Back to Portfolio

EML Framework: Symbolic Regression Representation Study

A controlled empirical study on how constrained operator representations affect tree structure and search efficiency in grammar-guided symbolic regression.

Structural Inflation

14.8 vs 10.6 nodes

4.4 vs 3.2 depth

EML grammars produce systematically larger and deeper trees.

Rejection Dynamics

75.2% depth rejection

85.9% aggregate (EML)

Depth rejection dominates — structural inflation is the primary bottleneck.

Exact Recovery Rate

44.4% EML

33.3% baseline

8/18 vs 6/18 exact recoveries (MSE < 1e-10).

Loading executive summary...

Raw Data

FileDescriptionAction
symbolic_regression_per_target.csvPer-target SR results with rejection metricsView ↗
depth_vs_error.csvTree depth and MSE per target × grammarView ↗
size_vs_error.csvNode count and MSE per target × grammarView ↗
domain_ablation.csvPer-expression transform outcomes (strict vs relaxed)View ↗
success_rates.csvExact recovery counts and success ratesView ↗
summary.mdFull results summary (Markdown)View Raw ↗