https://github.com/adamcrume/robust-binary-search

Robust Binary Search provides a binary search implementation which is robust against errors during the search.
https://github.com/adamcrume/robust-binary-search

Last synced: about 2 months ago
JSON representation

Robust Binary Search provides a binary search implementation which is robust against errors during the search.

• Host: GitHub
• URL: https://github.com/adamcrume/robust-binary-search
• Owner: adamcrume
• License: apache-2.0
• Created: 2020-10-24T03:33:05.000Z (almost 4 years ago)
• Default Branch: main
• Last Pushed: 2023-01-17T03:53:45.000Z (over 1 year ago)
• Last Synced: 2024-07-17T22:53:54.728Z (2 months ago)
• Language: Rust
• Size: 73.2 KB
• Stars: 42
• Watchers: 3
• Forks: 1
• Open Issues: 2
• Metadata Files:
  • Readme: README.md
  • Contributing: CONTRIBUTING.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# Robust Binary Search

Robust Binary Search provides a binary search implementation which is robust against errors during

the search. In other words, if the comparison function sometimes returns an incorrect result, the

search in this project will still converge on the correct solution.

This is adapted from the multiplicative weights algorithm in ["Noisy binary search and its

applications" by Karp and Kleinberg](https://www.cs.cornell.edu/~rdk/papers/karpr2.pdf), with

adjustments to make it deterministic and then extended to support directed acyclic graphs.

## Usage

To use the git bisect replacement, install with `cargo install robust-git-bisect`, and then

`~/.cargo/bin/robust-git-bisect $start_commit $end_commit $command_to_test_commit`

For library usage, see `AutoSearcher` for binary search over a linear range and

`AutoCompressedDAGSearcher` for binary search over a graph.

## Performance

This code is optimized to minimize the number of tests executed (i.e. number of iterations) and not

necessarily the CPU time of the search algorithm itself, so this will be slower than a plain binary

search if the test is deterministic.

The linear algorithm (`Searcher` and `AutoSearcher`) takes approximately `O(log N)` time per

iteration. The graph algorithm (`CompressedDAGSearcher` and `AutoCompressedDAGSearcher`) takes

approximately `O(segments)` time per iteration.

robust-git-bisect shows improved performance compared with git bisect (higher accuracy with fewer

iterations):

Method                             | Iterations | Accuracy

---------------------------------- | ---------- | --------

robust-git-bisect with 0.99 target | 29.6558    | 99.5392%

robust-git-bisect with 0.9 target  | 26.1828    | 98.8950%

git bisect                         | 16.1907    | 31.7972%

git bisect with tests repeated     | 35.0465    | 86.6359%

git bisect repeated                | 72.3674    | 86.1751%

This test is run over the `git` git repo from e83c516331 to 54e85e7af1, simulating 9c3592cf3c as the

bad commit, with a test that returns an incorrect result 5% of the time. See benchmark.rs for

details.