fit - Class: NaiveBayes

Create Naive Bayes classifier object by fitting training data

Syntax

Description

nb = NaiveBayes.fit(training, class) builds a NaiveBayes classifier object nb. training is an N-by-D numeric matrix of training data. Rows of training correspond to observations; columns correspond to features. class is a classing variable for training (see Grouped Data) taking K distinct levels. Each element of class defines which class the corresponding row of training belongs to. training and class must have the same number of rows.

nb = NaiveBayes.fit(..., 'param1',val1, 'param2',val2, ...) specifies one or more of the following name/value pairs:

• 'Distribution' – a string or a 1-by-D cell vector of strings, specifying which distributions fit uses to model the data. If the value is a string, fit models all the features using one type of distribution. fit can also model different features using different types of distributions. If the value is a cell vector, its jth element specifies the distribution fit uses for the jth feature. The available types of distributions are:

  'normal' (default)	Normal (Gaussian) distribution.
  'kernel'	Kernel smoothing density estimate.
  'mvmn'	Multivariate multinomial distribution for discrete data. fit assumes each individual feature follows a multinomial model within a class. The parameters for a feature include the probabilities of all possible values that the corresponding feature can take.
  'mn'	Multinomial distribution for classifying the count-based data such as the bag-of-tokens model. In the bag-of-tokens model, the value of the jth feature is the number of occurrences of the jth token in this observation, so it must be a non-negative integer. When 'mn' is used, fit considers each observation as multiple trials of a multinomial distribution, and considers each occurrence of a token as one trial. The number of categories (bins) in this multinomial model is the number of distinct tokens, i.e., the number of columns of training.

• 'Prior' – The prior probabilities for the classes, specified as one of the following:

  'empirical' (default)	fit estimates the prior probabilities from the relative frequencies of the classes in training.
  'uniform'	The prior probabilities are equal for all classes.
  vector	A numeric vector of length K specifying the prior probabilities in the class order of class.
  structure	A structure S containing class levels and their prior probabilities. S must have two fields: S.prob: A numeric vector of prior probabilities. S.class: A vector of the same type as class, containing unique class levels indicating the class for the corresponding element of prob. S.class must contain all the K levels in class. It can also contain classes that do not appear in class. This can be useful if training is a subset of a larger training set. fit ignores any classes that appear in S.class but not in class.

  If the prior probabilities don't sum to one, fit will normalize them.

• 'KSWidth' – The bandwidth of the kernel smoothing window. The default is to select a default bandwidth automatically for each combination of feature and class, using a value that is optimal for a Gaussian distribution. You can specify the value as one of the following:

  scalar	Width for all features in all classes.
  row vector	1-by-D vector where the jth element is the bandwidth for the jth feature in all classes.
  column vector	K-by-1 vector where the ith element specifies the bandwidth for all features in the ith class. K represents the number of class levels.
  matrix	K-by-D matrix M where M(i,j) specifies the bandwidth for the jth feature in the ith class.
  structure	A structure S containing class levels and their bandwidths. S must have two fields: S.width – A numeric array of bandwidths specified as a row vector, or a matrix with D columns. S.class – A vector of the same type as class, containing unique class levels indicating the class for the corresponding row of width.

• 'KSSupport' – The regions where the density can be applied. It can be a string, a two-element vector as shown below, or a 1-by-D cell array of these values:

  'unbounded' (default)	The density can extend over the whole real line.
  'positive'	The density is restricted to positive values.
  [L,U]	A two-element vector specifying the finite lower bound L and upper bound U for the support of the density.

• 'KSType' – The type of kernel smoother to use. It can be a string or a 1-by-D cell array of strings. Each string can be 'normal' (default), 'box', 'triangle', or 'epanechnikov'.

How To

• Naive Bayes Classification
• Grouped Data

© 1984-2012- The MathWorks, Inc.    -   Site Help   -   Patents   -   Trademarks   -   Privacy Policy   -   Preventing Piracy   -   RSS