lightgbm的sklearn接口和原生接口参数详细说明及调参指点

class lightgbm.LGBMClassifier(boosting_type='gbdt', num_leaves=31, max_depth=-1, learning_rate=0.1, n_estimators=10, max_bin=255, subsample_for_bin=200000, objective=None, min_split_gain=0.0, min_child_weight=0.001, min_child_samples=20, subsample=1.0, subsample_freq=1, colsample_bytree=1.0, reg_alpha=0.0, reg_lambda=0.0, random_state=None, n_jobs=-1, silent=True, **kwargs)

boosting_type	default="gbdt"	"gbdt":Gradient Boosting Decision Tree "dart":Dropouts meet Multiple Additive Regression Trees "goss":Gradient-based One-Side Sampling "rf": Random Forest
num_leaves	(int, optional (default=31))	每个基学习器的最大叶子节点	<=2^max_depth
max_depth	(int, optional (default=-1))	每个基学习器的最大深度, -1 means no limit	当模型过拟合，首先降低max_depth
learning_rate	(float, optional (default=0.1))	Boosting learning rate
n_estimators	(int, optional (default=10))	基学习器的数量
max_bin	(int, optional (default=255))	feature将存入的bin的最大数量，应该是直方图的k值
subsample_for_bin	(int, optional (default=50000))	Number of samples for constructing bins
objective	(string, callable or None, optional (default=None))	default: ‘regression’ for LGBMRegressor,  ‘binary’ or ‘multiclass’ for LGBMClassifier,  ‘lambdarank’ for LGBMRanker.
min_split_gain	(float, optional (default=0.))	树的叶子节点上进行进一步划分所需的最小损失减少
min_child_weight	(float, optional (default=1e-3))	Minimum sum of instance weight(hessian) needed in a child(leaf)
min_child_samples	(int, optional (default=20))	叶子节点具有的最小记录数
subsample	(float, optional (default=1.))	训练时采样一定比例的数据
subsample_freq	(int, optional (default=1))	Frequence of subsample, <=0 means no enable
colsample_bytree	(float, optional (default=1.))	Subsample ratio of columns when constructing each tree
reg_alpha	(float, optional (default=0.))	L1 regularization term on weights
reg_lambda	(float, optional (default=0.))	L2 regularization term on weights
random_state	(int or None, optional (default=None))
silent	(bool, optional (default=True))
n_jobs	(int, optional (default=-1))

   

 

 

 

 

 

 

 

 

 ######################################################################################################

下表对应了Faster Spread，better accuracy，over-fitting三种目的时，可以调整的参数:

########################################################################################### 

类的属性：

n_features_	int	特征的数量
classes_	rray of shape = [n_classes]	类标签数组（只针对分类问题）
n_classes_	int	类别数量   （只针对分类问题）
best_score_	dict or None	最佳拟合模型得分
best_iteration_	int or None	如果已经指定了early_stopping_rounds，则拟合模型的最佳迭代次数
objective_	string or callable	拟合模型时的具体目标
booster_	Booster	这个模型的Booster
evals_result_	dict or None	如果已经指定了early_stopping_rounds，则评估结果
feature_importances_	array of shape = [n_features]	特征的重要性

 ###########################################################################################

类的方法： 

fit(X, y, sample_weight=None, init_score=None, eval_set=None, eval_names=None, eval_sample_weight=None, eval_init_score=None, eval_metric='logloss', early_stopping_rounds=None, verbose=True, feature_name='auto', categorical_feature='auto', callbacks=None)

X	array-like or sparse matrix of shape = [n_samples, n_features]	特征矩阵
y	array-like of shape = [n_samples]	The target values (class labels in classification, real numbers in regression)
sample_weight	array-like of shape = [n_samples] or None, optional (default=None))	样本权重,可以采用np.where设置
init_score	array-like of shape = [n_samples] or None, optional (default=None))	Init score of training data
group	array-like of shape = [n_samples] or None, optional (default=None)	Group data of training data.
eval_set	list or None, optional (default=None)	A list of (X, y) tuple pairs to use as a validation sets for early-stopping
eval_names	list of strings or None, optional (default=None)	Names of eval_set
eval_sample_weight	list of arrays or None, optional (default=None)	Weights of eval data
eval_init_score	list of arrays or None, optional (default=None)	Init score of eval data
eval_group	list of arrays or None, optional (default=None)	Group data of eval data
eval_metric	string, list of strings, callable or None, optional (default="logloss")	"mae","mse",...
early_stopping_rounds	int or None, optional (default=None)	一定rounds,即停止迭代
verbose	bool, optional (default=True)
feature_name	list of strings or 'auto', optional (default="auto")	If ‘auto’ and data is pandas DataFrame, data columns names are used
categorical_feature	list of strings or int, or 'auto', optional (default="auto")	If ‘auto’ and data is pandas DataFrame, pandas categorical columns are used
callbacks	list of callback functions or None, optional (default=None)

###############################################################################################

 predict_proba(X, raw_score=False, num_iteration=0)

X	array-like or sparse matrix of shape = [n_samples, n_features]	Input features matrix
raw_score	bool, optional (default=False)	Whether to predict raw scores
num_iteration	int, optional (default=0)	Limit number of iterations in the prediction; defaults to 0 (use all trees).
Returns	predicted_probability	The predicted probability for each class for each sample.
Return type	array-like of shape = [n_samples, n_classes]

 

 

 

不平衡处理的参数：

1.一个简单的方法是设置is_unbalance参数为True或者设置scale_pos_weight,二者只能选一个。 设置is_unbalance参数为True时会把负样本的权重设为：正样本数/负样本数。这个参数只能用于二分类。

2.自定义评价函数：

https://cloud.tencent.com/developer/article/1357671

 

lightGBM的原理总结：

http://www.cnblogs.com/gczr/p/9024730.html

论文翻译：https://blog.csdn.net/u010242233/article/details/79769950，https://zhuanlan.zhihu.com/p/42939089

处理分类变量的原理：https://blog.csdn.net/anshuai_aw1/article/details/83275299

CatBoost、LightGBM、XGBoost的对比

https://blog.csdn.net/LrS62520kV/article/details/79620615