TopolearnThe scikit-learn wrapper gets you a strong tabular model in a few lines and fits into pipelines and searches.
LightGBM is a gradient boosting library: it adds decision trees one at a time, each correcting the errors of the ensemble so far. The lightgbm package provides scikit-learn compatible estimators:
LGBMClassifier(binary and multi-class),LGBMRegressor,LGBMRankerfor learning to rank.
The usual API applies: constructor hyperparameters, fit(X, y), predict, predict_proba, score, and full compatibility with Pipeline, cross_val_score and GridSearchCV. Inputs can be NumPy arrays, pandas DataFrames (column names are kept; category columns are used natively) or SciPy sparse matrices.
Defaults: 100 trees (n_estimators), learning_rate=0.1, num_leaves=31, no depth limit. No feature scaling is needed and missing values are handled natively. Unlike XGBoost, class labels may be strings: the classifier encodes them and classes_ lists them.
LightGBM logs warnings such as "No further splits with positive gain" during training. Pass verbose=-1 to the constructor to silence them once you know what they mean.
Its main selling point is speed: histogram-based, leaf-wise training with careful engineering, often several times faster than other boosting libraries on large data with similar accuracy.
Example
import lightgbm as lgb
from sklearn.datasets import load_breast_cancer
from sklearn.model_selection import train_test_split
X, y = load_breast_cancer(return_X_y=True, as_frame=True)
X_tr, X_te, y_tr, y_te = train_test_split(X, y, stratify=y, random_state=0)
model = lgb.LGBMClassifier(n_estimators=200, learning_rate=0.05, verbose=-1)
model.fit(X_tr, y_tr)
print(round(model.score(X_te, y_te), 3)) # 0.944
print(model.predict_proba(X_te.iloc[:2]).round(3))
# [[0. 1.]
# [1. 0.]]
labels = y_tr.map({0: "malignant", 1: "benign"}) # string labels work too
print(lgb.LGBMClassifier(verbose=-1).fit(X_tr, labels).classes_) # ['benign' 'malignant']Remember
- LGBMClassifier/LGBMRegressor follow the scikit-learn API
- No scaling needed; NaN and pandas categories are handled natively
verbose=-1silences training warnings
Watch out for
- Silencing warnings with verbose=-1 before ever reading them; they often reveal data problems.