partitionedmultifilter.java

来自「Weka」· Java 代码 · 共 697 行 · 第 1/2 页

    }
    else {
      result  = filter.getClass().getName();
      if (filter instanceof OptionHandler)
        result += " " 
          + Utils.joinOptions(((OptionHandler) filter).getOptions());
    }

    return result;
  }

  /**
   * Sets the list of possible Ranges to choose from.
   * Also resets the state of the Range (this reset doesn't affect the 
   * options).
   *
   * @param Ranges	an array of Ranges with all options set.
   * @see #reset()
   */
  public void setRanges(Range[] Ranges) {
    m_Ranges = Ranges;
    reset();
  }

  /**
   * Gets the list of possible Ranges to choose from.
   *
   * @return 		the array of Ranges
   */
  public Range[] getRanges() {
    return m_Ranges;
  }
  
  /**
   * Returns the tip text for this property
   * 
   * @return    	tip text for this property suitable for
   *            	displaying in the explorer/experimenter gui
   */
  public String rangesTipText() {
    return "The attribute ranges to be used.";
  }
  
  /**
   * Gets a single Range from the set of available Ranges.
   *
   * @param index 	the index of the Range wanted
   * @return 		the Range
   */
  public Range getRange(int index) {
    return m_Ranges[index];
  }
  
  /**
   * determines the indices of unused attributes (ones that are not covered
   * by any of the range)
   * 
   * @param data	the data to base the determination on
   * @see 		#m_IndicesUnused
   */
  protected void determineUnusedIndices(Instances data) {
    Vector<Integer>	indices;
    int			i;
    int			n;
    boolean		covered;
    
    // traverse all ranges
    indices = new Vector<Integer>();
    for (i = 0; i < data.numAttributes(); i++) {
      if (i == data.classIndex())
	continue;
      
      covered = false;
      for (n = 0; n < getRanges().length; n++) {
	if (getRanges()[n].isInRange(i)) {
	  covered = true;
	  break;
	}
      }
      
      if (!covered)
	indices.add(new Integer(i));
    }
    
    // create array
    m_IndicesUnused = new int[indices.size()];
    for (i = 0; i < indices.size(); i++)
      m_IndicesUnused[i] = indices.get(i).intValue();
    
    if (getDebug())
      System.out.println(
	  "Unused indices: " + Utils.arrayToString(m_IndicesUnused));
  }
  
  /**
   * generates a subset of the dataset with only the attributes from the range
   * (class is always added if present)
   * 
   * @param data	the data to work on
   * @param range	the range of attribute to use
   * @return		the generated subset
   * @throws Exception	if creation fails
   */
  protected Instances generateSubset(Instances data, Range range) throws Exception {
    Remove	filter;
    String	atts;
    Instances	result;
 
    // determine attributes
    atts = range.getRanges();
    if ((data.classIndex() > -1) && (!range.isInRange(data.classIndex())))
      atts += "," + (data.classIndex() + 1);
    
    // setup filter
    filter = new Remove();
    filter.setAttributeIndices(atts);
    filter.setInvertSelection(true);
    filter.setInputFormat(data);
    
    // generate output
    result = Filter.useFilter(data, filter);
    
    return result;
  }
  
  /**
   * renames all the attributes in the dataset (excluding the class if present)
   * by adding the prefix to the name.
   * 
   * @param data	the data to work on
   * @param prefix	the prefix for the attributes
   * @return		a copy of the data with the attributes renamed
   * @throws Exception	if renaming fails
   */
  protected Instances renameAttributes(Instances data, String prefix) throws Exception {
    Instances	result;
    int		i;
    FastVector	atts;
    
    // rename attributes
    atts = new FastVector();
    for (i = 0; i < data.numAttributes(); i++) {
      if (i == data.classIndex())
	atts.addElement(data.attribute(i).copy());
      else
	atts.addElement(data.attribute(i).copy(prefix + data.attribute(i).name()));
    }
    
    // create new dataset
    result = new Instances(data.relationName(), atts, data.numInstances());
    for (i = 0; i < data.numInstances(); i++) {
      result.add((Instance) data.instance(i).copy());
    }
    
    // set class if present
    if (data.classIndex() > -1)
      result.setClassIndex(data.classIndex());
    
    return result;
  }
  
  /**
   * Determines the output format based only on the full input dataset and 
   * returns this otherwise null is returned. In case the output format cannot 
   * be returned immediately, i.e., immediateOutputFormat() returns false, 
   * then this method will be called from batchFinished().
   *
   * @param inputFormat     the input format to base the output format on
   * @return                the output format
   * @throws Exception      in case the determination goes wrong
   * @see                   #hasImmediateOutputFormat()
   * @see                   #batchFinished()
   */
  protected Instances determineOutputFormat(Instances inputFormat) throws Exception {
    Instances   result;
    Instances	processed;
    int         i;
    int		n;
    FastVector	atts;
    Attribute	att;
    
    if (!isFirstBatchDone()) {
      // we need the full dataset here, see process(Instances)
      if (inputFormat.numInstances() == 0)
	return null;

      checkDimensions();

