/**
* Builds a new LinearRegression without the 'bad' data found by buildWeight
*
* @throws Exception if building fails
*/
private void buildRLSRegression() throws Exception {
buildWeight();
m_RLSData = new Instances(m_Data);
int x = 0;
int y = 0;
int n = m_RLSData.numInstances();
while (y < n) {
if (m_weight[x] == 0) {
m_RLSData.delete(y);
n = m_RLSData.numInstances();
y--;
}
x++;
y++;
}
if (m_RLSData.numInstances() == 0) {
System.err.println("rls regression unbuilt");
m_ls = m_currentRegression;
} else {
m_ls = new LinearRegression();
m_ls.setOptions(new String[] {"-S", "1"});
m_ls.buildClassifier(m_RLSData);
m_currentRegression = m_ls;
}
}
/**
* Signify that this batch of input to the filter is finished. If the filter requires all
* instances prior to filtering, output() may now be called to retrieve the filtered instances.
*
* @return true if there are instances pending output
* @exception Exception if an error occurs
* @exception IllegalStateException if no input structure has been defined
*/
public boolean batchFinished() throws Exception {
if (getInputFormat() == null) {
throw new IllegalStateException("No input instance format defined");
}
if (m_Means == null) {
Instances input = getInputFormat();
m_Means = new double[input.numAttributes()];
m_StdDevs = new double[input.numAttributes()];
for (int i = 0; i < input.numAttributes(); i++) {
if (input.attribute(i).isNumeric() && (input.classIndex() != i)) {
m_Means[i] = input.meanOrMode(i);
m_StdDevs[i] = Math.sqrt(input.variance(i));
}
}
// Convert pending input instances
for (int i = 0; i < input.numInstances(); i++) {
convertInstance(input.instance(i));
}
}
// Free memory
flushInput();
m_NewBatch = true;
return (numPendingOutput() != 0);
}
/**
* Calculates the performance stats for the desired class and return results as a set of
* Instances.
*
* @param predictions the predictions to base the curve on
* @param classIndex index of the class of interest.
* @return datapoints as a set of instances.
*/
public Instances getCurve(FastVector predictions, int classIndex) {
if ((predictions.size() == 0)
|| (((NominalPrediction) predictions.elementAt(0)).distribution().length <= classIndex)) {
return null;
}
ThresholdCurve tc = new ThresholdCurve();
Instances threshInst = tc.getCurve(predictions, classIndex);
Instances insts = makeHeader();
int fpind = threshInst.attribute(ThresholdCurve.FP_RATE_NAME).index();
int tpind = threshInst.attribute(ThresholdCurve.TP_RATE_NAME).index();
int threshind = threshInst.attribute(ThresholdCurve.THRESHOLD_NAME).index();
double[] vals;
double fpval, tpval, thresh;
for (int i = 0; i < threshInst.numInstances(); i++) {
fpval = threshInst.instance(i).value(fpind);
tpval = threshInst.instance(i).value(tpind);
thresh = threshInst.instance(i).value(threshind);
vals = new double[3];
vals[0] = 0;
vals[1] = fpval;
vals[2] = thresh;
insts.add(new Instance(1.0, vals));
vals = new double[3];
vals[0] = 1;
vals[1] = 1.0 - tpval;
vals[2] = thresh;
insts.add(new Instance(1.0, vals));
}
return insts;
}
/**
* Builds a weight function removing instances with an abnormally high scaled residual
*
* @throws Exception if weight building fails
*/
private void buildWeight() throws Exception {
findResiduals();
m_scalefactor =
1.4826
* (1 + 5 / (m_Data.numInstances() - m_Data.numAttributes()))
* Math.sqrt(m_bestMedian);
m_weight = new double[m_Residuals.length];
for (int i = 0; i < m_Residuals.length; i++)
m_weight[i] = ((Math.sqrt(m_Residuals[i]) / m_scalefactor < 2.5) ? 1.0 : 0.0);
}
/**
* Finds residuals (squared) for the current regression.
