8000 MAINT Adapt `PairwiseDistancesReduction` heuristic for `strategy="auto"` by jjerphan · Pull Request #24043 · scikit-learn/scikit-learn · GitHub
[go: up one dir, main page]
Skip to content

MAINT Adapt PairwiseDistancesReduction heuristic for strategy="auto" #24043

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

Merged
merged 7 commits into from
Aug 11, 2022
Merged

MAINT Adapt PairwiseDistancesReduction heuristic for strategy="auto" #24043

Changes from all commits
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
06e733e
ASV Add benchmarks for PairwiseDistancesReductions
jjerphan Aug 5, 2022
fec7843
MAINT Adapt heuristic to take the number of samples of Y in
jjerphan Jul 29, 2022
09b7060
Remove old sub-optimal condition
jjerphan Aug 5, 2022
9c890df
Well, Julien
jjerphan Aug 5, 2022
5d0a337
MAINT Revert heuristic to use 2 conditions and document it
jjerphan Aug 11, 2022
7056ce2
MAINT Remove ASV benchmark
jjerphan Aug 11, 2022
9e5155e
Merge branch 'main' into maint/pdr-adapt-heuristic
jjerphan Aug 11, 2022
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
13 changes: 12 additions & 1 deletion sklearn/metrics/_pairwise_distances_reduction/_base.pyx.tp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -158,7 +158,18 @@ cdef class PairwiseDistancesReduction{{name_suffix}}:
if strategy == 'auto':
# This is a simple heuristic whose constant for the
# comparison has been chosen based on experiments.
if 4 * self.chunk_size * self.effective_n_threads < self.n_samples_X:
# parallel_on_X has less synchronization overhead than
# parallel_on_Y and should therefore be used whenever
# n_samples_X is large enough to not starve any of the
# available hardware threads.
if self.n_samples_Y < self.n_samples_X:
# No point to even consider parallelizing on Y in this case. This
# is in particular important to do this on machines with a large
# number of hardware threads.
strategy = 'parallel_on_X'
elif 4 * self.chunk_size * self.effective_n_threads < self.n_samples_X:
# If Y is larger than X, but X is still large enough to allow for
# parallelism, we might still want to favor parallelizing on X.
strategy = 'parallel_on_X'
else:
strategy = 'parallel_on_Y'
Expand Down
0