WashU-Astroparticle-Lab / straxion / 21339757694

64%

Build Type

push

github

Committed by web-flow

Commit Message

Optimize computational performance (#77)

* Optimize hit classification with numba and FFT

Introduces numba-accelerated helper functions for moving average, chi-squared batch computation, and profile fitting. Refactors optimal filter logic to use pre-computed signal FFT, vectorized template generation via FFT time-shift property, and batch chi2 computation for improved performance. Debug output and kappa fitting also leverage optimized routines.

* Optimize DxRecords with numba and array-based interpolation

Replaces scipy interp1d with numba-accelerated linear interpolation using sorted arrays, adds numba-accelerated helper functions for IQ correction, convolution, and truth pulse injection, and updates DxRecords to use these optimizations. Updates tests to check for new attributes and array-based interpolation data instead of interpolation model objects.

* Accelerate DxHits hit finding with numba batch processing

Replaces the previous numpy-based hit candidate detection and per-hit processing in DxHits with numba-accelerated batch functions. This improves performance by using single-pass hit detection and batch computation of hit boundaries and amplitudes, reducing Python overhead and enabling faster processing of large signals.

* Add unit tests for numba-accelerated helper functions

Introduces comprehensive unit tests for numba-accelerated helper functions in DxHits and DxRecords, including hit candidate finding, hit boundary computation, linear interpolation, IQ correction, convolution, and truth pulse injection. These tests ensure correctness, edge case handling, and consistency with original implementations.

Run Details

253 of 302 new or added lines in 3 files covered. (83.77%)

5 existing lines in 1 file now uncovered.

1154 of 1846 relevant lines covered (62.51%)

0.63 hits per line