*
* @throws Exception if an error occurs
*/
private void findResiduals() throws Exception {
m_SSR = 0;
m_Residuals = new double[m_Data.numInstances()];
for (int i = 0; i < m_Data.numInstances(); i++) {
m_Residuals[i] = m_currentRegression.classifyInstance(m_Data.instance(i));
m_Residuals[i] -= m_Data.instance(i).value(m_Data.classAttribute());
m_Residuals[i] *= m_Residuals[i];
m_SSR += m_Residuals[i];
}
}
/**
* Gets the number of samples to use.
*
* @throws Exception if an error occurs
*/
private void getSamples() throws Exception {
int stuf[] = new int[] {500, 50, 22, 17, 15, 14};
if (m_samplesize < 7) {
if (m_Data.numInstances() < stuf[m_samplesize - 1])
m_samples = combinations(m_Data.numInstances(), m_samplesize);
else m_samples = m_samplesize * 500;
} else m_samples = 3000;
if (m_debug) {
System.out.println("m_samplesize: " + m_samplesize);
System.out.println("m_samples: " + m_samples);
System.out.println("m_randomseed: " + m_randomseed);
}
}
/**
* Creates a new <code>TestSetEvent</code>
*
* @param source the source of the event
* @param testSet the test instances
*/
public TestSetEvent(Object source, Instances testSet) {
super(source);
m_testSet = testSet;
if (m_testSet != null && m_testSet.numInstances() == 0) {
m_structureOnly = true;
}
}
/**
* Cleans up data
*
* @param data data to be cleaned up
* @throws Exception if an error occurs
*/
private void cleanUpData(Instances data) throws Exception {
m_Data = data;
m_TransformFilter = new NominalToBinary();
m_TransformFilter.setInputFormat(m_Data);
m_Data = Filter.useFilter(m_Data, m_TransformFilter);
m_MissingFilter = new ReplaceMissingValues();
m_MissingFilter.setInputFormat(m_Data);
m_Data = Filter.useFilter(m_Data, m_MissingFilter);
m_Data.deleteWithMissingClass();
}
/**
* Returns a string suitable for passing to RemoveRange consisting of m_samplesize indices.
*
* @param data dataset from which to take indicese
* @return string of indices suitable for passing to RemoveRange
*/
private String selectIndices(Instances data) {
StringBuffer text = new StringBuffer();
for (int i = 0, x = 0; i < m_samplesize; i++) {
do {
x = (int) (m_random.nextDouble() * data.numInstances());
} while (x == 0);
text.append(Integer.toString(x));
if (i < m_samplesize - 1) text.append(",");
else text.append("\n");
}
return text.toString();
}
/**
* Tests the CostCurve generation from the command line. The classifier is currently hardcoded.
* Pipe in an arff file.
*
* @param args currently ignored
*/
public static void main(String[] args) {
try {
Instances inst = new Instances(new java.io.InputStreamReader(System.in));
inst.setClassIndex(inst.numAttributes() - 1);
CostCurve cc = new CostCurve();
EvaluationUtils eu = new EvaluationUtils();
Classifier classifier = new features.classifiers.functions.Logistic();
FastVector predictions = new FastVector();
for (int i = 0; i < 2; i++) { // Do two runs.
eu.setSeed(i);
predictions.appendElements(eu.getCVPredictions(classifier, inst, 10));
// System.out.println("\n\n\n");
}
Instances result = cc.getCurve(predictions);
System.out.println(result);
} catch (Exception ex) {
ex.printStackTrace();
}
}
/**
* Build lms regression
*
* @param data training data
* @throws Exception if an error occurs
*/
public void buildClassifier(Instances data) throws Exception {
// can classifier handle the data?
getCapabilities().testWithFail(data);
// remove instances with missing class
data = new Instances(data);
data.deleteWithMissingClass();
cleanUpData(data);
getSamples();
findBestRegression();
buildRLSRegression();
} // buildClassifier
/**
* Converts the header info of the given set of instances into a set of item sets (singletons).
