Jump to content

Gradient boosting

From Wikipedia, the free encyclopedia

Gradient boosting is a machine learning technique based on boosting in a functional space, where the target is pseudo-residuals rather than the typical residuals used in traditional boosting. It gives a prediction model in the form of an ensemble of weak prediction models, i.e., models that make very few assumptions about the data, which are typically simple decision trees.[1][2] When a decision tree is the weak learner, the resulting algorithm is called gradient-boosted trees; it usually outperforms random forest.[1] A gradient-boosted trees model is built in a stage-wise fashion as in other boosting methods, but it generalizes the other methods by allowing optimization of an arbitrary differentiable loss function.

History

[edit]

The idea of gradient boosting originated in the observation by Leo Breiman that boosting can be interpreted as an optimization algorithm on a suitable cost function.[3] Explicit regression gradient boosting algorithms were subsequently developed, by Jerome H. Friedman,[4][2] (in 1999 and later in 2001) simultaneously with the more general functional gradient boosting perspective of Llew Mason, Jonathan Baxter, Peter Bartlett and Marcus Frean.[5][6] The latter two papers introduced the view of boosting algorithms as iterative functional gradient descent algorithms. That is, algorithms that optimize a cost function over function space by iteratively choosing a function (weak hypothesis) that points in the negative gradient direction. This functional gradient view of boosting has led to the development of boosting algorithms in many areas of machine learning and statistics beyond regression and classification.

Informal introduction

[edit]

(This section follows the exposition of gradient boosting by Cheng Li.[7])

Like other boosting methods, gradient boosting combines weak "learners" into a single strong learner in an iterative fashion. It is easiest to explain in the least-squares regression setting, where the goal is to "teach" a model to predict values of the form by minimizing the mean squared error , where indexes over some training set of size of actual values of the output variable :

  • the predicted value
  • the observed value
  • the number of samples in

Now, let us consider a gradient boosting algorithm with stages. At each stage () of gradient boosting, suppose some imperfect model (for low , this model may simply return , where the RHS is the mean of ). In order to improve , our algorithm should add some new estimator, . Thus,

or, equivalently,

.

Therefore, gradient boosting will fit to the residual . As in other boosting variants, each attempts to correct the errors of its predecessor . A generalization of this idea to loss functions other than squared error, and to classification and ranking problems, follows from the observation that residuals for a given model are proportional to the negative gradients of the mean squared error (MSE) loss function (with respect to ):

.

So, gradient boosting could be specialized to a gradient descent algorithm, and generalizing it entails "plugging in" a different loss and its gradient.

Algorithm

[edit]

In many supervised learning problems there is an output variable y and a vector of input variables x, related to each other with some probabilistic distribution. The goal is to find some function that best approximates the output variable from the values of input variables. This is formalized by introducing some loss function and minimizing it in expectation:

.

The gradient boosting method assumes a real-valued y. It seeks an approximation in the form of a weighted sum of M functions from some class , called base (or weak) learners:

.

We are usually given a training set of known sample values of x and corresponding values of y. In accordance with the empirical risk minimization principle, the method tries to find an approximation that minimizes the average value of the loss function on the training set, i.e., minimizes the empirical risk. It does so by starting with a model, consisting of a constant function , and incrementally expands it in a greedy fashion:

,
,

for , where is a base learner function.

Unfortunately, choosing the best function at each step for an arbitrary loss function L is a computationally infeasible optimization problem in general. Therefore, we restrict our approach to a simplified version of the problem.

The idea is to apply a steepest descent step to this minimization problem (functional gradient descent).

The basic idea behind the steepest descent is to find a local minimum of the loss function by iterating on . In fact, the local maximum-descent direction of the loss function is the negative gradient.[8]

Hence, moving a small amount such that the linear approximation remains valid:

where . For small , this implies that .

Proof of functional form of Derivative
To prove the following, consider the objective

Doing a Taylor expansion around the fixed point up to first order

Now differentiating w.r.t to , only the derivative of the second term remains . This is the direction of steepest ascent and hence we must move in the opposite (i.e., negative) direction in order to move in the direction of steepest descent.

