Benchmark results
These studies compare TamerOp's algebra and barcode computations with specialized libraries, and measure how long its live viewers take to update.
Results at a glance
| Comparison | Main result | What it computes |
|---|---|---|
| QPA: finite module algebra | TamerOp was 135.7× as fast across 200 completed comparisons. QPA failed on 20 tensor requests. | Maps, kernels, resolutions and related algebraic operations on finite modules. |
| PHAT: ordinary persistence | TamerOp took about 25% less time overall, with similar times on the algebraic controls. | Barcodes from supplied boundary matrices, which describe the cells and how their boundaries fit together. |
| Ripser.py: Vietoris–Rips persistence | TamerOp was 1.32× as fast overall on the development inputs. Sparse inputs and small controls favored Ripser.py. | Barcodes from points or distances, with requests that also select representative points or describe the detected classes. |
| C++ Ripser: ordinary Rips barcodes | TamerOp was about 1.8× as fast on circles. C++ Ripser was faster on sparse spheres. | Barcodes from supplied distance files, using larger point sets. |
| CubicalRipser: cubical barcodes | TamerOp was 2.18× as fast overall. Binary volumes with values assigned to vertices favored CubicalRipser. | Barcodes from images and volumes, with values assigned to vertices or to pixels and voxels. |
| Continuous inspection: time from input to picture | Median updates took 0.274 s for the ring, 4.97 s for the image and 0.582 s for a line restriction. | The wait from changing a browser control to seeing the figure update after Julia's computation. |
Here, 2× as fast means TamerOp took half as long. Each report shows the variation across inputs and the actual waiting times.
Computer and reproducibility
The studies use the same Intel Core i7-1365U workstation with 32 GB of installed memory. How the benchmarks were measured explains what was timed and how the comparisons were made, with the full computer specifications in one place.
Follow Reproduce a benchmark report to check the downloaded JSON against its SHA-256 fingerprint, redraw the figures or run the computation yourself.