Sequential Model-based Algorithm Configuration
Algorithms AIMech:SequentialModelBasedAlgorithmConfiguration DRAFT
Aliases: SMAC
A sequential algorithm-configuration procedure that uses random-forest performance models to propose parameter settings and compares candidate settings with an incumbent through adaptive evaluations on shared problem instances and random seeds.
Identity Notes & Concept Scope
The primary scope is the original 2011 SMAC procedure, not every sequential model-based optimizer or every later SMAC software release. It configures a target algorithm over a set or distribution of instances and returns one incumbent configuration; it is not per-instance algorithm selection, a prediction forest by itself, or population-based transfer of trained model weights. Its model-guided proposals distinguish it from ROAR, which shares the intensification procedure but samples configurations at random. No normalized identity match was found across all 442 pinned AIO labels/scoped synonyms and 262 preceding native IDs/names/aliases. Full-checkout fixed-string searches included ignored and hidden files in both worktree and main checkout, returning 9 lines in each checkout. These were prior candidate leads, source references and neighboring-method mentions, not existing records for these identities. Ten exact AIO CLASS nodes with synonyms and outgoing edges and eight neighboring native mechanisms were inspected. Absence is bounded to this snapshot and semantic neighborhood.
Claims & Grounded Evidence
1 Claim
Mechanism
#method
Initialize with an evaluation of the default configuration, or a random setting when no default is available. Fit a random-forest surrogate using configuration values and, when available, training-instance features. Aggregate each tree’s predictions over instances according to the chosen cost metric, then derive predictive moments across trees. For runtime optimization the paper uses log-transformed responses while preserving the original aggregation objective through inverse transformation at leaves. Rank proposals by expected improvement, using its log-cost formulation, and search for promising settings with multistart one-coordinate local changes plus randomly sampled acquisition candidates. Interleave guided proposals with uniformly sampled configurations. The intensification routine compares a challenger and incumbent on shared instance/seed evaluations, enlarges the challenger’s evaluated subset in doubling batches, rejects it when its empirical cost is worse, and replaces the incumbent when the challenger has matched the available evaluation set and is at least as good. Repeat model fitting and configuration comparisons within the overall configuration budget.
⚠️ Curated Limitations & Negative Results
- The original method explicitly lacks adaptive capping of individual poor runs. Its adaptive allocation of additional evaluations must not be equated with the censored-runtime handling or other features of later SMAC releases.
- Surrogate uncertainty and early empirical comparisons do not certify an optimal configuration or generalization to unseen instance distributions. The paper’s asymptotic argument assumes a finite configuration space, consistent cost estimates and unbounded evaluation growth; the displayed experimental procedure caps incumbent evaluations with maxR=2000. These conditions are not an unconditional guarantee for the capped implementation or continuous spaces.
- Instance features are optional; the result is a single configuration and does not require test-instance feature computation. The reviewed source is an author-hosted manuscript, not a claim that its 15 PDF pages match the publisher’s page layout. No benchmark result or executable implementation was reproduced.
Artificial Intelligence Ontology (AIO) Grounding
Mapped AIO Classes
| AIO Target Class | Mapping Relation | Mapping Rationale |
|---|---|---|
AIO:RandomForest
Random Forest
|
related_match | The original SMAC algorithm uses a random-forest regression surrogate as a component. The full proposal-and-comparison procedure is neither identical to that ensemble model nor a subclass of it; this is a component-based conceptual relation. |
Evidence Sources & Primary Literature
1 Source| Source ID | Publication / Document | Type / Access | Reference & Checksum |
|---|---|---|---|
method-source
|
Frank Hutter, Holger H. Hoos, Kevin Leyton-Brown
|
primary paper
full_text
|
ID: doi:10.1007/978-3-642-25566-3_40
sha256:aecf80aba313…
|
Curation Provenance
Curator: Codex
Method: ai_assisted
Model: GPT-6
Created: 2026-10-11T11:53:28Z
Updated: 2026-10-11T11:53:28Z