XGBoostClassifier¶

class
getml.predictors.
XGBoostClassifier
(booster='gbtree', colsample_bylevel=1.0, colsample_bytree=1.0, gamma=0.0, learning_rate=0.1, max_delta_step=0.0, max_depth=3, min_child_weights=1.0, n_estimators=100, normalize_type='tree', num_parallel_tree=1, n_jobs=1, objective='binary:logistic', one_drop=False, rate_drop=0.0, reg_alpha=0.0, reg_lambda=1.0, sample_type='uniform', silent=True, skip_drop=0.0, subsample=1.0)¶ Bases:
getml.predictors._Predictor
Gradient boosting classifier based on xgboost.
XGBoost is an implementation of the gradient tree boosting algorithm that is widely recognized for its efficiency and predictive accuracy.
Gradient tree boosting trains an ensemble of decision trees by training each tree to predict the prediction error of all previous trees in the ensemble:
where is the prediction generated by the newest decision tree for sample and is the prediction generated by all previous trees, is the loss function used and is the target we are trying to predict.
XGBoost implements this general approach by adding two specific components:
The loss function is approximated using a Taylor series.
The leaves of the decision tree contain weights that can be regularized.
These weights are calculated as follows:
where and are the first and second order derivative of w.r.t. , denotes the weight on leaf and denotes all samples on that leaf.
is the regularization parameter reg_lambda. This hyperparameter can be set by the users or the hyperparameter optimization algorithm to avoid overfitting.
 Parameters
booster (string, optional) –
Which base classifier to use.
Possible values:
’gbtree’: normal gradient boosted decision trees
’gblinear’: uses a linear model instead of decision trees
’dart’: adds dropout to the standard gradient boosting algorithm. Please also refer to the remarks on rate_drop for further explanation on ‘dart’.
colsample_bylevel (float, optional) –
Subsample ratio for the columns used, for each level inside a tree.
Note that XGBoost grows its trees levelbylevel, not nodebynode. At each level, a subselection of the features will be randomly picked and the best feature for each split will be chosen. This hyperparameter determines the share of features randomly picked at each level. When set to 1, then now such sampling takes place.
Decreasing this hyperparameter reduces the likelihood of overfitting.
Range: (0, 1]
colsample_bytree (float, optional) –
Subsample ratio for the columns used, for each tree. This means that for each tree, a subselection of the features will be randomly chosen. This hyperparameter determines the share of features randomly picked for each tree.
Decreasing this hyperparameter reduces the likelihood of overfitting.
Range: (0, 1]
gamma (float, optional) –
Minimum loss reduction required for any update to the tree. This means that every potential update will first be evaluated for its improvement to the loss function. If the improvement exceeds gamma, the update will be accepted.
Increasing this hyperparameter reduces the likelihood of overfitting.
Range: [0, ]
learning_rate (float, optional) –
Learning rate for the gradient boosting algorithm. When a new tree is trained, it will be added to the existing trees . Before doing so, it will be multiplied by the learning_rate.
Decreasing this hyperparameter reduces the likelihood of overfitting.
Range: [0, 1]
max_delta_step (float, optional) –
The maximum delta step allowed for the weight estimation of each tree.
Decreasing this hyperparameter reduces the likelihood of overfitting.
Range: [0, )
max_depth (int, optional) –
Maximum allowed depth of the trees.
Decreasing this hyperparameter reduces the likelihood of overfitting.
Range: [0, ]
min_child_weights (float, optional) –
Minimum sum of weights needed in each child node for a split. The idea here is that any leaf should have a minimum number of samples in order to avoid overfitting. This very common form of regularizing decision trees is slightly modified to refer to weights instead of number of samples, but the basic idea is the same.
Increasing this hyperparameter reduces the likelihood of overfitting.
Range: [0, ]
n_estimators (int, optional) –
Number of estimators (trees).
Decreasing this hyperparameter reduces the likelihood of overfitting.
Range: [10, ]
normalize_type (string, optional) –
This determines how to normalize trees during ‘dart’.
Possible values:
’tree’: a new tree has the same weight as a single dropped tree.
’forest’: a new tree has the same weight as a the sum of all dropped trees.
Please also refer to the remarks on rate_drop for further explanation.
Will be ignored if booster is not set to ‘dart’.
n_jobs (int, optional) –
Number of parallel threads.
Range: [1, ]
objective (string, optional) –
Specify the learning task and the corresponding learning objective.
Possible values:
’reg:logistic’
’binary:logistic’
’binary:logitraw’
one_drop (bool, optional) –
If set to True, then at least one tree will always be dropped out. Setting this hyperparameter to true reduces the likelihood of overfitting.
Please also refer to the remarks on rate_drop for further explanation.
Will be ignored if booster is not set to ‘dart’.
rate_drop (float, optional) –
Dropout rate for trees  determines the probability that a tree will be dropped out. Dropout is an algorithm that enjoys considerable popularity in the deep learning community. It means that every node can be randomly removed during training.
This approach can also be applied to gradient boosting, where it means that every tree can be randomly removed with a certain probability. Said probability is determined by rate_drop. Dropout for gradient boosting is referred to as the ‘dart’ algorithm.
Increasing this hyperparameter reduces the likelihood of overfitting.
Will be ignored if booster is not set to ‘dart’.
reg_alpha (float, optional) –
L1 regularization on the weights.
Increasing this hyperparameter reduces the likelihood of overfitting.
Range: [0, ]
reg_lambda (float, optional) –
L2 regularization on the weights. Please refer to the introductory remarks to understand how this hyperparameter influences your weights.
Increasing this hyperparameter reduces the likelihood of overfitting.
Range: [0, ]
sample_type (string, optional) –
Possible values:
’uniform’: every tree is equally likely to be dropped out
’weighted’: the dropout probability will be proportional to a tree’s weight
Please also refer to the remarks on rate_drop for further explanation.
Will be ignored if booster is not set to ‘dart’.
silent (bool, optional) – In silent mode, XGBoost will not print out information on the training progress.
skip_drop (float, optional) –
Probability of skipping the dropout during a given iteration. Please also refer to the remarks on rate_drop for further explanation.
Increasing this hyperparameter reduces the likelihood of overfitting.
Will be ignored if booster is not set to ‘dart’.
Range: [0, 1]
subsample (float, optional) –
Subsample ratio from the training set. This means that for every tree a subselection of samples from the training set will be included into training. Please note that this samples without replacement  the common approach for random forests is to sample with replace.
Decreasing this hyperparameter reduces the likelihood of overfitting.
Range: (0, 1]
 Raises
TypeError – If any of the input arguments does not match its expected type.
Methods Summary
validate
([params])Checks both the types and the values of all instance variables and raises an exception if something is off.
Methods Documentation

validate
(params=None)¶ Checks both the types and the values of all instance variables and raises an exception if something is off.
 Parameters
params (dict, optional) – A dictionary containing the parameters to validate. If not is passed, the own parameters will be validated.
Examples
x = getml.predictors.XGBoostClassifier() x.gamma = 200 x.validate()
 Raises
KeyError – If an unsupported instance variable is encountered.
TypeError – If any instance variable is of wrong type.
ValueError – If any instance variable does not match its possible choices (string) or is out of the expected bounds (numerical).
Note
This method is called at end of the __init__ constructor and every time before the predictor  or a class holding it as an instance variable  is send to the getML engine.