Support multidimensional HSGP (numpyro.contrib.hsgp) #1803
Conversation
wip(hsgp_nd): test list input for m, make dim required arg wip(hsgp_nd): support batched eigenfunctions, swap dims for m and d
|
Check out this pull request on See visual diffs & provide feedback on Jupyter Notebooks. Powered by ReviewNB |
doc(hsgp_nd): polish contrib documentation rm TODO fix eigenindices docstring cleanup nb
brendancooley
force-pushed
the
feat/hsgp-nd
branch
from
fda9312
to
8b064eb
Compare
|
^ fix pushed for python 3.9 typing issues causing CI failure |
|
^once more, fix for linting |
|
This looks super cool @brendancooley ! I this one ready for review? |
looks like matplotlib snuck in a deprecation here
There was a problem hiding this comment.
@brendancooley This is amazing! I could have never done this in a cleaner way 🙌 ! Thank you!
I left two suggestions regarding documentation and some additional simple unit-tests to make this bullet-proof and to help other devs get what are the auxiliary functions expected to output. Besides that is a ✅ from my side.
docs/source/contrib.rst
Outdated
| \times | ||
| \overbrace{\color{green}{\beta_{j}}}^{\sim \: \text{Normal}(0,1)} | ||
|
|
||
| where :math:`\lambda_j` are the eigenvalues of the Laplacian operator, :math:`\phi_{j}(x)` are the eigenfunctions of the | ||
| where :math:`\boldsymbol{x}` is a :math:`D` vector of inputs, :math:`\boldsymbol{\lambda}_j^\star` are the eigenvalues of the Laplacian operator, :math:`\phi_{j}(\boldsymbol{x})` are the eigenfunctions of the |
There was a problem hiding this comment.
can you define or explain what \boldsymbol{\lambda}_j^\ is :)
There was a problem hiding this comment.
good catch, thanks :)
numpyro/contrib/hsgp/laplacian.py
Outdated
| m1 = jnp.tile(w0 * x[:, None], m) | ||
| m2 = jnp.diag(jnp.arange(m, dtype=jnp.float32)) | ||
| mw0x = m1 @ m2 | ||
| cosines = jnp.cos(mw0x) | ||
| sines = jnp.sin(mw0x) | ||
| return cosines, sines | ||
|
|
||
|
|
||
| def _convert_ell(ell: float | list[float] | ArrayImpl, dim: int) -> ArrayImpl: |
There was a problem hiding this comment.
Could we add a unit-test for this function? It would be great for documenting how is supposed to work.
| import jax.numpy as jnp | ||
|
|
||
|
|
||
| # TODO: Adapt to dim >= 1. | ||
| def sqrt_eigenvalues(ell: float, m: int) -> ArrayImpl: | ||
| def eigenindices(m: list[int] | int, dim: int) -> ArrayImpl: |
There was a problem hiding this comment.
Could we add a unit-test for this function? It would be great for documenting how is supposed to work.
|
Regarding the TODOS:
|
|
@fehiepsi any hints on the examples CI failing because other tests not relevant to this PR 😄 ? |
Thanks the feedback @juanitorduz! Will push a commit including responses to your suggestions shortly. |
It looks like the |
issue for vector-valued |
Thanks again! |
There was a problem hiding this comment.
Beautiful work, @brendancooley!!
Thanks for reviewing, @juanitorduz!
eigenindicesfunction, following Eq 10 in Riutort-Mayol et aleigenfunctionsto handle multidimensionalx. Vector-valued inputs are treated as unidimensional problems. Otherwise the trailing dimension ofxis inferred as the dimension of the input space. All preceding dims are treated as batch dims.laplacianandapproximationfunctions to accept list-valued inputs formandell. If an int (m) or float (ell) is passed when the problem is multidimensional, the same value is used for all dimensions.The results in @juanitorduz notebook replicate with one change to function arguments.
TODO (on this PR or future)
numpyro.contrib.hsgpnumpyro.contrib.hsgp