s2fft
s2fft copied to clipboard
astro-informatics
Differentiable and accelerated spherical transforms with JAX
In the notebooks the generated flm test signals have imaginary components for `m=0`. These will just be ignored so it's fine but could be confusing. We should also check how...
Switches from current `setup.py` approach of specifying package metadata to instead use [a `pyproject.toml` configuration file](https://packaging.python.org/en/latest/guides/writing-pyproject-toml/) and sets up [`setuptools_scm`](https://github.com/pypa/setuptools_scm) to automatically set package version number from Git release tags....
Include functionality for existing plots in notebooks and also Mollweide plots.
Currently we force users to install both JAX and PyTorch to gain access to either. We should add a switch with optional dependencies to reduce the requirement overhead.
We should cache quadrature weights to avoid recalculating them. While the computation cost of quadrature weights is fairly low we should nevertheless avoid re-calculating. It should be possible to do...
I wrote a little test looking at the round-trip accuracy of the forward and inverse SHTs and noticed a surprising behaviour: accuracy is really good up to spin 5, and...
Currently back propagation is not being passed the precomputed array, and instead must compute it on a grad call. We can pass this in the residual.
Hi, Would it be possible to compute only the low frequencies from a high resolution grid? It seems to work with `sampling = "mw"` but not with `sampling = "dh"`:...
Metadata
Owner
Metadata
Differentiable and accelerated spherical transforms with JAX