

WDM.filter remains a float64 JAX array for both backends, as in v0.3.1. Numba wrappers convert inputs to NumPy arrays at the kernel boundary. NumPy remains a dependency for array storage and numerical operations; it is not a separate execution backend. Class inverse methods return GWpy TimeSeries.