* The ordering of values in the header file determines the lexicographic order.
*
* @param instances the set of instances whose header info is to be used
* @return a set of item sets, each containing a single item
* @exception Exception if singletons can't be generated successfully
*/
public static FastVector singletons(Instances instances) throws Exception {
FastVector setOfItemSets = new FastVector();
ItemSet current;
for (int i = 0; i < instances.numAttributes(); i++) {
if (instances.attribute(i).isNumeric())
throw new Exception("Can't handle numeric attributes!");
for (int j = 0; j < instances.attribute(i).numValues(); j++) {
current = new AprioriItemSet(instances.numInstances());
current.m_items = new int[instances.numAttributes()];
for (int k = 0; k < instances.numAttributes(); k++) current.m_items[k] = -1;
current.m_items[i] = j;
setOfItemSets.addElement(current);
}
}
return setOfItemSets;
}
/**
* Accept a test set
*
* @param e a <code>TestSetEvent</code> value
*/
public void acceptTestSet(final TestSetEvent e) {
if (e.isStructureOnly()) notifyTestListeners(e);
if (m_trainingSet != null
&& m_trainingSet.equalHeaders(e.getTestSet())
&& m_filterThread == null) {
try {
if (m_state == IDLE) {
m_state = FILTERING_TEST;
}
m_testingSet = e.getTestSet();
// final String oldText = m_visual.getText();
m_filterThread =
new Thread() {
public void run() {
try {
if (m_testingSet != null) {
m_visual.setAnimated();
// m_visual.setText("Filtering test data...");
if (m_log != null) {
m_log.statusMessage(
statusMessagePrefix()
+ Messages.getInstance()
.getString("Filter_AcceptTestSet_StatusMessage_Text_First")
+ m_testingSet.relationName()
+ Messages.getInstance()
.getString("Filter_AcceptTestSet_StatusMessage_Text_Second"));
}
Instances filteredTest =
features.filters.Filter.useFilter(m_testingSet, m_Filter);
// m_visual.setText(oldText);
m_visual.setStatic();
TestSetEvent ne =
new TestSetEvent(features.gui.beans.Filter.this, filteredTest);
ne.m_setNumber = e.m_setNumber;
ne.m_maxSetNumber = e.m_maxSetNumber;
notifyTestListeners(ne);
}
} catch (Exception ex) {
ex.printStackTrace();
if (m_log != null) {
m_log.logMessage(
Messages.getInstance()
.getString("Filter_AcceptTestSet_LogMessage_Text_First")
+ statusMessagePrefix()
+ ex.getMessage());
m_log.statusMessage(
statusMessagePrefix()
+ Messages.getInstance()
.getString("Filter_AcceptTestSet_StatusMessage_Text_Third"));
}
Filter.this.stop();
} finally {
// m_visual.setText(oldText);
m_visual.setStatic();
m_state = IDLE;
if (isInterrupted()) {
m_trainingSet = null;
if (m_log != null) {
m_log.logMessage(
Messages.getInstance()
.getString("Filter_AcceptTestSet_LogMessage_Text_Second")
+ statusMessagePrefix()
+ Messages.getInstance()
.getString("Filter_AcceptTestSet_LogMessage_Text_Third"));
m_log.statusMessage(
statusMessagePrefix()
+ Messages.getInstance()
.getString("Filter_AcceptTestSet_StatusMessage_Text_Fourth"));
// m_log.statusMessage("OK");
}
} else {
if (m_log != null) {
m_log.statusMessage(
statusMessagePrefix()
+ Messages.getInstance()
.getString("Filter_AcceptTestSet_StatusMessage_Text_Fifth"));
}
}
block(false);
}
}
};
m_filterThread.setPriority(Thread.MIN_PRIORITY);
m_filterThread.start();
block(true);
m_filterThread = null;
m_state = IDLE;
} catch (Exception ex) {
ex.printStackTrace();
}
}
}
/**
* Accept an instance for processing by StreamableFilters only
*
* @param e an <code>InstanceEvent</code> value
*/
public void acceptInstance(InstanceEvent e) {
// to do!
