| English | 中文 |
Regression relates a numeric outcome to a set of variables. A useful study defines the question, separates model development from evaluation, and explains what its results mean. This guide follows that sequence; the regression skill supplies configuration and code.
Read common design for the question, variable roles, product lineage, sampling, preprocessing, spatial/time validation and result review. Then choose models, predictive contribution, or causal design. Contribution describes fitted predictions; causal effects require a separately justified identification design.
The model directory groups linear regression, tree-based models and time-series models. Decision trees and random forests share the tree family. The training-mean baseline is a common evaluation reference, not a separate research-model category. These method pages explain scientific choices; matching skill pages own configuration, code and actual runtime status.
The cross-task Transformer architecture method explains attention and study design; its data guide distinguishes tables, sequences, grids and mixed sources. This is a method reference, not a newly enabled model. Existing specialist entries keep their current locations.
The planning method, requirements-driven building, and reproduction method remain authoritative for their own decisions. Select the runtime only after the scientific design: the skill routes R and Python implementations. Stage 02 use remains conceptual only.
The automatic parameter-selection method explains development-only search, grouped validation, final refitting and test isolation; its linked skill reference records actual configuration and runtime limits.