Bias in algorithm portfolio performance evaluation
Bibliographic record
Abstract
A Virtual Best Solver (VBS) is a hypothetical algorithm that selects the best solver from a given portfolio of alternatives on a per-instance basis. The VBS idealizes performance when all solvers in a portfolio are run in parallel, and also gives a valuable bound on the performance of portfolio-based algorithm selectors. Typically, VBS performance is measured by running every solver in a portfolio once on a given instance and reporting the best performance over all solvers. Here, we argue that doing so results in a flawed measure that is biased to reporting better performance when a randomized solver is present in an algorithm portfolio. Specifically, this flawed notion of VBS tends to show performance better than that achievable by a perfect selector that for each given instance runs the solver with the best expected running time. We report results from an empirical study using solvers and instances submitted to several SAT competitions, in which we observe significant bias on many random instances and some combinatorial instances. We also show that the bias increases with the number of randomized solvers and decreases as we average solver performance over many independent runs per instance. We propose an alternative VBS performance measure by (1) empirically obtaining the solver with best expected performance for each instance and (2) taking bootstrap samples for this solver on every instance, to obtain a confidence interval on VBS performance. Our findings shed new light on widely studied algorithm selection benchmarks and help explain performance gaps observed between VBS and state-of-the-art algorithm selection approaches.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.001 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.001 |
| Open science | 0.001 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.002 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; both teacher heads agree on what is shown here.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".