if (m_filterThread != null) {
String messg =
Messages.getInstance().getString("Filter_AcceptInstance_Mess_Text_First")
+ statusMessagePrefix()
+ Messages.getInstance().getString("Filter_AcceptInstance_Mess_Text_Second");
if (m_log != null) {
m_log.logMessage(messg);
m_log.statusMessage(
statusMessagePrefix()
+ Messages.getInstance()
.getString("Filter_AcceptInstance_StatusMessage_Text_First"));
} else {
System.err.println(messg);
}
return;
}
if (!(m_Filter instanceof StreamableFilter)) {
stop(); // stop all processing
if (m_log != null) {
m_log.logMessage(
Messages.getInstance().getString("Filter_AcceptInstance_LogMessage_Text_First")
+ statusMessagePrefix()
+ Messages.getInstance().getString("Filter_AcceptInstance_LogMessage_Text_Second")
+ m_Filter.getClass().getName()
+ Messages.getInstance().getString("Filter_AcceptInstance_LogMessage_Text_Third"));
m_log.statusMessage(
statusMessagePrefix()
+ Messages.getInstance()
.getString("Filter_AcceptInstance_StatusMessage_Text_Second"));
}
return;
}
if (e.getStatus() == InstanceEvent.FORMAT_AVAILABLE) {
try {
m_instanceCount = 0;
// notifyInstanceListeners(e);
// Instances dataset = e.getInstance().dataset();
Instances dataset = e.getStructure();
if (m_Filter instanceof SupervisedFilter) {
// defualt to last column if no class is set
if (dataset.classIndex() < 0) {
dataset.setClassIndex(dataset.numAttributes() - 1);
}
}
// initialize filter
m_Filter.setInputFormat(dataset);
// attempt to determine post-filtering
// structure. If successful this can be passed on to instance
// listeners as a new FORMAT_AVAILABLE event.
m_structurePassedOn = false;
try {
if (m_Filter.isOutputFormatDefined()) {
// System.err.println("Filter - passing on output format...");
// System.err.println(m_Filter.getOutputFormat());
m_ie.setStructure(m_Filter.getOutputFormat());
notifyInstanceListeners(m_ie);
m_structurePassedOn = true;
}
} catch (Exception ex) {
stop(); // stop all processing
if (m_log != null) {
m_log.logMessage(
Messages.getInstance().getString("Filter_AcceptInstance_LogMessage_Text_Fourth")
+ statusMessagePrefix()
+ Messages.getInstance()
.getString("Filter_AcceptInstance_LogMessage_Text_Fifth")
+ ex.getMessage());
m_log.statusMessage(
statusMessagePrefix()
+ Messages.getInstance()
.getString("Filter_AcceptInstance_StatusMessage_Text_Third"));
} else {
System.err.println(
Messages.getInstance().getString("Filter_AcceptInstance_Error_Text_First")
+ statusMessagePrefix()
+ Messages.getInstance().getString("Filter_AcceptInstance_Error_Text_Second"));
}
}
} catch (Exception ex) {
ex.printStackTrace();
}
return;
}
if (e.getStatus() == InstanceEvent.BATCH_FINISHED) {
// get the last instance (if available)
try {
if (m_log != null) {
m_log.statusMessage(
statusMessagePrefix()
+ Messages.getInstance()
.getString("Filter_AcceptInstance_StatusMessage_Text_Fourth"));
}
if (m_Filter.input(e.getInstance())) {
Instance filteredInstance = m_Filter.output();
if (filteredInstance != null) {
if (!m_structurePassedOn) {
// pass on the new structure first
m_ie.setStructure(new Instances(filteredInstance.dataset(), 0));
notifyInstanceListeners(m_ie);
m_structurePassedOn = true;
}
m_ie.setInstance(filteredInstance);
// if there are instances pending for output don't want to send
// a batch finisehd at this point...
