Parameters¶

For more detail, see API reference.

Core boosting¶

Parameter	Default	Effect
`n_estimators`	`2000`	Maximum boosting rounds (trees).
`learning_rate`	`None` (auto)	Per-tree shrinkage. `None` resolves to 0.1 with early stopping. Lower trades more trees for slightly better fit.
`depth`	`None`→auto (reg) / `6` (clf)	Tree depth (a depth-d tree is d splits). The regressor's `None` resolves to 6 for `"RMSE"`/`"MAE"` and 4 for `loss="Quantile"` (deep leaves overfit the tail quantile). Conservative by default; raise to 8–10 for large, interaction-heavy regression.
`l2_leaf_reg`	`1.0`	L2 penalty on leaf values. Higher is smoother.
`min_child_weight`	`1.0` (reg) / `None`→auto (clf)	Minimum hessian mass on each side of a split. The classifier's `None` is size-adaptive: the full veto (1.0) below ~500 training rows, fading linearly to 0 above ~2000 — oblivious trees underfit large data under a fixed veto, while small data still needs one.
`leaf_estimation_iterations`	`1` (reg) / `None`→`3` (clf)	Extra Newton refinement steps per leaf. The classifier's `None` is an auto default resolving to 3 (it helps small/categorical-heavy binary fits). Applies to the plain constant-leaf path only: inert while linear leaves are active (default-on for binary ≥ ~1000 rows — the per-leaf ridge already fits the second-order-optimal leaf), and not implemented for multiclass or `loss="MAE"`/`"Quantile"`. An explicitly-set value that will be ignored on the path warns.

Binning¶

Parameter	Default	Effect
`max_bins`	`128`	Histogram bins per numeric feature. Raising it can improve fit in some scenarios.
`quantize_gradients`	`True`	Split search on ~15-bit quantized grad/hess packed into integer histograms: ~20-25% faster fits, benchmark-flat accuracy. Leaf values always use exact float gradients. Deterministic per `random_state`. `False` = exact float64 histograms.

Row and column sampling¶

Parameter	Default	Effect
`subsample`	`1.0`	Row fraction per tree. Below 1.0, uses Minimum Variance Sampling (gradient-weighted, unbiased).
`colsample`	`None`	Feature fraction eligible per tree. `None` = 1.0 for a single model, 0.85 for members inside `n_ensembles > 1` (see recipes: bagging).

Categorical features¶

Parameter	Default	Effect
`cat_smoothing`	`1.0`	Prior strength for ordered target statistics; higher shrinks rare categories toward the global mean. Must be `> 0`.
`cat_n_permutations`	`4`	Random orderings averaged by the ordered target encoder.
`cat_combinations`	`None`→auto	Add all pairwise category-by-category features. `None` turns them on automatically only when the data is entirely categorical (where they help without crowding out numeric splits); set `True`/`False` to force it. Auto is skipped for very wide all-categorical data (a resource guard against the `C(n_cat, 2)` blow-up) — pass `True` there if you want them anyway.

Which columns are categorical can be passed either to fit(..., cat_features=[...]) or as the cat_features to your ChimeraBoostRegressor/ChimeraBoostClassifier arguments depending on your use case. Columns may be named by integer position or by column name (resolved against the DataFrame), or a mix — e.g. cat_features=["city", "brand"] or cat_features=[0, 3].

Loss (regressor only)¶

Parameter	Default	Effect
`loss`	`"RMSE"`	`"RMSE"`, `"MAE"` (median), or `"Quantile"`.
`alpha`	`0.5`	Quantile level for `loss="Quantile"`.

The classifier picks its loss automatically: binary logloss for 2 classes, softmax for 3+.

Leaf models¶

Parameter	Default	Effect
`linear_leaves`	`None` → auto	Fit a ridge linear model per leaf over the numeric split features instead of a constant. Binary classification: on by default. Regression: the default fits both variants and keeps the validation winner (~2× fit; `True`/`False` skips the double fit). Falls back to constant below ~1000 rows. Not available with MAE/Quantile or multiclass.
`linear_lambda`	`1.0`	Ridge penalty on per-leaf slopes; larger is closer to a constant.

