pymc-devs / pymc-devs/pytensor
Rewrite homogenous repeat vector
Open
Nobody has claimed this yet.
graph rewriting
performance
- Dominant language
- Python
- Stars
- 644
- Forks
- 208
- Avg merge
- 2d 14h
- Merged PRs (30d)
- 16
Description
Description
If the number of repetitions in pt.repeat is homogeneous, we could replace it by repeat(x, unique_value). Under the hood this ends up as an alloc, skipping the non-C Op altogether
import numpy as np
x = np.arange(600)
r1 = np.array(2, dtype=int)
r2 = np.array([2] * 600, dtype=int)
np.testing.assert_allclose(np.repeat(x, r1), np.repeat(x, r2))
%timeit np.repeat(x, r1) # 1.32 μs ± 3.21 ns per loop (mean ± std. dev. of 7 runs, 1,000,000 loops each)
%timeit np.repeat(x, r2) # 1.85 μs ± 3.69 ns per loop (mean ± std. dev. of 7 runs, 1,000,000 loops each)
Not a high priority issue
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Start at the pt.repeat entry point and compare the homogeneous repetition case with the scalar repeat path shown in the issue. Confirm the optimization preserves np.repeat equivalence for scalar and length-600 repetition inputs, and verify that the non-C Op is skipped as intended.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- numpy, python
- Domain
- performance
- Issue type
- Refactor
- Difficulty
- 3/5
- Estimated time
- 1-2 days
- Activity status
- Stale
- Clarity
- Mostly clear
- Newbie friendliness
- 45/100