      // determine unused indices
      determineUnusedIndices(inputFormat);

      atts = new FastVector();
      for (i = 0; i < getFilters().length; i++) {
	if (!isFirstBatchDone()) {
	  // generate subset
	  processed = generateSubset(inputFormat, getRange(i));
	  // set input format
	  if (!getFilter(i).setInputFormat(processed))
	    Filter.useFilter(processed, getFilter(i));
	}

	// get output format
	processed = getFilter(i).getOutputFormat();

	// rename attributes
	processed = renameAttributes(processed, "filtered-" + i + "-");

	// add attributes
	for (n = 0; n < processed.numAttributes(); n++) {
	  if (n == processed.classIndex())
	    continue;
	  atts.addElement(processed.attribute(n).copy());
	}
      }

      // add unused attributes
      if (!getRemoveUnused()) {
	for (i = 0; i < m_IndicesUnused.length; i++) {
	  att = inputFormat.attribute(m_IndicesUnused[i]);
	  atts.addElement(att.copy("unfiltered-" + att.name()));
	}
      }

      // add class if present
      if (inputFormat.classIndex() > -1)
	atts.addElement(inputFormat.classAttribute().copy());

      // generate new dataset
      result = new Instances(inputFormat.relationName(), atts, 0);
      if (inputFormat.classIndex() > -1)
	result.setClassIndex(result.numAttributes() - 1);
    }
    else {
      result = getOutputFormat();
    }
    
    return result;
  }

  /**
   * Processes the given data (may change the provided dataset) and returns
   * the modified version. This method is called in batchFinished().
   *
   * @param instances   the data to process
   * @return            the modified data
   * @throws Exception  in case the processing goes wrong
   * @see               #batchFinished()
   */
  protected Instances process(Instances instances) throws Exception {
    Instances		result;
    int        		i;
    int			n;
    int			m;
    int			index;
    Instances[]		processed;
    Instance		inst;
    Instance		newInst;
    double[]		values;
    Vector		errors;

    if (!isFirstBatchDone()) {
      checkDimensions();

      // set upper limits
      for (i = 0; i < m_Ranges.length; i++)
	m_Ranges[i].setUpper(instances.numAttributes() - 1);

      // determine unused indices
      determineUnusedIndices(instances);
    }

    // pass data through all filters
    processed = new Instances[getFilters().length];
    for (i = 0; i < getFilters().length; i++) {
      processed[i] = generateSubset(instances, getRange(i));
      if (!isFirstBatchDone())
	getFilter(i).setInputFormat(processed[i]);
      processed[i] = Filter.useFilter(processed[i], getFilter(i));
    }

    // set output format (can only be determined with full dataset, hence here)
    if (!isFirstBatchDone()) {
      result = determineOutputFormat(instances);
      setOutputFormat(result);
    }
    else {
      result = getOutputFormat();
    }
    
    // check whether all filters didn't change the number of instances
    errors = new Vector();
    for (i = 0; i < processed.length; i++) {
      if (processed[i].numInstances() != instances.numInstances())
	errors.add(new Integer(i));
    }
    if (errors.size() > 0)
      throw new IllegalStateException(
	  "The following filter(s) changed the number of instances: " + errors);
    
    // assemble data
    for (i = 0; i < instances.numInstances(); i++) {
      inst   = instances.instance(i);
      values = new double[result.numAttributes()];

      // filtered data
      index = 0;
      for (n = 0; n < processed.length; n++) {
	for (m = 0; m < processed[n].numAttributes(); m++) {
	  if (m == processed[n].classIndex())
	    continue;
	  values[index] = processed[n].instance(i).value(m);
	  index++;
	}
      }
      
      // unused attributes
      if (!getRemoveUnused()) {
	for (n = 0; n < m_IndicesUnused.length; n++) {
	  values[index] = inst.value(m_IndicesUnused[n]);
	  index++;
	}
      }
      
      // class
      if (instances.classIndex() > -1)
	values[values.length - 1] = inst.value(instances.classIndex());

      // generate and add instance
      if (inst instanceof SparseInstance)
	newInst = new SparseInstance(instances.instance(i).weight(), values);
      else
	newInst = new Instance(instances.instance(i).weight(), values);
      result.add(newInst);
    }
    
    return result;
  }

  /**
   * Main method for executing this class.
   *
   * @param args should contain arguments for the filter: use -h for help
   */
  public static void main(String[] args) {
    runFilter(new PartitionedMultiFilter(), args);
  }
}