predict

Predict responses using ensemble of bagged decision trees

Syntax

Yfit = predict(B,X)
Yfit = predict(B,X,Name,Value)
[Yfit,stdevs] = predict(___)
[Yfit,scores] = predict(___)
[Yfit,scores,stdevs] = predict(___)


Description

Yfit = predict(B,X) returns a vector of predicted responses for the predictor data in the table or matrix X, based on the compact ensemble of bagged decision trees B. Yfit is a cell array of character vectors for classification and a numeric array for regression. By default, predict takes a democratic (nonweighted) average vote from all trees in the ensemble.

B is a trained CompactTreeBagger model object, that is, a model returned by compact.

X is a table or matrix of predictor data used to generate responses. Rows represent observations and columns represent variables.

• If X is a numeric matrix:

  • The variables making up the columns of X must have the same order as the predictor variables that trained B.

  • If you trained B using a table (for example, Tbl), then X can be a numeric matrix if Tbl contains all numeric predictor variables. To treat numeric predictors in Tbl as categorical during training, identify categorical predictors using the CategoricalPredictors name-value pair argument of the TreeBagger function. If Tbl contains heterogeneous predictor variables (for example, numeric and categorical data types) and X is a numeric matrix, then predict throws an error.

• If X is a table:

  • predict does not support multicolumn variables or cell arrays other than cell arrays of character vectors.

  • If you trained B using a table (for example, Tbl), then all predictor variables in X must have the same variable names and be of the same data types as those that trained B (stored in B.PredictorNames). However, the column order of X does not need to correspond to the column order of Tbl. Tbl and X can contain additional variables (response variables, observation weights, etc.), but predict ignores them.

  • If you trained B using a numeric matrix, then the predictor names in B.PredictorNames and corresponding predictor variable names in X must be the same. To specify predictor names during training, see the PredictorNames name-value pair argument of TreeBagger. All predictor variables in X must be numeric vectors. X can contain additional variables (response variables, observation weights, etc.), but predict ignores them.

Yfit = predict(B,X,Name,Value) specifies additional options using one or more name-value pair arguments:

• 'Trees' — Array of tree indices to use for computation of responses. The default is 'all'.

• 'TreeWeights' — Array of NTrees weights for weighting votes from the specified trees, where NTrees is the number of trees in the ensemble.

• 'UseInstanceForTree' — Logical matrix of size Nobs-by-NTrees indicating which trees to use to make predictions for each observation, where Nobs is the number of observations. By default all trees are used for all observations.

For regression, [Yfit,stdevs] = predict(___) also returns standard deviations of the computed responses over the ensemble of the grown trees using any of the input argument combinations in previous syntaxes.

For classification, [Yfit,scores] = predict(___) also returns scores for all classes. scores is a matrix with one row per observation and one column per class. For each observation and each class, the score generated by each tree is the probability of the observation originating from the class, computed as the fraction of observations of the class in a tree leaf. predict averages these scores over all trees in the ensemble.

[Yfit,scores,stdevs] = predict(___) also returns standard deviations of the computed scores for classification. stdevs is a matrix with one row per observation and one column per class, with standard deviations taken over the ensemble of the grown trees.

Algorithms

• For regression problems, the predicted response for an observation is the weighted average of the predictions using selected trees only. That is,

  $${\widehat{y}}_{\text{bag}}=\frac{1}{{\displaystyle \sum _{t=1}^T{\alpha }_{t}I(t\in S)}}{\displaystyle \sum _{t=1}^T{\alpha }_t{\widehat{y}}_{t}I(t\in S)}.$$

  • $${\widehat{y}}_t$$ is the prediction from tree t in the ensemble.

  • S is the set of indices of selected trees that comprise the prediction (see 'Trees' and 'UseInstanceForTree'). $$I(t\in S)$$ is 1 if t is in the set S, and 0 otherwise.

  • α_t is the weight of tree t (see 'TreeWeights').

• For classification problems, the predicted class for an observation is the class that yields the largest weighted average of the class posterior probabilities (i.e., classification scores) computed using selected trees only. That is,

  1. For each class c ∊ C and each tree t = 1,...,T, predict computes $${\widehat{P}}_t\left(c|x\right)$$, which is the estimated posterior probability of class c given observation x using tree t. C is the set of all distinct classes in the training data. For more details on classification tree posterior probabilities, see fitctree and predict.

  2. predict computes the weighted average of the class posterior probabilities over the selected trees.

     $${\widehat{P}}_{\text{bag}}\left(c|x\right)=\frac{1}{{\displaystyle \sum _{t=1}^T{\alpha }_{t}I(t\in S)}}{\displaystyle \sum _{t=1}^T{\alpha }_t{\widehat{P}}_t\left(c|x\right)I(t\in S)}.$$

  3. The predicted class is the class that yields the largest weighted average.

  $${\widehat{y}}_{\text{bag}}=\underset{c\in C}{\arg \max }\left\{{\widehat{P}}_{\text{bag}}\left(c|x\right)\right\}.$$