Cross features¶

Parameter	Default	Effect
`cross_features`	`None` → auto	Refit with cross columns for the base fit's top features and keep whichever model has the lower validation loss (`cross_features_selected_`, `cross_pairs_` record the outcome): difference and product columns for the top-6 numeric pairs, plus group-centered columns `x_i − mean(x_i \\| category)` for the top-4 numerics × top-3 categoricals. Oblivious trees can only staircase "x_i < x_j" or "above this row's category baseline"; a cross column makes each one split. Large wins on interaction-heavy data (coordinates, prices, physical units) and entity-heavy categorical data. Auto applies to RMSE regression and to binary and multiclass classification with ≥ 2000 rows and ≥ 2 numeric features — or ≥ 1 numeric plus categoricals — (multiclass judges on softmax log loss), and skips everything else. `False` turns it off.
`selection_rounds`	`100`	Round budget for the internal selection fits (the linear-leaves double fit and the pre-cross base fit). Candidates are judged on their best validation loss within the budget; only the winner continues to full early stopping (~1.5× faster fits). An audition that early-stops before the budget already is the full fit. `None` runs every variant to full early stopping; the audition can occasionally pick a different variant than full runs would.

Ordered boosting¶

Parameter	Default	Effect
`ordered_boosting`	`False`	Leave-one-out leaf training step. Off by default; mutually exclusive with `leaf_estimation_iterations` in the booster. Ignored while linear leaves are active (the linear-leaf update owns the training step — set `linear_leaves=False` to use it) and with `loss="MAE"`/`"Quantile"`; supported for multiclass.

Early stopping¶

Parameter	Default	Effect
`early_stopping`	`True`	Hold out a validation split and stop on a plateau. Set `False` to build a fixed `n_estimators` trees.
`early_stopping_rounds`	`None`→`50`	Patience when early stopping is active. `50` is the sweet spot across the Grinsztajn suite; raising it to `100`–`300` helps only large, high-signal datasets (e.g. covertype, electricity, pol) and costs ~25–35% more trees, so it is not worth it as a default — bump it yourself for that kind of data.
`validation_fraction`	`0.2`	Held-out fraction (stratified for classifiers). Ignored when `eval_set` is passed to `fit`.

See Recipes → early stopping for eval_set and groups.

Bagging¶

Parameter	Default	Effect
`n_ensembles`	`None`	`None`/`1` is a single model; `≥2` averages members fit on random row subsamples (see `max_samples`). Reduces variance.
`ensemble_n_jobs`	`-1`	Worker processes fitting members concurrently, each on an equal share of the thread budget (same total cores, identical models, 1.2–2x faster wall-clock). `1` fits members sequentially.
`max_samples`	`0.8`	Fraction of rows each member trains on, drawn without replacement. Beats the classic bootstrap on accuracy and fit time; `1.0` restores the full-size with-replacement bootstrap.

System¶

Parameter	Default	Effect
`thread_count`	`None`	numba threads. `None`/`-1` uses all cores. Affects determinism of floating-point reductions.
`random_state`	`None`	Seed (deterministic for a fixed `thread_count`).
`verbose`	`False`	Print per-round metrics.

`fit()` arguments¶

Argument	Effect
`cat_features`	Columns to treat as categorical, by integer position and/or column name.
`eval_set`	`(X_val, y_val)` validation set; overrides the internal split.
`groups`	Group labels; keeps each group entirely in train or validation when auto-splitting.
`sample_weight`	Per-sample training weights (normalized to mean 1).
`callbacks`	A callable or list of callables `cb(iteration, train_loss, val_loss, model)` invoked each boosting round; returning `True` requests an early stop.

Fitted attributes¶

Attribute	Meaning
`feature_importances_`	Split-gain importance per input feature, summing to 1.
`best_iteration_`	Trees kept after early stopping.
`classes_` (classifier)	Label values, in `predict_proba` column order.
`temperature_` (classifier)	Calibration temperature; > 1 means scores were over-confident.
`quantile_offset_` (regressor)	Split-conformal correction added to `loss="Quantile"` predictions, fitted on the validation split; 0.0 for other losses or without one.
`expected_value_`	SHAP baseline; set after `shap_values` (see SHAP).
`estimators_`	Fitted members when `n_ensembles > 1`, else `None`.
`validation_history_`	Per-round validation loss recorded during fit; empty without a validation split, a list of member histories when bagged.
`linear_leaves_selected_` (regressor)	Leaf variant the auto-audition kept (`True` = linear leaves won on validation).
`cross_features_selected_` / `cross_pairs_`	Whether the cross-feature refit won, and the feature pairs it used.
`member_params_`	Member hyperparameters applied when `n_ensembles > 1` auto-resolves them (see recipes: bagging).