Performing a 10-fold cross validation: Difference between revisions

From Jstacs
Jump to navigationJump to search
No edit summary
No edit summary
Line 3: Line 3:
AlphabetContainer container = new AlphabetContainer( new DNAAlphabet() );
AlphabetContainer container = new AlphabetContainer( new DNAAlphabet() );
//the length of our input sequences
//the length of our input sequences
int length = 16;
int length = 7;


//create a Sample for each class from the input data, using the alphabet from above
//create a Sample for each class from the input data, using the alphabet from above
Sample[] data = new Sample[]{
Sample[] data = new Sample[]{ new Sample( container, new StringExtractor( new File(args[0]), 100 ) ),
                            new Sample( container, new StringExtractor( new File(args[0]), 100 ) ),
new Sample( container, new StringExtractor( new File(args[1]), 100 ), length ) };
                            new Sample( container, new StringExtractor( new File(args[1]), 100 ), length )
};


//create a new PWM
//create a new PWM
BayesianNetworkModel pwm = new BayesianNetworkModel( new BayesianNetworkModelParameterSet(
BayesianNetworkModel pwm = new BayesianNetworkModel( new BayesianNetworkModelParameterSet(
//the alphabet:
//the alphabet and the length of the model:
container,
container, length,  
//the length of the model
length,  
//the equivalent sample size to compute hyper-parameters
//the equivalent sample size to compute hyper-parameters
4,  
4,  
Line 36: Line 32:
//the equivalent sample sizes
//the equivalent sample sizes
new double[]{pwm.getESS(),pwm.getESS()},
new double[]{pwm.getESS(),pwm.getESS()},
//the hyper-parameters to draw the initial component weights (hidden variables)
//the hyper-parameters to draw the initial sequence-specific component weights (hidden variables)
1,
1,
//stopping criterion
//stopping criterion
1E-6,
1E-6,
//parameterization of the model, THETA complies with the
//parameterization of the model, LAMBDA complies with the
//parameterization by probabilities
//parameterization by log-probabilities
Parameterization.THETA);
Parameterization.LAMBDA);


//create a new inhomogeneous Markov model of order 3
//create a new inhomogeneous Markov model of order 3
BayesianNetworkModel mm = new BayesianNetworkModel( new BayesianNetworkModelParameterSet( container, length, 256, "my PWM", ModelType.IMM, (byte) 3, LearningType.ML_OR_MAP ) );
BayesianNetworkModel mm = new BayesianNetworkModel(  
new BayesianNetworkModelParameterSet( container, length, 256, "my PWM", ModelType.IMM, (byte) 3, LearningType.ML_OR_MAP ) );


//create a new PWM scoring function
//create a new PWM scoring function
BayesianNetworkScoringFunction dPwm = new BayesianNetworkScoringFunction(
BayesianNetworkScoringFunction dPwm = new BayesianNetworkScoringFunction(
//the alphabet
//the alphabet and the length of the scoring function
container,
container, length,  
//the length of the scoring function
length,  
//the equivalent sample size for the plug-in parameters
//the equivalent sample size for the plug-in parameters
4,  
4,  
Line 74: Line 69:
//create the classifiers
//create the classifiers
AbstractScoreBasedClassifier[] classifiers = new AbstractScoreBasedClassifier[]{
AbstractScoreBasedClassifier[] classifiers = new AbstractScoreBasedClassifier[]{
                              //model based with mixture model and Markov model
  //model based with mixture model and Markov model
                              new ModelBasedClassifier( mixPwms, mm ),
  new ModelBasedClassifier( mixPwms, mm ),
                              //conditional likelihood based classifier
  //conditional likelihood based classifier
                              new CLLClassifier( new CLLClassifierParameterSet(container, length,  
  new CLLClassifier( new CLLClassifierParameterSet(container, length,  
                              //method for optimizing the conditional likelihood
  //method for optimizing the conditional likelihood and
                              Optimizer.QUASI_NEWTON_BFGS,
  //other parameters of the numerical optimization
                              //parameters of the numerical optimization
  Optimizer.QUASI_NEWTON_BFGS, 1E-6, 1E-6, 1E-2, true, true, false),
                              1E-6, 1E-6, 1E-2, true, true, false),
  //mixture scoring function and Markov model scoring function
                              //mixture scoring function and Markov model scoring function
  dMixPwms,dMm )
                              dMixPwms,dMm )
};
};