Furthermore, we can optimize by finding the value for which the loss function has a minimum:

If we considered the continuous case, i.e., where is the set of arbitrary differentiable functions on , we would update the model in accordance with the following equations

where is the step length, defined as In the discrete case however, i.e. when the set is finite[clarification needed], we choose the candidate function h closest to the gradient of L for which the coefficient γ may then be calculated with the aid of line search on the above equations. Note that this approach is a heuristic and therefore doesn't yield an exact solution to the given problem, but rather an approximation. In pseudocode, the generic gradient boosting method is:[4][1]

Input: training set a differentiable loss function number of iterations M.

Algorithm:

  1. Initialize model with a constant value:
  2. For m = 1 to M:
    1. Compute so-called pseudo-residuals:
    2. Fit a base learner (or weak learner, e.g. tree) closed under scaling to pseudo-residuals, i.e. train it using the training set .
    3. Compute multiplier by solving the following one-dimensional optimization problem:
    4. Update the model:
  3. Output

Gradient tree boosting

[edit]

Gradient boosting is typically used with decision trees (especially CARTs) of a fixed size as base learners. For this special case, Friedman proposes a modification to gradient boosting method which improves the quality of fit of each base learner.

Generic gradient boosting at the m-th step would fit a decision tree to pseudo-residuals. Let be the number of its leaves. The tree partitions the input space into disjoint regions and predicts a constant value in each region. Using the indicator notation, the output of for input x can be written as the sum:

where is the value predicted in the region .[9]

Then the coefficients are multiplied by some value , chosen using line search so as to minimize the loss function, and the model is updated as follows:

Friedman proposes to modify this algorithm so that it chooses a separate optimal value for each of the tree's regions, instead of a single for the whole tree. He calls the modified algorithm "TreeBoost". The coefficients from the tree-fitting procedure can be then simply discarded and the model update rule becomes:

Size of trees

[edit]

, the number of terminal nodes in trees, is the method's parameter which can be adjusted for a data set at hand. It controls the maximum allowed level of interaction between variables in the model. With (decision stumps), no interaction between variables is allowed. With the model may include effects of the interaction between up to two variables, and so on.

Hastie et al.[1] comment that typically work well for boosting and results are fairly insensitive to the choice of in this range, is insufficient for many applications, and is unlikely to be required.

Regularization

[edit]

Fitting the training set too closely can lead to degradation of the model's generalization ability. Several so-called regularization techniques reduce this overfitting effect by constraining the fitting procedure.

One natural regularization parameter is the number of gradient boosting iterations M (i.e. the number of trees in the model when the base learner is a decision tree). Increasing M reduces the error on training set, but setting it too high may lead to overfitting. An optimal value of M is often selected by monitoring prediction error on a separate validation data set. Besides controlling M, several other regularization techniques are used.

Another regularization parameter is the depth of the trees. The higher this value the more likely the model will overfit the training data.

Shrinkage

[edit]

An important part of gradient boosting method is regularization by shrinkage which consists in modifying the update rule as follows:

where parameter is called the "learning rate".

Empirically it has been found that using small learning rates (such as ) yields dramatic improvements in models' generalization ability over gradient boosting without shrinking ().[1] However, it comes at the price of increasing computational time both during training and querying: lower learning rate requires more iterations.

Stochastic gradient boosting

[edit]

Soon after the introduction of gradient boosting, Friedman proposed a minor modification to the algorithm, motivated by Breiman's bootstrap aggregation ("bagging") method.[2] Specifically, he proposed that at each iteration of the algorithm, a base learner should be fit on a subsample of the training set drawn at random without replacement.[10] Friedman observed a substantial improvement in gradient boosting's accuracy with this modification.

Subsample size is some constant fraction of the size of the training set. When , the algorithm is deterministic and identical to the one described above. Smaller values of introduce randomness into the algorithm and help prevent overfitting, acting as a kind of regularization. The algorithm also becomes faster, because regression trees have to be fit to smaller datasets at each iteration. Friedman[2] obtained that leads to good results for small and moderate sized training sets. Therefore, is typically set to 0.5, meaning that one half of the training set is used to build each base learner.

