Wavelet Codec

WaveletCodec compresses data by transforming it into a wavelet basis and keeping only the largest coefficients.

For background on the method, see the PyWavelets documentation.

Configuration

Set exactly one of keep_fraction or threshold.

Option	Type	Meaning
wavelet	str	PyWavelets wavelet name (default "db4"). Any discrete wavelet from pywt.wavelist(kind="discrete"), e.g. "haar", "db2", "sym8", "coif3", "bior4.4".
mode	str	PyWavelets signal extension mode (default "symmetric"). See pywt.Modes.modes.
keep_fraction	float in (0, 1]	Fraction of all coefficients to keep. Gives a predictable compression ratio.
threshold	float >= 0	Keep coefficients with magnitude greater than this value. Gives a predictable noise floor.

As with all codecs, constructor options act as defaults and can be overridden per call:

codec = WaveletCodec(wavelet="sym8", keep_fraction=0.05)
compress_patch(patch, codec)                      # 5% of coefficients
compress_patch(patch, codec, keep_fraction=0.01)  # 1% for this call only
compress_patch(patch, codec, wavelet="haar")      # different wavelet

Example

import numpy as np
import dascore as dc
from comdas import WaveletCodec, compress_patch

patch = dc.get_example_patch("example_event_2")

for wavelet in ("haar", "db4", "sym8"):
    codec = WaveletCodec(wavelet=wavelet, keep_fraction=0.05)
    compressed = compress_patch(patch, codec)
    error = np.linalg.norm(patch.data - np.asarray(compressed.data))
    error /= np.linalg.norm(patch.data)
    ratio = codec.compression_ratio(compressed.data.payload)
    print(f"{wavelet:5s} ratio={ratio:5.1f}x  relative error={error:.3f}")

API

See the module reference.