Revision as of 12:35, 5 September 2008

//create a DNA-alphabet
AlphabetContainer container = new AlphabetContainer( new DNAAlphabet() );
//the length of our input sequences
int length = 7;

//create a Sample for each class from the input data, using the alphabet from above
Sample[] data = new Sample[]{	new Sample( container, new StringExtractor( new File(args[0]), 100 ) ),
								new Sample( container, new StringExtractor( new File(args[1]), 100 ), length ) };

//create a new PWM
BayesianNetworkModel pwm = new BayesianNetworkModel( new BayesianNetworkModelParameterSet(
		//the alphabet and the length of the model:
		container, length, 
		//the equivalent sample size to compute hyper-parameters
		4, 
		//some identifier for the model
		"my PWM", 
		//we want a PWM, which is an inhomogeneous Markov model (IMM) of order 0
		ModelType.IMM, (byte) 0, 
		//we want to estimate the MAP-parameters
		LearningType.ML_OR_MAP ) );

//create a new mixture model using 2 PWMs
MixtureModel mixPwms = new MixtureModel(
		//the length of the mixture model
		length, 
		//the two components, which are PWMs
		new Model[]{pwm,pwm},
		//the number of starts of the EM
		10,
		//the equivalent sample sizes
		new double[]{pwm.getESS(),pwm.getESS()},
		//the hyper-parameters to draw the initial sequence-specific component weights (hidden variables)
		1,
		//stopping criterion
		1E-6,
		//parameterization of the model, LAMBDA complies with the
		//parameterization by log-probabilities
		Parameterization.LAMBDA);

//create a new inhomogeneous Markov model of order 3
BayesianNetworkModel mm = new BayesianNetworkModel( 
		new BayesianNetworkModelParameterSet( container, length, 256, "my PWM", ModelType.IMM, (byte) 3, LearningType.ML_OR_MAP ) );

//create a new PWM scoring function
BayesianNetworkScoringFunction dPwm = new BayesianNetworkScoringFunction(
		//the alphabet and the length of the scoring function
		container, length, 
		//the equivalent sample size for the plug-in parameters
		4, 
		//we use plug-in parameters
		true, 
		//a PWM is an inhomogeneous Markov model of order 0
		new InhomogeneousMarkov(0));

//create a new mixture scoring function
MixtureScoringFunction dMixPwms = new MixtureScoringFunction(
		//the number of starts
		10,
		//we use plug-in parameters
		true,
		//the two components, which are PWMs
		dPwm,dPwm);

//create a new scoring function that is an inhomogeneous Markov model of order 3
BayesianNetworkScoringFunction dMm = new BayesianNetworkScoringFunction(container, length, 4, true, new InhomogeneousMarkov(3));

//create the classifiers
AbstractScoreBasedClassifier[] classifiers = new AbstractScoreBasedClassifier[]{
							   //model based with mixture model and Markov model
							   new ModelBasedClassifier( mixPwms, mm ),
							   //conditional likelihood based classifier
							   new CLLClassifier( new CLLClassifierParameterSet(container, length, 
									   //method for optimizing the conditional likelihood and 
									   //other parameters of the numerical optimization
									   Optimizer.QUASI_NEWTON_BFGS, 1E-6, 1E-6, 1E-2, true, true, false),
									   //mixture scoring function and Markov model scoring function
									   dMixPwms,dMm )
};

//create an new k-fold cross validation using above classifiers
KFoldCrossValidation cv = new KFoldCrossValidation( classifiers );

//we use a specificity of 0.999 to compute the sensitivity and a sensitivity of 0.95 to compute FPR and PPV
MeasureParameters mp = new MeasureParameters(false, 0.999, 0.95, 0.95);
//we do a 10-fold cross validation and partition the data by means of the number of symbols
KFoldCVAssessParameterSet cvpars = new KFoldCVAssessParameterSet(PartitionMethod.PARTITION_BY_NUMBER_OF_SYMBOLS, length, true, 10);

//compute the result of the cross validation and print them to System.out
System.out.println( cv.assess( mp, cvpars, data ) );