Also, like in bagging, subsampling allows one to define an out-of-bag error of the prediction performance improvement by evaluating predictions on those observations which were not used in the building of the next base learner. Out-of-bag estimates help avoid the need for an independent validation dataset, but often underestimate actual performance improvement and the optimal number of iterations.[11][12]

Number of observations in leaves

[edit]

Gradient tree boosting implementations often also use regularization by limiting the minimum number of observations in trees' terminal nodes. It is used in the tree building process by ignoring any splits that lead to nodes containing fewer than this number of training set instances.

Imposing this limit helps to reduce variance in predictions at leaves.

Penalize complexity of tree

[edit]

Another useful regularization techniques for gradient boosted trees is to penalize model complexity of the learned model.[13] The model complexity can be defined as the proportional number of leaves in the learned trees. The joint optimization of loss and model complexity corresponds to a post-pruning algorithm to remove branches that fail to reduce the loss by a threshold. Other kinds of regularization such as an penalty on the leaf values can also be added to avoid overfitting.

Usage

[edit]

Gradient boosting can be used in the field of learning to rank. The commercial web search engines Yahoo[14] and Yandex[15] use variants of gradient boosting in their machine-learned ranking engines. Gradient boosting is also utilized in High Energy Physics in data analysis. At the Large Hadron Collider (LHC), variants of gradient boosting Deep Neural Networks (DNN) were successful in reproducing the results of non-machine learning methods of analysis on datasets used to discover the Higgs boson.[16] Gradient boosting decision tree was also applied in earth and geological studies – for example quality evaluation of sandstone reservoir.[17]

Names

[edit]

The method goes by a variety of names. Friedman introduced his regression technique as a "Gradient Boosting Machine" (GBM).[4] Mason, Baxter et al. described the generalized abstract class of algorithms as "functional gradient boosting".[5][6] Friedman et al. describe an advancement of gradient boosted models as Multiple Additive Regression Trees (MART);[18] Elith et al. describe that approach as "Boosted Regression Trees" (BRT).[19]

A popular open-source implementation for R calls it a "Generalized Boosting Model",[11] however packages expanding this work use BRT.[20] Yet another name is TreeNet, after an early commercial implementation from Salford System's Dan Steinberg, one of researchers who pioneered the use of tree-based methods.[21]

Feature importance ranking

[edit]

Gradient boosting can be used for feature importance ranking, which is usually based on aggregating importance function of the base learners.[22] For example, if a gradient boosted trees algorithm is developed using entropy-based decision trees, the ensemble algorithm ranks the importance of features based on entropy as well with the caveat that it is averaged out over all base learners.[22][1]

Disadvantages

[edit]

While boosting can increase the accuracy of a base learner, such as a decision tree or linear regression, it sacrifices intelligibility and interpretability.[22][23] For example, following the path that a decision tree takes to make its decision is trivial and self-explained, but following the paths of hundreds or thousands of trees is much harder. To achieve both performance and interpretability, some model compression techniques allow transforming an XGBoost into a single "born-again" decision tree that approximates the same decision function.[24] Furthermore, its implementation may be more difficult due to the higher computational demand.

See also

[edit]

