Skip to content

RF: Rearranging matmul order and using hermitian flag in ICC_rep_anova #3453

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 8 commits into from
Apr 21, 2022
Merged

RF: Rearranging matmul order and using hermitian flag in ICC_rep_anova #3453

Changes from 6 commits
Commits
Show all changes
8 commits
Select commit Hold shift + click to select a range
e3d08bd
np.dot(X.T, T) is Hermitian
Apr 20, 2022
9f2fd97
Reordering matmul operations
Apr 20, 2022
49b7ed5
Using @ operator instead of np.dot
Apr 21, 2022
1b2eddd
Merge branch 'nipy:master' into linalg-speedup
kristoferm94 Apr 21, 2022
835dfe7
Merge branch 'nipy:master' into linalg-speedup
kristoferm94 Apr 21, 2022
883a84a
revert
kristoferm94 Apr 21, 2022
2aaa8a5
removing unneccesary import
Apr 21, 2022
9cb58c3
Update .pre-commit-config.yaml
kristoferm94 Apr 21, 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
2 changes: 1 addition & 1 deletion nipype/algorithms/icc.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -114,7 +114,7 @@ def ICC_rep_anova(Y):
X = hstack([x, x0])

# Sum Square Error
predicted_Y = dot(dot(dot(X, pinv(dot(X.T, X))), X.T), Y.flatten("F"))
predicted_Y = X @ (pinv(X.T @ X, hermitian=True) @ (X.T @ Y.flatten("F")))
Copy link
Member

Choose a reason for hiding this comment

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

For a simulated Y of shape 50 x 10, I get:

In []: %timeit dot(dot(dot(X, pinv(dot(X.T, X))), X.T), Y.flatten("F"))
1.42 ms ± 128 µs per loop (mean ± std. dev. of 7 runs, 1,000 loops each)

Convert to matmul:

In []: %timeit (X @ pinv(X.T @ X) @ X.T) @ Y.flatten("F")
1.4 ms ± 188 µs per loop (mean ± std. dev. of 7 runs, 1,000 loops each)

No change. Reorder multiplications:

In []: %timeit X @ (pinv(X.T @ X) @ (X.T @ Y.flatten("F")))
838 µs ± 43.3 µs per loop (mean ± std. dev. of 7 runs, 1,000 loops each)

41% speedup. Use hermetian optimization:

In []: %timeit X @ (pinv(X.T @ X, hermitian=True) @ (X.T @ Y.flatten("F")))
785 µs ± 17.8 µs per loop (mean ± std. dev. of 7 runs, 1,000 loops each)

6% relative speedup for total of 45%.

residuals = Y.flatten("F") - predicted_Y
SSE = (residuals**2).sum()

Expand Down