A batch of small efficiency edits suggested by codex.#192

Merged

rmjarvis merged 30 commits into

mainfrom

efficiency

Feb 28, 2026

Commits on Feb 25, 2026

More efficient implementation of _sum functions
rmjarvis
committed
Use start, stop limits rather than an array for my_indices in MPI job determination
rmjarvis
committed
Use reverse index range rather than reversing an instantiated list
rmjarvis
committed

Commits on Feb 26, 2026

Use python sum for scalar accumulation
rmjarvis
committed
Use _zero_copy rather than template for the classes that were not doing so yet.
rmjarvis
committed
Use a more efficient algorithm for finding the indices for each patch in a Catalog.
rmjarvis
committed
Remove unused variable.
rmjarvis
committed
Add reader add_col method and mildly optimize it for HDF
rmjarvis
committed
Add nrows param to make_cov_design_matrix so plist doesn't need to be instantiated as an actual list
rmjarvis
committed
Add calc_weights=False option for make_cov_design_matrix to skip weights when unneeded.
rmjarvis
committed
Pull an arange call outside of a loop, so it doesn't get recomputed many times.
rmjarvis
committed
Use a slightly faster build of the design matrix when func is None
rmjarvis
committed
Avoid making list of keys for logging output when logger won't use them.
rmjarvis
committed
Mark objects as corr_only when building from _sum(..., corr_only=True)
rmjarvis
committed
any/all can take generators rather than lists
rmjarvis
committed
Remove unused variables, imports
rmjarvis
committed
Add linter preferences to agents.md
rmjarvis
committed
Switch to unordered_set in Floyd's algorithm, since don't need ordering.
rmjarvis
committed
More efficient version of getAllLeaves and getAllIndices
rmjarvis
committed
Use references rather than copy when trivially equivalent.
rmjarvis
committed
Use swap rather than copy when updating centers in kmeans algo
rmjarvis
committed
Vector with sort is a bit faster for this selection algorithm
rmjarvis
committed
Use emplace_back rather than push_back when we're constructing the object just to add it to the vector
rmjarvis
committed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

A batch of small efficiency edits suggested by codex.#192

A batch of small efficiency edits suggested by codex.#192
rmjarvis merged 30 commits into
mainfrom
efficiency

Commits on Feb 25, 2026

More efficient implementation of _sum functions

Use start, stop limits rather than an array for my_indices in MPI job determination

Use reverse index range rather than reversing an instantiated list

Commits on Feb 26, 2026

Use python sum for scalar accumulation

Use _zero_copy rather than template for the classes that were not doing so yet.

Use a more efficient algorithm for finding the indices for each patch in a Catalog.

Remove unused variable.

Add reader add_col method and mildly optimize it for HDF

Add nrows param to make_cov_design_matrix so plist doesn't need to be instantiated as an actual list

Add calc_weights=False option for make_cov_design_matrix to skip weights when unneeded.

Pull an arange call outside of a loop, so it doesn't get recomputed many times.

Use a slightly faster build of the design matrix when func is None

Avoid making list of keys for logging output when logger won't use them.

Mark objects as corr_only when building from _sum(..., corr_only=True)

any/all can take generators rather than lists

Remove unused variables, imports

Add linter preferences to agents.md

Switch to unordered_set in Floyd's algorithm, since don't need ordering.

More efficient version of getAllLeaves and getAllIndices

Use references rather than copy when trivially equivalent.

Use swap rather than copy when updating centers in kmeans algo

Vector with sort is a bit faster for this selection algorithm

Use emplace_back rather than push_back when we're constructing the object just to add it to the vector

Commits on Feb 27, 2026

Use empty() rather than size()==0

More efficient initialize centers for the random case

Allocate temp vector once and reuse

Multiply u once rather than divide p[k] many times.

Keep track of min_dsq in KPP initialization rather than recalculate each time from scratch.

Fix bug in KMPP initialization. Should only consider already selected centers for dups, not the whole vector

coverage