References

[edit]
  1. ^ a b c d e f Hastie, T.; Tibshirani, R.; Friedman, J. H. (2009). "10. Boosting and Additive Trees". The Elements of Statistical Learning (2nd ed.). New York: Springer. pp. 337–384. ISBN 978-0-387-84857-0. Archived from the original on 2009-11-10.
  2. ^ a b c d Friedman, J. H. (March 1999). "Stochastic Gradient Boosting" (PDF). Archived from the original (PDF) on 2014-08-01. Retrieved 2013-11-13.
  3. ^ Breiman, L. (June 1997). "Arcing The Edge" (PDF). Technical Report 486. Statistics Department, University of California, Berkeley.
  4. ^ a b c Friedman, J. H. (February 1999). "Greedy Function Approximation: A Gradient Boosting Machine" (PDF). Archived from the original (PDF) on 2019-11-01. Retrieved 2018-08-27.
  5. ^ a b Mason, L.; Baxter, J.; Bartlett, P. L.; Frean, Marcus (1999). "Boosting Algorithms as Gradient Descent" (PDF). In S.A. Solla and T.K. Leen and K. Müller (ed.). Advances in Neural Information Processing Systems 12. MIT Press. pp. 512–518.
  6. ^ a b Mason, L.; Baxter, J.; Bartlett, P. L.; Frean, Marcus (May 1999). "Boosting Algorithms as Gradient Descent in Function Space" (PDF). Archived from the original (PDF) on 2018-12-22.
  7. ^ Cheng Li. "A Gentle Introduction to Gradient Boosting" (PDF).
  8. ^ Lambers, Jim (2011–2012). "The Method of Steepest Descent" (PDF).
  9. ^ Note: in case of usual CART trees, the trees are fitted using least-squares loss, and so the coefficient for the region is equal to just the value of output variable, averaged over all training instances in .
  10. ^ Note that this is different from bagging, which samples with replacement because it uses samples of the same size as the training set.
  11. ^ a b Ridgeway, Greg (2007). Generalized Boosted Models: A guide to the gbm package.
  12. ^ Learn Gradient Boosting Algorithm for better predictions (with codes in R)
  13. ^ Tianqi Chen. Introduction to Boosted Trees
  14. ^ Cossock, David and Zhang, Tong (2008). Statistical Analysis of Bayes Optimal Subset Ranking Archived 2010-08-07 at the Wayback Machine, page 14.
  15. ^ Yandex corporate blog entry about new ranking model "Snezhinsk" Archived 2012-03-01 at the Wayback Machine (in Russian)
  16. ^ Lalchand, Vidhi (2020). "Extracting more from boosted decision trees: A high energy physics case study". arXiv:2001.06033 [stat.ML].
  17. ^ Ma, Longfei; Xiao, Hanmin; Tao, Jingwei; Zheng, Taiyi; Zhang, Haiqin (1 January 2022). "An intelligent approach for reservoir quality evaluation in tight sandstone reservoir using gradient boosting decision tree algorithm". Open Geosciences. 14 (1): 629–645. Bibcode:2022OGeo...14..354M. doi:10.1515/geo-2022-0354. ISSN 2391-5447.
  18. ^ Friedman, Jerome (2003). "Multiple Additive Regression Trees with Application in Epidemiology". Statistics in Medicine. 22 (9): 1365–1381. doi:10.1002/sim.1501. PMID 12704603. S2CID 41965832.
  19. ^ Elith, Jane (2008). "A working guide to boosted regression trees". Journal of Animal Ecology. 77 (4): 802–813. Bibcode:2008JAnEc..77..802E. doi:10.1111/j.1365-2656.2008.01390.x. PMID 18397250.
  20. ^ Elith, Jane. "Boosted Regression Trees for ecological modeling" (PDF). CRAN. Archived from the original (PDF) on 25 July 2020. Retrieved 31 August 2018.
  21. ^ "Exclusive: Interview with Dan Steinberg, President of Salford Systems, Data Mining Pioneer".
  22. ^ a b c Piryonesi, S. Madeh; El-Diraby, Tamer E. (2020-03-01). "Data Analytics in Asset Management: Cost-Effective Prediction of the Pavement Condition Index". Journal of Infrastructure Systems. 26 (1): 04019036. doi:10.1061/(ASCE)IS.1943-555X.0000512. ISSN 1943-555X. S2CID 213782055.
  23. ^ Wu, Xindong; Kumar, Vipin; Ross Quinlan, J.; Ghosh, Joydeep; Yang, Qiang; Motoda, Hiroshi; McLachlan, Geoffrey J.; Ng, Angus; Liu, Bing; Yu, Philip S.; Zhou, Zhi-Hua (2008-01-01). "Top 10 algorithms in data mining". Knowledge and Information Systems. 14 (1): 1–37. doi:10.1007/s10115-007-0114-2. hdl:10983/15329. ISSN 0219-3116. S2CID 2367747.
  24. ^ Sagi, Omer; Rokach, Lior (2021). "Approximating XGBoost with an interpretable decision tree". Information Sciences. 572 (2021): 522–542. doi:10.1016/j.ins.2021.05.055.

Further reading

[edit]
  • Boehmke, Bradley; Greenwell, Brandon (2019). "Gradient Boosting". Hands-On Machine Learning with R. Chapman & Hall. pp. 221–245. ISBN 978-1-138-49568-5.
[edit]