// System.err.println("Filter - in batch finisehd...");
if (m_Filter.batchFinished() && m_Filter.numPendingOutput() > 0) {
m_ie.setStatus(InstanceEvent.INSTANCE_AVAILABLE);
} else {
m_ie.setStatus(e.getStatus());
}
notifyInstanceListeners(m_ie);
}
}
if (m_log != null) {
m_log.statusMessage(
statusMessagePrefix()
+ Messages.getInstance()
.getString("Filter_AcceptInstance_StatusMessage_Text_Fourth_Alpha"));
}
} catch (Exception ex) {
stop(); // stop all processing
if (m_log != null) {
m_log.logMessage(
Messages.getInstance().getString("Filter_AcceptInstance_LogMessage_Text_Sixth")
+ statusMessagePrefix()
+ ex.getMessage());
m_log.statusMessage(
statusMessagePrefix()
+ Messages.getInstance()
.getString("Filter_AcceptInstance_StatusMessage_Text_Fifth"));
}
ex.printStackTrace();
}
// check for any pending instances that we might need to pass on
try {
if (m_Filter.batchFinished() && m_Filter.numPendingOutput() > 0) {
if (m_log != null) {
m_log.statusMessage(
statusMessagePrefix()
+ Messages.getInstance()
.getString("Filter_AcceptInstance_StatusMessage_Text_Sixth"));
}
Instance filteredInstance = m_Filter.output();
if (filteredInstance != null) {
if (!m_structurePassedOn) {
// pass on the new structure first
m_ie.setStructure(new Instances(filteredInstance.dataset(), 0));
notifyInstanceListeners(m_ie);
m_structurePassedOn = true;
}
m_ie.setInstance(filteredInstance);
// TODO here is the problem I think
m_ie.setStatus(InstanceEvent.INSTANCE_AVAILABLE);
notifyInstanceListeners(m_ie);
}
while (m_Filter.numPendingOutput() > 0) {
filteredInstance = m_Filter.output();
m_ie.setInstance(filteredInstance);
// System.err.println("Filter - sending pending...");
if (m_Filter.numPendingOutput() == 0) {
m_ie.setStatus(InstanceEvent.BATCH_FINISHED);
} else {
m_ie.setStatus(InstanceEvent.INSTANCE_AVAILABLE);
}
notifyInstanceListeners(m_ie);
}
if (m_log != null) {
m_log.statusMessage(
statusMessagePrefix()
+ Messages.getInstance()
.getString("Filter_AcceptInstance_StatusMessage_Text_Seventh"));
}
}
} catch (Exception ex) {
stop(); // stop all processing
if (m_log != null) {
m_log.logMessage(
Messages.getInstance().getString("Filter_AcceptInstance_LogMessage_Text_Seventh")
+ statusMessagePrefix()
+ ex.toString());
m_log.statusMessage(
statusMessagePrefix()
+ Messages.getInstance()
.getString("Filter_AcceptInstance_StatusMessage_Text_Eighth"));
}
ex.printStackTrace();
}
} else {
// pass instance through the filter
try {
if (!m_Filter.input(e.getInstance())) {
// System.err.println("Filter - inputing instance into filter...");
/* if (m_log != null) {
m_log.logMessage("ERROR : filter not ready to output instance");
} */
// quietly return. Filter might be able to output some instances
// once the batch is finished.
return;
}
// collect output instance.
