Skip to content

Fix: sepal numba compilation options changes the results significantly #1045

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 2 commits into
base: main
Choose a base branch
from
Open

Fix: sepal numba compilation options changes the results significantly #1045

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
ff51a53
add numerical stability
selmanozleyen Oct 1, 2025
fbc31ad
simplify unnecessary parameters and arrays
selmanozleyen Oct 1, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
22 changes: 8 additions & 14 deletions src/squidpy/gr/_sepal.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -202,30 +202,28 @@ def _score_helper(
@njit(fastmath=True)
def _diffusion(
conc: NDArrayA,
laplacian: Callable[[NDArrayA, NDArrayA, NDArrayA], float],
laplacian: Callable[[NDArrayA, NDArrayA], float],
n_iter: int,
sat: NDArrayA,
sat_idx: NDArrayA,
unsat: NDArrayA,
unsat_idx: NDArrayA,
dt: float = 0.001,
D: float = 1.0,
thresh: float = 1e-8,
) -> float:
"""Simulate diffusion process on a regular graph."""
sat_shape, conc_shape = sat.shape[0], conc.shape[0]
entropy_arr = np.zeros(n_iter)
prev_ent = 1.0
nhood = np.zeros(sat_shape)
weights = np.ones(sat_shape)

for i in range(n_iter):
for j in range(sat_shape):
nhood[j] = np.sum(conc[sat_idx[j]])
d2 = laplacian(conc[sat], nhood, weights)
d2 = laplacian(conc[sat], nhood)

dcdt = np.zeros(conc_shape)
dcdt[sat] = D * d2
dcdt[sat] = d2
conc[sat] += dcdt[sat] * dt
conc[unsat] += dcdt[unsat_idx] * dt
# set values below zero to 0
Expand All @@ -246,16 +244,13 @@ def _diffusion(
def _laplacian_rect(
centers: NDArrayA,
nbrs: NDArrayA,
h: float,
) -> NDArrayA:
"""
Five point stencil approximation on rectilinear grid.

See `Wikipedia <https://en.wikipedia.org/wiki/Five-point_stencil>`_ for more information.
"""
d2f: NDArrayA = nbrs - 4 * centers
d2f = d2f / h**2

return d2f


Expand All @@ -264,7 +259,6 @@ def _laplacian_rect(
def _laplacian_hex(
centers: NDArrayA,
nbrs: NDArrayA,
h: float,
) -> NDArrayA:
"""
Seven point stencil approximation on hexagonal grid.
Expand All @@ -275,10 +269,7 @@ def _laplacian_hex(
Curtis D. Benster, L.V. Kantorovich, V.I. Krylov,
ISBN-13: 978-0486821603.
"""
d2f: NDArrayA = nbrs - 6 * centers
d2f = d2f / h**2
d2f = (d2f * 2) / 3

d2f: NDArrayA = (2.0 * nbrs - 12.0 * centers) / 3.0
return d2f


Expand All @@ -290,8 +281,11 @@ def _entropy(
"""Get entropy of an array."""
Copy link
Member

@flying-sheep flying-sheep Oct 2, 2025 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This is what that does right?

Suggested change
"""Get entropy of an array."""
"""Compute Shannon entropy of an array of probability values (in nats)."""

Nats instead of Shannons, because you use the natural logarithm instead of log2.

Maybe also quote why the result is 0 if the sum is 0? https://stats.stackexchange.com/a/433096

xnz = xx[xx > 0]
xs: np.float64 = np.sum(xnz)
if xs == 0:
return 0.0
eps = np.finfo(np.float64).eps # ~2.22e-16
xn = xnz / xs
xl = np.log(xn)
xl = np.log(np.maximum(xn, eps))
return float((-xl * xn).sum())


Expand Down