Stage 6: Integration tests (new tests/ dir, end-to-end workflows) (tracking #391)

[Mec-iS]

Tracking issue: #391 — Improve test coverage significantly (staged plan).

Scope

Create a top-level tests/ directory for end-to-end integration tests. This is the largest structural coverage gap: the crate currently has zero integration tests (all 311 existing tests are inline unit tests). Cargo.toml already excludes tests/ from the published package, so this is safe.

Tasks

• Create tests/ directory at repo root.
• One integration test file per estimator family, each exercising a full workflow (load dataset → preprocess → train → predict → evaluate):
  • tests/linear_workflow.rs — LinearRegression / LogisticRegression / RidgeRegression end-to-end.
  • tests/tree_workflow.rs — DecisionTreeClassifier / DecisionTreeRegressor.
  • tests/ensemble_workflow.rs — RandomForestClassifier / RandomForestRegressor.
  • tests/svm_workflow.rs — SVC / SVR with at least two kernels.
  • tests/naive_bayes_workflow.rs — all four NB variants.
  • tests/neighbors_workflow.rs — KNNClassifier / KNNRegressor.
  • tests/cluster_workflow.rs — DBSCAN / KMeans.
  • tests/decomposition_workflow.rs — PCA / SVD fit → transform → reconstruct.
  • tests/preprocessing_workflow.rs — OneHotEncoder + StandardScaler in a pipeline.
  • tests/model_selection_workflow.rs — train_test_split → fit → cross_validate → score.
• Each workflow test should assert non-trivial outcomes (e.g., accuracy above a sane threshold, reconstruction error near zero) rather than just "it runs".
• Gate dataset-dependent tests behind #[cfg(feature = "datasets")]; provide tiny inline fixtures for the no-feature path where possible.

Verification

• cargo test --all-features (integration tests run as part of the suite)
• cargo fmt --all -- --check
• cargo clippy --all-features -- -Drust-2018-idioms -Dwarnings

Do you want to work on this issue?

yes — PR will target development.