BluSky is a Python library for that implements the Mallat wavelet scattering transform using Keras/Tensorflow. Features include:
- 1D, 2D transforms. 3D on the way.
- arbitrary order
- Morlets & Gabor, modular library allows for arbitrary wavelets(?)
- built in visualization of transform coefficients
Install the dependencies in a Conda environment:
conda create -n blusky python=3.8.5 traits matplotlib scikit-learn
conda activate blusky
python -m pip install tensorflow==2.4
Install BluSky from source:
git clone git@github.com:enthought/blusky.git
cd blusky
python -m pip install -e .
- wavelet scattering transform, implemented as a series of convolutions using Keras and Tensorflow.
- following the approach of Bruna, and Mallat (refs)
- the transform is implemented as a convolutional neural network. intermediate activations are easily interrogated to enable easy interrogation of the signal as it propagates through the cascade
This effort was supported by Sandia National Labs with development and maintenance support by Enthought.