Instance filteredInstance = m_Filter.output();
if (filteredInstance == null) {
return;
}
m_instanceCount++;
if (!m_structurePassedOn) {
// pass on the new structure first
m_ie.setStructure(new Instances(filteredInstance.dataset(), 0));
notifyInstanceListeners(m_ie);
m_structurePassedOn = true;
}
m_ie.setInstance(filteredInstance);
m_ie.setStatus(e.getStatus());
if (m_log != null && (m_instanceCount % 10000 == 0)) {
m_log.statusMessage(
statusMessagePrefix()
+ Messages.getInstance()
.getString("Filter_AcceptInstance_StatusMessage_Text_Nineth")
+ m_instanceCount
+ Messages.getInstance()
.getString("Filter_AcceptInstance_StatusMessage_Text_Tenth"));
}
notifyInstanceListeners(m_ie);
} catch (Exception ex) {
stop(); // stop all processing
if (m_log != null) {
m_log.logMessage(
Messages.getInstance().getString("Filter_AcceptInstance_LogMessage_Text_Eighth")
+ statusMessagePrefix()
+ ex.toString());
m_log.statusMessage(
statusMessagePrefix()
+ Messages.getInstance()
.getString("Filter_AcceptInstance_StatusMessage_Text_Eleventh"));
}
ex.printStackTrace();
}
}
}
/**
* Returns a string that describes the filter as source. The filter will be contained in a class
* with the given name (there may be auxiliary classes), and will contain two methods with these
* signatures:
*
* <pre><code>
* // converts one row
* public static Object[] filter(Object[] i);
* // converts a full dataset (first dimension is row index)
* public static Object[][] filter(Object[][] i);
* </code></pre>
*
* where the array <code>i</code> contains elements that are either Double, String, with missing
* values represented as null. The generated code is public domain and comes with no warranty.
*
* @param className the name that should be given to the source class.
* @param data the dataset used for initializing the filter
* @return the object source described by a string
* @throws Exception if the source can't be computed
*/
public String toSource(String className, Instances data) throws Exception {
StringBuffer result;
boolean[] process;
int i;
result = new StringBuffer();
// determine what attributes were processed
process = new boolean[data.numAttributes()];
for (i = 0; i < data.numAttributes(); i++) {
process[i] = (data.attribute(i).isNumeric() && (i != data.classIndex()));
}
result.append("class " + className + " {\n");
result.append("\n");
result.append(" /** lists which attributes will be processed */\n");
result.append(
" protected final static boolean[] PROCESS = new boolean[]{"
+ Utils.arrayToString(process)
+ "};\n");
result.append("\n");
result.append(" /** the computed means */\n");
result.append(
" protected final static double[] MEANS = new double[]{"
+ Utils.arrayToString(m_Means)
+ "};\n");
result.append("\n");
result.append(" /** the computed standard deviations */\n");
result.append(
" protected final static double[] STDEVS = new double[]{"
+ Utils.arrayToString(m_StdDevs)
+ "};\n");
result.append("\n");
result.append(" /**\n");
result.append(" * filters a single row\n");
result.append(" * \n");
result.append(" * @param i the row to process\n");
result.append(" * @return the processed row\n");
result.append(" */\n");
result.append(" public static Object[] filter(Object[] i) {\n");
result.append(" Object[] result;\n");
result.append("\n");
result.append(" result = new Object[i.length];\n");
result.append(" for (int n = 0; n < i.length; n++) {\n");
result.append(" if (PROCESS[n] && (i[n] != null)) {\n");
result.append(" if (STDEVS[n] > 0)\n");
result.append(" result[n] = (((Double) i[n]) - MEANS[n]) / STDEVS[n];\n");
result.append(" else\n");
result.append(" result[n] = ((Double) i[n]) - MEANS[n];\n");
result.append(" }\n");
result.append(" else {\n");
result.append(" result[n] = i[n];\n");
result.append(" }\n");
result.append(" }\n");
result.append("\n");
result.append(" return result;\n");
result.append(" }\n");
result.append("\n");
result.append(" /**\n");
result.append(" * filters multiple rows\n");
result.append(" * \n");
result.append(" * @param i the rows to process\n");
result.append(" * @return the processed rows\n");
result.append(" */\n");
result.append(" public static Object[][] filter(Object[][] i) {\n");
result.append(" Object[][] result;\n");
result.append("\n");
result.append(" result = new Object[i.length][];\n");
result.append(" for (int n = 0; n < i.length; n++) {\n");
result.append(" result[n] = filter(i[n]);\n");
result.append(" }\n");
result.append("\n");
result.append(" return result;\n");
result.append(" }\n");
result.append("}\n");
return result.toString();
}